@point3/node-rdkafka 3.6.0-1

package/deps/librdkafka/tests/0105-transactions_mock.c

@@ -0,0 +1,3930 @@
+/*
+ * librdkafka - Apache Kafka C library
+ *
+ * Copyright (c) 2019-2022, Magnus Edenhill
+ *               2025, Confluent Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "test.h"
+
+#include "rdkafka.h"
+
+#include "../src/rdkafka_proto.h"
+#include "../src/rdstring.h"
+#include "../src/rdunittest.h"
+
+#include <stdarg.h>
+
+
+/**
+ * @name Producer transaction tests using the mock cluster
+ *
+ */
+
+
+static int allowed_error;
+static int allowed_error_2;
+
+/**
+ * @brief Decide what error_cb's will cause the test to fail.
+ */
+static int
+error_is_fatal_cb(rd_kafka_t *rk, rd_kafka_resp_err_t err, const char *reason) {
+        if (err == allowed_error || err == allowed_error_2 ||
+            /* If transport errors are allowed then it is likely
+             * that we'll also see ALL_BROKERS_DOWN. */
+            (allowed_error == RD_KAFKA_RESP_ERR__TRANSPORT &&
+             err == RD_KAFKA_RESP_ERR__ALL_BROKERS_DOWN)) {
+                TEST_SAY("Ignoring allowed error: %s: %s\n",
+                         rd_kafka_err2name(err), reason);
+                return 0;
+        }
+        return 1;
+}
+
+
+static rd_kafka_resp_err_t (*on_response_received_cb)(rd_kafka_t *rk,
+                                                      int sockfd,
+                                                      const char *brokername,
+                                                      int32_t brokerid,
+                                                      int16_t ApiKey,
+                                                      int16_t ApiVersion,
+                                                      int32_t CorrId,
+                                                      size_t size,
+                                                      int64_t rtt,
+                                                      rd_kafka_resp_err_t err,
+                                                      void *ic_opaque);
+
+/**
+ * @brief Simple on_response_received interceptor that simply calls the
+ *        sub-test's on_response_received_cb function, if set.
+ */
+static rd_kafka_resp_err_t
+on_response_received_trampoline(rd_kafka_t *rk,
+                                int sockfd,
+                                const char *brokername,
+                                int32_t brokerid,
+                                int16_t ApiKey,
+                                int16_t ApiVersion,
+                                int32_t CorrId,
+                                size_t size,
+                                int64_t rtt,
+                                rd_kafka_resp_err_t err,
+                                void *ic_opaque) {
+        TEST_ASSERT(on_response_received_cb != NULL, "");
+        return on_response_received_cb(rk, sockfd, brokername, brokerid, ApiKey,
+                                       ApiVersion, CorrId, size, rtt, err,
+                                       ic_opaque);
+}
+
+
+/**
+ * @brief on_new interceptor to add an on_response_received interceptor.
+ */
+static rd_kafka_resp_err_t on_new_producer(rd_kafka_t *rk,
+                                           const rd_kafka_conf_t *conf,
+                                           void *ic_opaque,
+                                           char *errstr,
+                                           size_t errstr_size) {
+        rd_kafka_resp_err_t err = RD_KAFKA_RESP_ERR_NO_ERROR;
+
+        if (on_response_received_cb)
+                err = rd_kafka_interceptor_add_on_response_received(
+                    rk, "on_response_received", on_response_received_trampoline,
+                    ic_opaque);
+
+        return err;
+}
+
+
+/**
+ * @brief Create a transactional producer and a mock cluster.
+ *
+ * The var-arg list is a NULL-terminated list of
+ * (const char *key, const char *value) config properties.
+ *
+ * Special keys:
+ *   "on_response_received", "" - enable the on_response_received_cb
+ *                                interceptor,
+ *                                which must be assigned prior to
+ *                                calling create_tnx_producer().
+ */
+static RD_SENTINEL rd_kafka_t *
+create_txn_producer(rd_kafka_mock_cluster_t **mclusterp,
+                    const char *transactional_id,
+                    int broker_cnt,
+                    ...) {
+        rd_kafka_conf_t *conf;
+        rd_kafka_t *rk;
+        char numstr[8];
+        va_list ap;
+        const char *key;
+        rd_bool_t add_interceptors = rd_false;
+
+        rd_snprintf(numstr, sizeof(numstr), "%d", broker_cnt);
+
+        test_conf_init(&conf, NULL, 60);
+
+        test_conf_set(conf, "transactional.id", transactional_id);
+        /* When mock brokers are set to down state they're still binding
+         * the port, just not listening to it, which makes connection attempts
+         * stall until socket.connection.setup.timeout.ms expires.
+         * To speed up detection of brokers being down we reduce this timeout
+         * to just a couple of seconds. */
+        test_conf_set(conf, "socket.connection.setup.timeout.ms", "5000");
+        /* Speed up reconnects */
+        test_conf_set(conf, "reconnect.backoff.max.ms", "2000");
+        test_conf_set(conf, "test.mock.num.brokers", numstr);
+        rd_kafka_conf_set_dr_msg_cb(conf, test_dr_msg_cb);
+
+        test_curr->ignore_dr_err = rd_false;
+
+        va_start(ap, broker_cnt);
+        while ((key = va_arg(ap, const char *))) {
+                if (!strcmp(key, "on_response_received")) {
+                        add_interceptors = rd_true;
+                        (void)va_arg(ap, const char *);
+                } else {
+                        test_conf_set(conf, key, va_arg(ap, const char *));
+                }
+        }
+        va_end(ap);
+
+        /* Add an on_.. interceptors */
+        if (add_interceptors)
+                rd_kafka_conf_interceptor_add_on_new(conf, "on_new_producer",
+                                                     on_new_producer, NULL);
+
+        rk = test_create_handle(RD_KAFKA_PRODUCER, conf);
+
+        if (mclusterp) {
+                *mclusterp = rd_kafka_handle_mock_cluster(rk);
+                TEST_ASSERT(*mclusterp, "failed to create mock cluster");
+
+                /* Create some of the common consumer "input" topics
+                 * that we must be able to commit to with
+                 * send_offsets_to_transaction().
+                 * The number depicts the number of partitions in the topic. */
+                TEST_CALL_ERR__(
+                    rd_kafka_mock_topic_create(*mclusterp, "srctopic4", 4, 1));
+                TEST_CALL_ERR__(rd_kafka_mock_topic_create(
+                    *mclusterp, "srctopic64", 64, 1));
+        }
+
+        return rk;
+}
+
+
+/**
+ * @brief Test recoverable errors using mock broker error injections
+ *        and code coverage checks.
+ */
+static void do_test_txn_recoverable_errors(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+        const char *groupid = "myGroupId";
+        const char *txnid   = "myTxnId";
+
+        SUB_TEST_QUICK();
+
+        rk = create_txn_producer(&mcluster, txnid, 3, "batch.num.messages", "1",
+                                 NULL);
+
+        /* Make sure transaction and group coordinators are different.
+         * This verifies that AddOffsetsToTxnRequest isn't sent to the
+         * transaction coordinator but the group coordinator. */
+        rd_kafka_mock_coordinator_set(mcluster, "group", groupid, 1);
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", txnid, 2);
+
+        /*
+         * Inject som InitProducerId errors that causes retries
+         */
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_InitProducerId, 3,
+            RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE,
+            RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+            RD_KAFKA_RESP_ERR_COORDINATOR_LOAD_IN_PROGRESS);
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        (void)RD_UT_COVERAGE_CHECK(0); /* idemp_request_pid_failed(retry) */
+        (void)RD_UT_COVERAGE_CHECK(1); /* txn_idemp_state_change(READY) */
+
+        /*
+         * Start a transaction
+         */
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+
+        /* Produce a message without error first */
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        rd_kafka_flush(rk, -1);
+
+        /*
+         * Produce a message, let it fail with a non-idempo/non-txn
+         * retryable error
+         */
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_Produce, 1,
+            RD_KAFKA_RESP_ERR_NOT_ENOUGH_REPLICAS);
+
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        /* Make sure messages are produced */
+        rd_kafka_flush(rk, -1);
+
+        /*
+         * Send some arbitrary offsets, first with some failures, then
+         * succeed.
+         */
+        offsets = rd_kafka_topic_partition_list_new(4);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 3)->offset = 12;
+        rd_kafka_topic_partition_list_add(offsets, "srctopic64", 39)->offset =
+            999999111;
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 0)->offset =
+            999;
+        rd_kafka_topic_partition_list_add(offsets, "srctopic64", 19)->offset =
+            123456789;
+
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_AddPartitionsToTxn, 1,
+            RD_KAFKA_RESP_ERR_UNKNOWN_TOPIC_OR_PART);
+
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_TxnOffsetCommit, 2,
+            RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+            RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS);
+
+        cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+        TEST_CALL_ERROR__(
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1));
+
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+        /*
+         * Commit transaction, first with som failures, then succeed.
+         */
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_EndTxn, 3,
+            RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE,
+            RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+            RD_KAFKA_RESP_ERR_COORDINATOR_LOAD_IN_PROGRESS);
+
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, 5000));
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief KIP-360: Test that fatal idempotence errors triggers abortable
+ *        transaction errors and that the producer can recover.
+ */
+static void do_test_txn_fatal_idempo_errors(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_error_t *error;
+        const char *txnid = "myTxnId";
+
+        SUB_TEST_QUICK();
+
+        rk = create_txn_producer(&mcluster, txnid, 3, "batch.num.messages", "1",
+                                 NULL);
+
+        test_curr->ignore_dr_err = rd_true;
+        test_curr->is_fatal_cb   = error_is_fatal_cb;
+        allowed_error            = RD_KAFKA_RESP_ERR_UNKNOWN_PRODUCER_ID;
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        /*
+         * Start a transaction
+         */
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+
+        /* Produce a message without error first */
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        /* Produce a message, let it fail with a fatal idempo error. */
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_Produce, 1,
+            RD_KAFKA_RESP_ERR_UNKNOWN_PRODUCER_ID);
+
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        /* Commit the transaction, should fail */
+        error = rd_kafka_commit_transaction(rk, -1);
+        TEST_ASSERT(error != NULL, "Expected commit_transaction() to fail");
+
+        TEST_SAY("commit_transaction() failed (expectedly): %s\n",
+                 rd_kafka_error_string(error));
+
+        TEST_ASSERT(!rd_kafka_error_is_fatal(error),
+                    "Did not expect fatal error");
+        TEST_ASSERT(rd_kafka_error_txn_requires_abort(error),
+                    "Expected abortable error");
+        rd_kafka_error_destroy(error);
+
+        /* Abort the transaction */
+        TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, -1));
+
+        /* Run a new transaction without errors to verify that the
+         * producer can recover. */
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        allowed_error = RD_KAFKA_RESP_ERR_NO_ERROR;
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief KIP-360: Test that fatal idempotence errors triggers abortable
+ *        transaction errors, but let the broker-side bumping of the
+ *        producer PID take longer than the remaining transaction timeout
+ *        which should raise a retriable error from abort_transaction().
+ *
+ * @param with_sleep After the first abort sleep longer than it takes to
+ *                   re-init the pid so that the internal state automatically
+ *                   transitions.
+ */
+static void do_test_txn_slow_reinit(rd_bool_t with_sleep) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_error_t *error;
+        int32_t txn_coord = 2;
+        const char *txnid = "myTxnId";
+        test_timing_t timing;
+
+        SUB_TEST("%s sleep", with_sleep ? "with" : "without");
+
+        rk = create_txn_producer(&mcluster, txnid, 3, "batch.num.messages", "1",
+                                 NULL);
+
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", txnid,
+                                      txn_coord);
+
+        test_curr->ignore_dr_err = rd_true;
+        test_curr->is_fatal_cb   = NULL;
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, -1));
+
+        /*
+         * Start a transaction
+         */
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+
+        /* Produce a message without error first */
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        test_flush(rk, -1);
+
+        /* Set transaction coordinator latency higher than
+         * the abort_transaction() call timeout so that the automatic
+         * re-initpid takes longer than abort_transaction(). */
+        rd_kafka_mock_broker_push_request_error_rtts(
+            mcluster, txn_coord, RD_KAFKAP_InitProducerId, 1,
+            RD_KAFKA_RESP_ERR_NO_ERROR, 10000 /*10s*/);
+
+        /* Produce a message, let it fail with a fatal idempo error. */
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_Produce, 1,
+            RD_KAFKA_RESP_ERR_UNKNOWN_PRODUCER_ID);
+
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+
+        /* Commit the transaction, should fail */
+        TIMING_START(&timing, "commit_transaction(-1)");
+        error = rd_kafka_commit_transaction(rk, -1);
+        TIMING_STOP(&timing);
+        TEST_ASSERT(error != NULL, "Expected commit_transaction() to fail");
+
+        TEST_SAY("commit_transaction() failed (expectedly): %s\n",
+                 rd_kafka_error_string(error));
+
+        TEST_ASSERT(!rd_kafka_error_is_fatal(error),
+                    "Did not expect fatal error");
+        TEST_ASSERT(rd_kafka_error_txn_requires_abort(error),
+                    "Expected abortable error");
+        rd_kafka_error_destroy(error);
+
+        /* Abort the transaction, should fail with retriable (timeout) error */
+        TIMING_START(&timing, "abort_transaction(100)");
+        error = rd_kafka_abort_transaction(rk, 100);
+        TIMING_STOP(&timing);
+        TEST_ASSERT(error != NULL, "Expected abort_transaction() to fail");
+
+        TEST_SAY("First abort_transaction() failed: %s\n",
+                 rd_kafka_error_string(error));
+        TEST_ASSERT(!rd_kafka_error_is_fatal(error),
+                    "Did not expect fatal error");
+        TEST_ASSERT(rd_kafka_error_is_retriable(error),
+                    "Expected retriable error");
+        rd_kafka_error_destroy(error);
+
+        if (with_sleep)
+                rd_sleep(12);
+
+        /* Retry abort, should now finish. */
+        TEST_SAY("Retrying abort\n");
+        TIMING_START(&timing, "abort_transaction(-1)");
+        TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, -1));
+        TIMING_STOP(&timing);
+
+        /* Run a new transaction without errors to verify that the
+         * producer can recover. */
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        allowed_error = RD_KAFKA_RESP_ERR_NO_ERROR;
+
+        SUB_TEST_PASS();
+}
+
+
+
+/**
+ * @brief KIP-360: Test that fatal idempotence errors triggers abortable
+ *        transaction errors, but let the broker-side bumping of the
+ *        producer PID fail with a fencing error.
+ *        Should raise a fatal error.
+ *
+ * @param error_code Which error code InitProducerIdRequest should fail with.
+ *                   Either RD_KAFKA_RESP_ERR_INVALID_PRODUCER_EPOCH (older)
+ *                   or RD_KAFKA_RESP_ERR_PRODUCER_FENCED (newer).
+ */
+static void do_test_txn_fenced_reinit(rd_kafka_resp_err_t error_code) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_error_t *error;
+        int32_t txn_coord = 2;
+        const char *txnid = "myTxnId";
+        char errstr[512];
+        rd_kafka_resp_err_t fatal_err;
+
+        SUB_TEST_QUICK("With error %s", rd_kafka_err2name(error_code));
+
+        rk = create_txn_producer(&mcluster, txnid, 3, "batch.num.messages", "1",
+                                 NULL);
+
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", txnid,
+                                      txn_coord);
+
+        test_curr->ignore_dr_err = rd_true;
+        test_curr->is_fatal_cb   = error_is_fatal_cb;
+        allowed_error            = RD_KAFKA_RESP_ERR__FENCED;
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, -1));
+
+        /*
+         * Start a transaction
+         */
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+
+        /* Produce a message without error first */
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        test_flush(rk, -1);
+
+        /* Fail the PID reinit */
+        rd_kafka_mock_broker_push_request_error_rtts(
+            mcluster, txn_coord, RD_KAFKAP_InitProducerId, 1, error_code, 0);
+
+        /* Produce a message, let it fail with a fatal idempo error. */
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_Produce, 1,
+            RD_KAFKA_RESP_ERR_UNKNOWN_PRODUCER_ID);
+
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        test_flush(rk, -1);
+
+        /* Abort the transaction, should fail with a fatal error */
+        error = rd_kafka_abort_transaction(rk, -1);
+        TEST_ASSERT(error != NULL, "Expected abort_transaction() to fail");
+
+        TEST_SAY("abort_transaction() failed: %s\n",
+                 rd_kafka_error_string(error));
+        TEST_ASSERT(rd_kafka_error_is_fatal(error), "Expected a fatal error");
+        rd_kafka_error_destroy(error);
+
+        fatal_err = rd_kafka_fatal_error(rk, errstr, sizeof(errstr));
+        TEST_ASSERT(fatal_err, "Expected a fatal error to have been raised");
+        TEST_SAY("Fatal error: %s: %s\n", rd_kafka_err2name(fatal_err), errstr);
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        allowed_error = RD_KAFKA_RESP_ERR_NO_ERROR;
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Test EndTxn errors.
+ */
+static void do_test_txn_endtxn_errors(void) {
+        rd_kafka_t *rk                    = NULL;
+        rd_kafka_mock_cluster_t *mcluster = NULL;
+        rd_kafka_resp_err_t err;
+        struct {
+                size_t error_cnt;
+                rd_kafka_resp_err_t errors[4];
+                rd_kafka_resp_err_t exp_err;
+                rd_bool_t exp_retriable;
+                rd_bool_t exp_abortable;
+                rd_bool_t exp_fatal;
+                rd_bool_t exp_successful_abort;
+        } scenario[] = {
+            /* This list of errors is from the EndTxnResponse handler in
+             * AK clients/.../TransactionManager.java */
+            {
+                /* #0 */
+                2,
+                {RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE,
+                 RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE},
+                /* Should auto-recover */
+                RD_KAFKA_RESP_ERR_NO_ERROR,
+            },
+            {
+                /* #1 */
+                2,
+                {RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                 RD_KAFKA_RESP_ERR_NOT_COORDINATOR},
+                /* Should auto-recover */
+                RD_KAFKA_RESP_ERR_NO_ERROR,
+            },
+            {
+                /* #2 */
+                1,
+                {RD_KAFKA_RESP_ERR_COORDINATOR_LOAD_IN_PROGRESS},
+                /* Should auto-recover */
+                RD_KAFKA_RESP_ERR_NO_ERROR,
+            },
+            {
+                /* #3 */
+                3,
+                {RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS,
+                 RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS,
+                 RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS},
+                /* Should auto-recover */
+                RD_KAFKA_RESP_ERR_NO_ERROR,
+            },
+            {
+                /* #4: the abort is auto-recovering thru epoch bump */
+                1,
+                {RD_KAFKA_RESP_ERR_UNKNOWN_PRODUCER_ID},
+                RD_KAFKA_RESP_ERR_UNKNOWN_PRODUCER_ID,
+                rd_false /* !retriable */,
+                rd_true /* abortable */,
+                rd_false /* !fatal */,
+                rd_true /* successful abort */
+            },
+            {
+                /* #5: the abort is auto-recovering thru epoch bump */
+                1,
+                {RD_KAFKA_RESP_ERR_INVALID_PRODUCER_ID_MAPPING},
+                RD_KAFKA_RESP_ERR_INVALID_PRODUCER_ID_MAPPING,
+                rd_false /* !retriable */,
+                rd_true /* abortable */,
+                rd_false /* !fatal */,
+                rd_true /* successful abort */
+            },
+            {
+                /* #6 */
+                1,
+                {RD_KAFKA_RESP_ERR_INVALID_PRODUCER_EPOCH},
+                /* This error is normalized */
+                RD_KAFKA_RESP_ERR__FENCED,
+                rd_false /* !retriable */,
+                rd_false /* !abortable */,
+                rd_true /* fatal */
+            },
+            {
+                /* #7 */
+                1,
+                {RD_KAFKA_RESP_ERR_PRODUCER_FENCED},
+                /* This error is normalized */
+                RD_KAFKA_RESP_ERR__FENCED,
+                rd_false /* !retriable */,
+                rd_false /* !abortable */,
+                rd_true /* fatal */
+            },
+            {
+                /* #8 */
+                1,
+                {RD_KAFKA_RESP_ERR_TRANSACTIONAL_ID_AUTHORIZATION_FAILED},
+                RD_KAFKA_RESP_ERR_TRANSACTIONAL_ID_AUTHORIZATION_FAILED,
+                rd_false /* !retriable */,
+                rd_false /* !abortable */,
+                rd_true /* fatal */
+            },
+            {
+                /* #9 */
+                1,
+                {RD_KAFKA_RESP_ERR_GROUP_AUTHORIZATION_FAILED},
+                RD_KAFKA_RESP_ERR_GROUP_AUTHORIZATION_FAILED,
+                rd_false /* !retriable */,
+                rd_true /* abortable */,
+                rd_false /* !fatal */
+            },
+            {
+                /* #10 */
+                /* Any other error should raise a fatal error */
+                1,
+                {RD_KAFKA_RESP_ERR_INVALID_MSG_SIZE},
+                RD_KAFKA_RESP_ERR_INVALID_MSG_SIZE,
+                rd_false /* !retriable */,
+                rd_true /* abortable */,
+                rd_false /* !fatal */,
+            },
+            {
+                /* #11 */
+                1,
+                {RD_KAFKA_RESP_ERR_PRODUCER_FENCED},
+                /* This error is normalized */
+                RD_KAFKA_RESP_ERR__FENCED,
+                rd_false /* !retriable */,
+                rd_false /* !abortable */,
+                rd_true /* fatal */
+            },
+            {0},
+        };
+        int i;
+
+        SUB_TEST_QUICK();
+
+        for (i = 0; scenario[i].error_cnt > 0; i++) {
+                int j;
+                /* For each scenario, test:
+                 *   commit_transaction()
+                 *   flush() + commit_transaction()
+                 *   abort_transaction()
+                 *   flush() + abort_transaction()
+                 */
+                for (j = 0; j < (2 + 2); j++) {
+                        rd_bool_t commit     = j < 2;
+                        rd_bool_t with_flush = j & 1;
+                        rd_bool_t exp_successful_abort =
+                            !commit && scenario[i].exp_successful_abort;
+                        const char *commit_str =
+                            commit ? (with_flush ? "commit&flush" : "commit")
+                                   : (with_flush ? "abort&flush" : "abort");
+                        rd_kafka_topic_partition_list_t *offsets;
+                        rd_kafka_consumer_group_metadata_t *cgmetadata;
+                        rd_kafka_error_t *error;
+                        test_timing_t t_call;
+
+                        TEST_SAY("Testing scenario #%d %s with %" PRIusz
+                                 " injected erorrs, expecting %s\n",
+                                 i, commit_str, scenario[i].error_cnt,
+                                 exp_successful_abort
+                                     ? "successful abort"
+                                     : rd_kafka_err2name(scenario[i].exp_err));
+
+                        if (!rk) {
+                                const char *txnid = "myTxnId";
+                                rk = create_txn_producer(&mcluster, txnid, 3,
+                                                         NULL);
+                                TEST_CALL_ERROR__(
+                                    rd_kafka_init_transactions(rk, 5000));
+                        }
+
+                        /*
+                         * Start transaction
+                         */
+                        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+                        /* Transaction aborts will cause DR errors:
+                         * ignore them. */
+                        test_curr->ignore_dr_err = !commit;
+
+                        /*
+                         * Produce a message.
+                         */
+                        err = rd_kafka_producev(rk, RD_KAFKA_V_TOPIC("mytopic"),
+                                                RD_KAFKA_V_VALUE("hi", 2),
+                                                RD_KAFKA_V_END);
+                        TEST_ASSERT(!err, "produce failed: %s",
+                                    rd_kafka_err2str(err));
+
+                        if (with_flush)
+                                test_flush(rk, -1);
+
+                        /*
+                         * Send some arbitrary offsets.
+                         */
+                        offsets = rd_kafka_topic_partition_list_new(4);
+                        rd_kafka_topic_partition_list_add(offsets, "srctopic4",
+                                                          3)
+                            ->offset = 12;
+                        rd_kafka_topic_partition_list_add(offsets, "srctopic64",
+                                                          60)
+                            ->offset = 99999;
+
+                        cgmetadata =
+                            rd_kafka_consumer_group_metadata_new("mygroupid");
+
+                        TEST_CALL_ERROR__(rd_kafka_send_offsets_to_transaction(
+                            rk, offsets, cgmetadata, -1));
+
+                        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+                        rd_kafka_topic_partition_list_destroy(offsets);
+
+                        /*
+                         * Commit transaction, first with som failures,
+                         * then succeed.
+                         */
+                        rd_kafka_mock_push_request_errors_array(
+                            mcluster, RD_KAFKAP_EndTxn, scenario[i].error_cnt,
+                            scenario[i].errors);
+
+                        TIMING_START(&t_call, "%s", commit_str);
+                        if (commit)
+                                error = rd_kafka_commit_transaction(
+                                    rk, tmout_multip(5000));
+                        else
+                                error = rd_kafka_abort_transaction(
+                                    rk, tmout_multip(5000));
+                        TIMING_STOP(&t_call);
+
+                        if (error)
+                                TEST_SAY(
+                                    "Scenario #%d %s failed: %s: %s "
+                                    "(retriable=%s, req_abort=%s, "
+                                    "fatal=%s)\n",
+                                    i, commit_str, rd_kafka_error_name(error),
+                                    rd_kafka_error_string(error),
+                                    RD_STR_ToF(
+                                        rd_kafka_error_is_retriable(error)),
+                                    RD_STR_ToF(
+                                        rd_kafka_error_txn_requires_abort(
+                                            error)),
+                                    RD_STR_ToF(rd_kafka_error_is_fatal(error)));
+                        else
+                                TEST_SAY("Scenario #%d %s succeeded\n", i,
+                                         commit_str);
+
+                        if (!scenario[i].exp_err || exp_successful_abort) {
+                                TEST_ASSERT(!error,
+                                            "Expected #%d %s to succeed, "
+                                            "got %s",
+                                            i, commit_str,
+                                            rd_kafka_error_string(error));
+                                continue;
+                        }
+
+
+                        TEST_ASSERT(error != NULL, "Expected #%d %s to fail", i,
+                                    commit_str);
+                        TEST_ASSERT(scenario[i].exp_err ==
+                                        rd_kafka_error_code(error),
+                                    "Scenario #%d: expected %s, not %s", i,
+                                    rd_kafka_err2name(scenario[i].exp_err),
+                                    rd_kafka_error_name(error));
+                        TEST_ASSERT(
+                            scenario[i].exp_retriable ==
+                                (rd_bool_t)rd_kafka_error_is_retriable(error),
+                            "Scenario #%d: retriable mismatch", i);
+                        TEST_ASSERT(
+                            scenario[i].exp_abortable ==
+                                (rd_bool_t)rd_kafka_error_txn_requires_abort(
+                                    error),
+                            "Scenario #%d: abortable mismatch", i);
+                        TEST_ASSERT(
+                            scenario[i].exp_fatal ==
+                                (rd_bool_t)rd_kafka_error_is_fatal(error),
+                            "Scenario #%d: fatal mismatch", i);
+
+                        /* Handle errors according to the error flags */
+                        if (rd_kafka_error_is_fatal(error)) {
+                                TEST_SAY("Fatal error, destroying producer\n");
+                                rd_kafka_error_destroy(error);
+                                rd_kafka_destroy(rk);
+                                rk = NULL; /* Will be re-created on the next
+                                            * loop iteration. */
+
+                        } else if (rd_kafka_error_txn_requires_abort(error)) {
+                                rd_kafka_error_destroy(error);
+                                TEST_SAY(
+                                    "Abortable error, "
+                                    "aborting transaction\n");
+                                TEST_CALL_ERROR__(
+                                    rd_kafka_abort_transaction(rk, -1));
+
+                        } else if (rd_kafka_error_is_retriable(error)) {
+                                rd_kafka_error_destroy(error);
+                                TEST_SAY("Retriable error, retrying %s once\n",
+                                         commit_str);
+                                if (commit)
+                                        TEST_CALL_ERROR__(
+                                            rd_kafka_commit_transaction(rk,
+                                                                        5000));
+                                else
+                                        TEST_CALL_ERROR__(
+                                            rd_kafka_abort_transaction(rk,
+                                                                       5000));
+                        } else {
+                                TEST_FAIL(
+                                    "Scenario #%d %s: "
+                                    "Permanent error without enough "
+                                    "hints to proceed: %s\n",
+                                    i, commit_str,
+                                    rd_kafka_error_string(error));
+                        }
+                }
+        }
+
+        /* All done */
+        if (rk)
+                rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Test that the commit/abort works properly with infinite timeout.
+ */
+static void do_test_txn_endtxn_infinite(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster = NULL;
+        const char *txnid                 = "myTxnId";
+        int i;
+
+        SUB_TEST_QUICK();
+
+        rk = create_txn_producer(&mcluster, txnid, 3, NULL);
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        for (i = 0; i < 2; i++) {
+                rd_bool_t commit       = i == 0;
+                const char *commit_str = commit ? "commit" : "abort";
+                rd_kafka_error_t *error;
+                test_timing_t t_call;
+
+                /* Messages will fail on as the transaction fails,
+                 * ignore the DR error */
+                test_curr->ignore_dr_err = rd_true;
+
+                TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+                TEST_CALL_ERR__(rd_kafka_producev(
+                    rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_VALUE("hi", 2),
+                    RD_KAFKA_V_END));
+
+                /*
+                 * Commit/abort transaction, first with som retriable failures,
+                 * then success.
+                 */
+                rd_kafka_mock_push_request_errors(
+                    mcluster, RD_KAFKAP_EndTxn, 10,
+                    RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_COORDINATOR_LOAD_IN_PROGRESS,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR);
+
+                rd_sleep(1);
+
+                TIMING_START(&t_call, "%s_transaction()", commit_str);
+                if (commit)
+                        error = rd_kafka_commit_transaction(rk, -1);
+                else
+                        error = rd_kafka_abort_transaction(rk, -1);
+                TIMING_STOP(&t_call);
+
+                TEST_SAY("%s returned %s\n", commit_str,
+                         error ? rd_kafka_error_string(error) : "success");
+
+                TEST_ASSERT(!error, "Expected %s to succeed, got %s",
+                            commit_str, rd_kafka_error_string(error));
+        }
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+
+/**
+ * @brief Test that the commit/abort user timeout is honoured.
+ */
+static void do_test_txn_endtxn_timeout(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster = NULL;
+        const char *txnid                 = "myTxnId";
+        int i;
+
+        SUB_TEST_QUICK();
+
+        rk = create_txn_producer(&mcluster, txnid, 3, NULL);
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        for (i = 0; i < 2; i++) {
+                rd_bool_t commit       = i == 0;
+                const char *commit_str = commit ? "commit" : "abort";
+                rd_kafka_error_t *error;
+                test_timing_t t_call;
+
+                /* Messages will fail as the transaction fails,
+                 * ignore the DR error */
+                test_curr->ignore_dr_err = rd_true;
+
+                TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+                TEST_CALL_ERR__(rd_kafka_producev(
+                    rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_VALUE("hi", 2),
+                    RD_KAFKA_V_END));
+
+                /*
+                 * Commit/abort transaction, first with some retriable failures
+                 * whos retries exceed the user timeout.
+                 */
+                rd_kafka_mock_push_request_errors(
+                    mcluster, RD_KAFKAP_EndTxn, 10,
+                    RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_COORDINATOR_LOAD_IN_PROGRESS,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR,
+                    RD_KAFKA_RESP_ERR_NOT_COORDINATOR);
+
+                rd_sleep(1);
+
+                TIMING_START(&t_call, "%s_transaction()", commit_str);
+                if (commit)
+                        error = rd_kafka_commit_transaction(rk, 100);
+                else
+                        error = rd_kafka_abort_transaction(rk, 100);
+                TIMING_STOP(&t_call);
+
+                TEST_SAY_ERROR(error, "%s returned: ", commit_str);
+                TEST_ASSERT(error != NULL, "Expected %s to fail", commit_str);
+                TEST_ASSERT(
+                    rd_kafka_error_code(error) == RD_KAFKA_RESP_ERR__TIMED_OUT,
+                    "Expected %s to fail with timeout, not %s: %s", commit_str,
+                    rd_kafka_error_name(error), rd_kafka_error_string(error));
+                TEST_ASSERT(rd_kafka_error_is_retriable(error),
+                            "%s failure should raise a retriable error",
+                            commit_str);
+                rd_kafka_error_destroy(error);
+
+                /* Now call it again with an infinite timeout, should work. */
+                TIMING_START(&t_call, "%s_transaction() nr 2", commit_str);
+                if (commit)
+                        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+                else
+                        TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, -1));
+                TIMING_STOP(&t_call);
+        }
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+
+/**
+ * @brief Test commit/abort inflight timeout behaviour, which should result
+ *        in a retriable error.
+ */
+static void do_test_txn_endtxn_timeout_inflight(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster = NULL;
+        const char *txnid                 = "myTxnId";
+        int32_t coord_id                  = 1;
+        int i;
+
+        SUB_TEST();
+
+        allowed_error          = RD_KAFKA_RESP_ERR__TIMED_OUT;
+        test_curr->is_fatal_cb = error_is_fatal_cb;
+
+        rk = create_txn_producer(&mcluster, txnid, 1, "transaction.timeout.ms",
+                                 "5000", NULL);
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, -1));
+
+        for (i = 0; i < 2; i++) {
+                rd_bool_t commit       = i == 0;
+                const char *commit_str = commit ? "commit" : "abort";
+                rd_kafka_error_t *error;
+                test_timing_t t_call;
+
+                /* Messages will fail as the transaction fails,
+                 * ignore the DR error */
+                test_curr->ignore_dr_err = rd_true;
+
+                TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+                TEST_CALL_ERR__(rd_kafka_producev(
+                    rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_VALUE("hi", 2),
+                    RD_KAFKA_V_END));
+
+                /* Let EndTxn & EndTxn retry timeout */
+                rd_kafka_mock_broker_push_request_error_rtts(
+                    mcluster, coord_id, RD_KAFKAP_EndTxn, 2,
+                    RD_KAFKA_RESP_ERR_NO_ERROR, 10000,
+                    RD_KAFKA_RESP_ERR_NO_ERROR, 10000);
+
+                rd_sleep(1);
+
+                TIMING_START(&t_call, "%s_transaction()", commit_str);
+                if (commit)
+                        error = rd_kafka_commit_transaction(rk, 4000);
+                else
+                        error = rd_kafka_abort_transaction(rk, 4000);
+                TIMING_STOP(&t_call);
+
+                TEST_SAY_ERROR(error, "%s returned: ", commit_str);
+                TEST_ASSERT(error != NULL, "Expected %s to fail", commit_str);
+                TEST_ASSERT(
+                    rd_kafka_error_code(error) == RD_KAFKA_RESP_ERR__TIMED_OUT,
+                    "Expected %s to fail with timeout, not %s: %s", commit_str,
+                    rd_kafka_error_name(error), rd_kafka_error_string(error));
+                TEST_ASSERT(rd_kafka_error_is_retriable(error),
+                            "%s failure should raise a retriable error",
+                            commit_str);
+                rd_kafka_error_destroy(error);
+
+                /* Now call it again with an infinite timeout, should work. */
+                TIMING_START(&t_call, "%s_transaction() nr 2", commit_str);
+                if (commit)
+                        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+                else
+                        TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, -1));
+                TIMING_STOP(&t_call);
+        }
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        allowed_error          = RD_KAFKA_RESP_ERR_NO_ERROR;
+        test_curr->is_fatal_cb = NULL;
+
+        SUB_TEST_PASS();
+}
+
+
+
+/**
+ * @brief Test that EndTxn is properly sent for aborted transactions
+ *        even if AddOffsetsToTxnRequest was retried.
+ *        This is a check for a txn_req_cnt bug.
+ */
+static void do_test_txn_req_cnt(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+        const char *txnid = "myTxnId";
+
+        SUB_TEST_QUICK();
+
+        rk = create_txn_producer(&mcluster, txnid, 3, NULL);
+
+        /* Messages will fail on abort(), ignore the DR error */
+        test_curr->ignore_dr_err = rd_true;
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        /*
+         * Send some arbitrary offsets, first with some failures, then
+         * succeed.
+         */
+        offsets = rd_kafka_topic_partition_list_new(2);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 3)->offset = 12;
+        rd_kafka_topic_partition_list_add(offsets, "srctopic64", 40)->offset =
+            999999111;
+
+        rd_kafka_mock_push_request_errors(mcluster, RD_KAFKAP_AddOffsetsToTxn,
+                                          2,
+                                          RD_KAFKA_RESP_ERR_REQUEST_TIMED_OUT,
+                                          RD_KAFKA_RESP_ERR_NOT_COORDINATOR);
+
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_TxnOffsetCommit, 2,
+            RD_KAFKA_RESP_ERR_COORDINATOR_LOAD_IN_PROGRESS,
+            RD_KAFKA_RESP_ERR_UNKNOWN_TOPIC_OR_PART);
+
+        cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+        TEST_CALL_ERROR__(
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1));
+
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+        TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, 5000));
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Test abortable errors using mock broker error injections
+ *        and code coverage checks.
+ */
+static void do_test_txn_requires_abort_errors(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_error_t *error;
+        rd_kafka_resp_err_t err;
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+        int r;
+
+        SUB_TEST_QUICK();
+
+        rk = create_txn_producer(&mcluster, "txnid", 3, NULL);
+
+        test_curr->ignore_dr_err = rd_true;
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        /*
+         * 1. Fail on produce
+         */
+        TEST_SAY("1. Fail on produce\n");
+
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_Produce, 1,
+            RD_KAFKA_RESP_ERR_TOPIC_AUTHORIZATION_FAILED);
+
+        err = rd_kafka_producev(rk, RD_KAFKA_V_TOPIC("mytopic"),
+                                RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END);
+        TEST_ASSERT(!err, "produce failed: %s", rd_kafka_err2str(err));
+
+        /* Wait for messages to fail */
+        test_flush(rk, 5000);
+
+        /* Any other transactional API should now raise an error */
+        offsets = rd_kafka_topic_partition_list_new(1);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 3)->offset = 12;
+
+        cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+        error =
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1);
+
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+        TEST_ASSERT(error, "expected error");
+        TEST_ASSERT(rd_kafka_error_txn_requires_abort(error),
+                    "expected abortable error, not %s",
+                    rd_kafka_error_string(error));
+        TEST_SAY("Error %s: %s\n", rd_kafka_error_name(error),
+                 rd_kafka_error_string(error));
+        rd_kafka_error_destroy(error);
+
+        TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, -1));
+
+        /*
+         * 2. Restart transaction and fail on AddPartitionsToTxn
+         */
+        TEST_SAY("2. Fail on AddPartitionsToTxn\n");
+
+        /* First refresh proper Metadata to clear the topic's auth error,
+         * otherwise the produce() below will fail immediately. */
+        r = test_get_partition_count(rk, "mytopic", 5000);
+        TEST_ASSERT(r > 0, "Expected topic %s to exist", "mytopic");
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_AddPartitionsToTxn, 1,
+            RD_KAFKA_RESP_ERR_TOPIC_AUTHORIZATION_FAILED);
+
+        err = rd_kafka_producev(rk, RD_KAFKA_V_TOPIC("mytopic"),
+                                RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END);
+        TEST_ASSERT(!err, "produce failed: %s", rd_kafka_err2str(err));
+
+        error = rd_kafka_commit_transaction(rk, 5000);
+        TEST_ASSERT(error, "commit_transaction should have failed");
+        TEST_SAY("commit_transaction() error %s: %s\n",
+                 rd_kafka_error_name(error), rd_kafka_error_string(error));
+        rd_kafka_error_destroy(error);
+
+        TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, -1));
+
+        /*
+         * 3. Restart transaction and fail on AddOffsetsToTxn
+         */
+        TEST_SAY("3. Fail on AddOffsetsToTxn\n");
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        err = rd_kafka_producev(rk, RD_KAFKA_V_TOPIC("mytopic"),
+                                RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END);
+        TEST_ASSERT(!err, "produce failed: %s", rd_kafka_err2str(err));
+
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_AddOffsetsToTxn, 1,
+            RD_KAFKA_RESP_ERR_GROUP_AUTHORIZATION_FAILED);
+
+        offsets = rd_kafka_topic_partition_list_new(1);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 3)->offset = 12;
+        cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+        error =
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1);
+        TEST_ASSERT(error, "Expected send_offsets..() to fail");
+        TEST_ASSERT(rd_kafka_error_code(error) ==
+                        RD_KAFKA_RESP_ERR_GROUP_AUTHORIZATION_FAILED,
+                    "expected send_offsets_to_transaction() to fail with "
+                    "group auth error: not %s",
+                    rd_kafka_error_name(error));
+        rd_kafka_error_destroy(error);
+
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+
+        error = rd_kafka_commit_transaction(rk, 5000);
+        TEST_ASSERT(error, "commit_transaction should have failed");
+        rd_kafka_error_destroy(error);
+
+        TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, -1));
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Test error handling and recover for when broker goes down during
+ *        an ongoing transaction.
+ */
+static void do_test_txn_broker_down_in_txn(rd_bool_t down_coord) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        int32_t coord_id, leader_id, down_id;
+        const char *down_what;
+        rd_kafka_resp_err_t err;
+        const char *topic            = "test";
+        const char *transactional_id = "txnid";
+        int msgcnt                   = 1000;
+        int remains                  = 0;
+
+        /* Assign coordinator and leader to two different brokers */
+        coord_id  = 1;
+        leader_id = 2;
+        if (down_coord) {
+                down_id   = coord_id;
+                down_what = "coordinator";
+        } else {
+                down_id   = leader_id;
+                down_what = "leader";
+        }
+
+        SUB_TEST_QUICK("Test %s down", down_what);
+
+        rk = create_txn_producer(&mcluster, transactional_id, 3, NULL);
+
+        /* Broker down is not a test-failing error */
+        allowed_error          = RD_KAFKA_RESP_ERR__TRANSPORT;
+        test_curr->is_fatal_cb = error_is_fatal_cb;
+
+        err = rd_kafka_mock_topic_create(mcluster, topic, 1, 3);
+        TEST_ASSERT(!err, "Failed to create topic: %s", rd_kafka_err2str(err));
+
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", transactional_id,
+                                      coord_id);
+        rd_kafka_mock_partition_set_leader(mcluster, topic, 0, leader_id);
+
+        /* Start transactioning */
+        TEST_SAY("Starting transaction\n");
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        test_produce_msgs2_nowait(rk, topic, 0, RD_KAFKA_PARTITION_UA, 0,
+                                  msgcnt / 2, NULL, 0, &remains);
+
+        TEST_SAY("Bringing down %s %" PRId32 "\n", down_what, down_id);
+        rd_kafka_mock_broker_set_down(mcluster, down_id);
+
+        rd_kafka_flush(rk, 3000);
+
+        /* Produce remaining messages */
+        test_produce_msgs2_nowait(rk, topic, 0, RD_KAFKA_PARTITION_UA,
+                                  msgcnt / 2, msgcnt / 2, NULL, 0, &remains);
+
+        rd_sleep(2);
+
+        TEST_SAY("Bringing up %s %" PRId32 "\n", down_what, down_id);
+        rd_kafka_mock_broker_set_up(mcluster, down_id);
+
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+
+        TEST_ASSERT(remains == 0, "%d message(s) were not produced\n", remains);
+
+        rd_kafka_destroy(rk);
+
+        test_curr->is_fatal_cb = NULL;
+
+        SUB_TEST_PASS();
+}
+
+
+
+/**
+ * @brief Advance the coord_id to the next broker.
+ */
+static void set_next_coord(rd_kafka_mock_cluster_t *mcluster,
+                           const char *transactional_id,
+                           int broker_cnt,
+                           int32_t *coord_idp) {
+        int32_t new_coord_id;
+
+        new_coord_id = 1 + ((*coord_idp) % (broker_cnt));
+        TEST_SAY("Changing transaction coordinator from %" PRId32 " to %" PRId32
+                 "\n",
+                 *coord_idp, new_coord_id);
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", transactional_id,
+                                      new_coord_id);
+
+        *coord_idp = new_coord_id;
+}
+
+/**
+ * @brief Switch coordinator during a transaction.
+ *
+ */
+static void do_test_txn_switch_coordinator(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        int32_t coord_id;
+        const char *topic            = "test";
+        const char *transactional_id = "txnid";
+        const int broker_cnt         = 5;
+        const int iterations         = 20;
+        int i;
+
+        test_timeout_set(iterations * 10);
+
+        SUB_TEST("Test switching coordinators");
+
+        rk = create_txn_producer(&mcluster, transactional_id, broker_cnt, NULL);
+
+        coord_id = 1;
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", transactional_id,
+                                      coord_id);
+
+        /* Start transactioning */
+        TEST_SAY("Starting transaction\n");
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        for (i = 0; i < iterations; i++) {
+                const int msgcnt = 100;
+                int remains      = 0;
+
+                set_next_coord(mcluster, transactional_id, broker_cnt,
+                               &coord_id);
+
+                TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+                test_produce_msgs2(rk, topic, 0, RD_KAFKA_PARTITION_UA, 0,
+                                   msgcnt / 2, NULL, 0);
+
+                if (!(i % 3))
+                        set_next_coord(mcluster, transactional_id, broker_cnt,
+                                       &coord_id);
+
+                /* Produce remaining messages */
+                test_produce_msgs2_nowait(rk, topic, 0, RD_KAFKA_PARTITION_UA,
+                                          msgcnt / 2, msgcnt / 2, NULL, 0,
+                                          &remains);
+
+                if ((i & 1) || !(i % 8))
+                        set_next_coord(mcluster, transactional_id, broker_cnt,
+                                       &coord_id);
+
+
+                if (!(i % 5)) {
+                        test_curr->ignore_dr_err = rd_false;
+                        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+
+                } else {
+                        test_curr->ignore_dr_err = rd_true;
+                        TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, -1));
+                }
+        }
+
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Switch coordinator during a transaction when AddOffsetsToTxn
+ *        are sent. #3571.
+ */
+static void do_test_txn_switch_coordinator_refresh(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        const char *topic            = "test";
+        const char *transactional_id = "txnid";
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+
+        SUB_TEST("Test switching coordinators (refresh)");
+
+        rk = create_txn_producer(&mcluster, transactional_id, 3, NULL);
+
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", transactional_id,
+                                      1);
+
+        /* Start transactioning */
+        TEST_SAY("Starting transaction\n");
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        /* Switch the coordinator so that AddOffsetsToTxnRequest
+         * will respond with NOT_COORDINATOR. */
+        TEST_SAY("Switching to coordinator 2\n");
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", transactional_id,
+                                      2);
+
+        /*
+         * Send some arbitrary offsets.
+         */
+        offsets = rd_kafka_topic_partition_list_new(4);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 3)->offset = 12;
+        rd_kafka_topic_partition_list_add(offsets, "srctopic64", 29)->offset =
+            99999;
+
+        cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+        TEST_CALL_ERROR__(rd_kafka_send_offsets_to_transaction(
+            rk, offsets, cgmetadata, 20 * 1000));
+
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+
+        /* Produce some messages */
+        test_produce_msgs2(rk, topic, 0, RD_KAFKA_PARTITION_UA, 0, 10, NULL, 0);
+
+        /* And commit the transaction */
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Test fatal error handling when transactions are not supported
+ *        by the broker.
+ */
+static void do_test_txns_not_supported(void) {
+        rd_kafka_t *rk;
+        rd_kafka_conf_t *conf;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_error_t *error;
+        rd_kafka_resp_err_t err;
+
+        SUB_TEST_QUICK();
+
+        test_conf_init(&conf, NULL, 10);
+
+        test_conf_set(conf, "transactional.id", "myxnid");
+        test_conf_set(conf, "bootstrap.servers", ",");
+        rd_kafka_conf_set_dr_msg_cb(conf, test_dr_msg_cb);
+
+        rk = test_create_handle(RD_KAFKA_PRODUCER, conf);
+
+        /* Create mock cluster */
+        mcluster = rd_kafka_mock_cluster_new(rk, 3);
+
+        /* Disable InitProducerId */
+        rd_kafka_mock_set_apiversion(mcluster, 22 /*InitProducerId*/, -1, -1);
+
+
+        rd_kafka_brokers_add(rk, rd_kafka_mock_cluster_bootstraps(mcluster));
+
+
+
+        error = rd_kafka_init_transactions(rk, 5 * 1000);
+        TEST_SAY("init_transactions() returned %s: %s\n",
+                 error ? rd_kafka_error_name(error) : "success",
+                 error ? rd_kafka_error_string(error) : "success");
+
+        TEST_ASSERT(error, "Expected init_transactions() to fail");
+        TEST_ASSERT(rd_kafka_error_code(error) ==
+                        RD_KAFKA_RESP_ERR__UNSUPPORTED_FEATURE,
+                    "Expected init_transactions() to fail with %s, not %s: %s",
+                    rd_kafka_err2name(RD_KAFKA_RESP_ERR__UNSUPPORTED_FEATURE),
+                    rd_kafka_error_name(error), rd_kafka_error_string(error));
+        rd_kafka_error_destroy(error);
+
+        err = rd_kafka_producev(rk, RD_KAFKA_V_TOPIC("test"),
+                                RD_KAFKA_V_KEY("test", 4), RD_KAFKA_V_END);
+        TEST_ASSERT(err == RD_KAFKA_RESP_ERR__FATAL,
+                    "Expected producev() to fail with %s, not %s",
+                    rd_kafka_err2name(RD_KAFKA_RESP_ERR__FATAL),
+                    rd_kafka_err2name(err));
+
+        rd_kafka_mock_cluster_destroy(mcluster);
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief CONCURRENT_TRANSACTION on AddOffsets.. should be retried.
+ */
+static void do_test_txns_send_offsets_concurrent_is_retried(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_resp_err_t err;
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+
+        SUB_TEST_QUICK();
+
+        rk = create_txn_producer(&mcluster, "txnid", 3, NULL);
+
+        test_curr->ignore_dr_err = rd_true;
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        err = rd_kafka_producev(rk, RD_KAFKA_V_TOPIC("mytopic"),
+                                RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END);
+        TEST_ASSERT(!err, "produce failed: %s", rd_kafka_err2str(err));
+
+        /* Wait for messages to be delivered */
+        test_flush(rk, 5000);
+
+
+        /*
+         * Have AddOffsetsToTxn fail but eventually succeed due to
+         * infinite retries.
+         */
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_AddOffsetsToTxn,
+            1 + 5, /* first request + some retries */
+            RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS,
+            RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS,
+            RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS,
+            RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS,
+            RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS,
+            RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS);
+
+        offsets = rd_kafka_topic_partition_list_new(1);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 3)->offset = 12;
+
+        cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+        TEST_CALL_ERROR__(
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1));
+
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, 5000));
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Verify that send_offsets_to_transaction() with no eligible offsets
+ *        is handled properly - the call should succeed immediately and be
+ *        repeatable.
+ */
+static void do_test_txns_send_offsets_non_eligible(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_resp_err_t err;
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+
+        SUB_TEST_QUICK();
+
+        rk = create_txn_producer(&mcluster, "txnid", 3, NULL);
+
+        test_curr->ignore_dr_err = rd_true;
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        err = rd_kafka_producev(rk, RD_KAFKA_V_TOPIC("mytopic"),
+                                RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END);
+        TEST_ASSERT(!err, "produce failed: %s", rd_kafka_err2str(err));
+
+        /* Wait for messages to be delivered */
+        test_flush(rk, 5000);
+
+        /* Empty offsets list */
+        offsets = rd_kafka_topic_partition_list_new(0);
+
+        cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+        TEST_CALL_ERROR__(
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1));
+
+        /* Now call it again, should also succeed. */
+        TEST_CALL_ERROR__(
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1));
+
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, 5000));
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Verify that request timeouts don't cause crash (#2913).
+ */
+static void do_test_txns_no_timeout_crash(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_error_t *error;
+        rd_kafka_resp_err_t err;
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+
+        SUB_TEST_QUICK();
+
+        rk =
+            create_txn_producer(&mcluster, "txnid", 3, "socket.timeout.ms",
+                                "1000", "transaction.timeout.ms", "5000", NULL);
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        err = rd_kafka_producev(rk, RD_KAFKA_V_TOPIC("mytopic"),
+                                RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END);
+        TEST_ASSERT(!err, "produce failed: %s", rd_kafka_err2str(err));
+
+        test_flush(rk, -1);
+
+        /* Delay all broker connections */
+        if ((err = rd_kafka_mock_broker_set_rtt(mcluster, 1, 2000)) ||
+            (err = rd_kafka_mock_broker_set_rtt(mcluster, 2, 2000)) ||
+            (err = rd_kafka_mock_broker_set_rtt(mcluster, 3, 2000)))
+                TEST_FAIL("Failed to set broker RTT: %s",
+                          rd_kafka_err2str(err));
+
+        /* send_offsets..() should now time out */
+        offsets = rd_kafka_topic_partition_list_new(1);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 3)->offset = 12;
+        cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+        error =
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1);
+        TEST_ASSERT(error, "Expected send_offsets..() to fail");
+        TEST_SAY("send_offsets..() failed with %serror: %s\n",
+                 rd_kafka_error_is_retriable(error) ? "retriable " : "",
+                 rd_kafka_error_string(error));
+        TEST_ASSERT(rd_kafka_error_code(error) == RD_KAFKA_RESP_ERR__TIMED_OUT,
+                    "expected send_offsets_to_transaction() to fail with "
+                    "timeout, not %s",
+                    rd_kafka_error_name(error));
+        TEST_ASSERT(rd_kafka_error_is_retriable(error),
+                    "expected send_offsets_to_transaction() to fail with "
+                    "a retriable error");
+        rd_kafka_error_destroy(error);
+
+        /* Reset delay and try again */
+        if ((err = rd_kafka_mock_broker_set_rtt(mcluster, 1, 0)) ||
+            (err = rd_kafka_mock_broker_set_rtt(mcluster, 2, 0)) ||
+            (err = rd_kafka_mock_broker_set_rtt(mcluster, 3, 0)))
+                TEST_FAIL("Failed to reset broker RTT: %s",
+                          rd_kafka_err2str(err));
+
+        TEST_SAY("Retrying send_offsets..()\n");
+        error =
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1);
+        TEST_ASSERT(!error, "Expected send_offsets..() to succeed, got: %s",
+                    rd_kafka_error_string(error));
+
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+        /* All done */
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Test auth failure handling.
+ */
+static void do_test_txn_auth_failure(int16_t ApiKey,
+                                     rd_kafka_resp_err_t ErrorCode) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_error_t *error;
+
+        SUB_TEST_QUICK("ApiKey=%s ErrorCode=%s", rd_kafka_ApiKey2str(ApiKey),
+                       rd_kafka_err2name(ErrorCode));
+
+        rk = create_txn_producer(&mcluster, "txnid", 3, NULL);
+
+        rd_kafka_mock_push_request_errors(mcluster, ApiKey, 1, ErrorCode);
+
+        error = rd_kafka_init_transactions(rk, 5000);
+        TEST_ASSERT(error, "Expected init_transactions() to fail");
+
+        TEST_SAY("init_transactions() failed: %s: %s\n",
+                 rd_kafka_err2name(rd_kafka_error_code(error)),
+                 rd_kafka_error_string(error));
+        TEST_ASSERT(rd_kafka_error_code(error) == ErrorCode,
+                    "Expected error %s, not %s", rd_kafka_err2name(ErrorCode),
+                    rd_kafka_err2name(rd_kafka_error_code(error)));
+        TEST_ASSERT(rd_kafka_error_is_fatal(error),
+                    "Expected error to be fatal");
+        TEST_ASSERT(!rd_kafka_error_is_retriable(error),
+                    "Expected error to not be retriable");
+        rd_kafka_error_destroy(error);
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Issue #3041: Commit fails due to message flush() taking too long,
+ *        eventually resulting in an unabortable error and failure to
+ *        re-init the transactional producer.
+ */
+static void do_test_txn_flush_timeout(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+        rd_kafka_error_t *error;
+        const char *txnid      = "myTxnId";
+        const char *topic      = "myTopic";
+        const int32_t coord_id = 2;
+        int msgcounter         = 0;
+        rd_bool_t is_retry     = rd_false;
+
+        SUB_TEST_QUICK();
+
+        rk = create_txn_producer(&mcluster, txnid, 3, "message.timeout.ms",
+                                 "10000", "transaction.timeout.ms", "10000",
+                                 /* Speed up coordinator reconnect */
+                                 "reconnect.backoff.max.ms", "1000", NULL);
+
+
+        /* Broker down is not a test-failing error */
+        test_curr->is_fatal_cb = error_is_fatal_cb;
+        allowed_error          = RD_KAFKA_RESP_ERR__TRANSPORT;
+
+        rd_kafka_mock_topic_create(mcluster, topic, 2, 3);
+
+        /* Set coordinator so we can disconnect it later */
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", txnid, coord_id);
+
+        /*
+         * Init transactions
+         */
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+retry:
+        if (!is_retry) {
+                /* First attempt should fail. */
+
+                test_curr->ignore_dr_err = rd_true;
+                test_curr->exp_dr_err    = RD_KAFKA_RESP_ERR__MSG_TIMED_OUT;
+
+                /* Assign invalid partition leaders for some partitions so
+                 * that messages will not be delivered. */
+                rd_kafka_mock_partition_set_leader(mcluster, topic, 0, -1);
+                rd_kafka_mock_partition_set_leader(mcluster, topic, 1, -1);
+
+        } else {
+                /* The retry should succeed */
+                test_curr->ignore_dr_err = rd_false;
+                test_curr->exp_dr_err    = is_retry
+                                               ? RD_KAFKA_RESP_ERR_NO_ERROR
+                                               : RD_KAFKA_RESP_ERR__MSG_TIMED_OUT;
+
+                rd_kafka_mock_partition_set_leader(mcluster, topic, 0, 1);
+                rd_kafka_mock_partition_set_leader(mcluster, topic, 1, 1);
+        }
+
+
+        /*
+         * Start a transaction
+         */
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        /*
+         * Produce some messages to specific partitions and random.
+         */
+        test_produce_msgs2_nowait(rk, topic, 0, 0, 0, 100, NULL, 10,
+                                  &msgcounter);
+        test_produce_msgs2_nowait(rk, topic, 1, 0, 0, 100, NULL, 10,
+                                  &msgcounter);
+        test_produce_msgs2_nowait(rk, topic, RD_KAFKA_PARTITION_UA, 0, 0, 100,
+                                  NULL, 10, &msgcounter);
+
+
+        /*
+         * Send some arbitrary offsets.
+         */
+        offsets = rd_kafka_topic_partition_list_new(4);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 3)->offset = 12;
+        rd_kafka_topic_partition_list_add(offsets, "srctopic64", 49)->offset =
+            999999111;
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 0)->offset =
+            999;
+        rd_kafka_topic_partition_list_add(offsets, "srctopic64", 34)->offset =
+            123456789;
+
+        cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+        TEST_CALL_ERROR__(
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1));
+
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+        rd_sleep(2);
+
+        if (!is_retry) {
+                /* Now disconnect the coordinator. */
+                TEST_SAY("Disconnecting transaction coordinator %" PRId32 "\n",
+                         coord_id);
+                rd_kafka_mock_broker_set_down(mcluster, coord_id);
+        }
+
+        /*
+         * Start committing.
+         */
+        error = rd_kafka_commit_transaction(rk, -1);
+
+        if (!is_retry) {
+                TEST_ASSERT(error != NULL, "Expected commit to fail");
+                TEST_SAY("commit_transaction() failed (expectedly): %s\n",
+                         rd_kafka_error_string(error));
+                rd_kafka_error_destroy(error);
+
+        } else {
+                TEST_ASSERT(!error, "Expected commit to succeed, not: %s",
+                            rd_kafka_error_string(error));
+        }
+
+        if (!is_retry) {
+                /*
+                 * Bring the coordinator back up.
+                 */
+                rd_kafka_mock_broker_set_up(mcluster, coord_id);
+                rd_sleep(2);
+
+                /*
+                 * Abort, and try again, this time without error.
+                 */
+                TEST_SAY("Aborting and retrying\n");
+                is_retry = rd_true;
+
+                TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, 60000));
+                goto retry;
+        }
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief ESC-4424: rko is reused in response handler after destroy in coord_req
+ *        sender due to bad state.
+ *
+ * This is somewhat of a race condition so we need to perform a couple of
+ * iterations before it hits, usually 2 or 3, so we try at least 15 times.
+ */
+static void do_test_txn_coord_req_destroy(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        int i;
+        int errcnt = 0;
+
+        SUB_TEST();
+
+        rk = create_txn_producer(&mcluster, "txnid", 3, NULL);
+
+        test_curr->ignore_dr_err = rd_true;
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        for (i = 0; i < 15; i++) {
+                rd_kafka_error_t *error;
+                rd_kafka_resp_err_t err;
+                rd_kafka_topic_partition_list_t *offsets;
+                rd_kafka_consumer_group_metadata_t *cgmetadata;
+
+                test_timeout_set(10);
+
+                TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+                /*
+                 * Inject errors to trigger retries
+                 */
+                rd_kafka_mock_push_request_errors(
+                    mcluster, RD_KAFKAP_AddPartitionsToTxn,
+                    2, /* first request + number of internal retries */
+                    RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS,
+                    RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS);
+
+                rd_kafka_mock_push_request_errors(
+                    mcluster, RD_KAFKAP_AddOffsetsToTxn,
+                    3, /* first request + number of internal retries */
+                    RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS,
+                    RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS,
+                    RD_KAFKA_RESP_ERR_CONCURRENT_TRANSACTIONS);
+
+                err = rd_kafka_producev(rk, RD_KAFKA_V_TOPIC("mytopic"),
+                                        RD_KAFKA_V_VALUE("hi", 2),
+                                        RD_KAFKA_V_END);
+                TEST_ASSERT(!err, "produce failed: %s", rd_kafka_err2str(err));
+
+                rd_kafka_mock_push_request_errors(
+                    mcluster, RD_KAFKAP_Produce, 4,
+                    RD_KAFKA_RESP_ERR_REQUEST_TIMED_OUT,
+                    RD_KAFKA_RESP_ERR_REQUEST_TIMED_OUT,
+                    RD_KAFKA_RESP_ERR_TOPIC_AUTHORIZATION_FAILED,
+                    RD_KAFKA_RESP_ERR_TOPIC_AUTHORIZATION_FAILED);
+                /* FIXME: When KIP-360 is supported, add this error:
+                 *        RD_KAFKA_RESP_ERR_OUT_OF_ORDER_SEQUENCE_NUMBER */
+
+                err = rd_kafka_producev(rk, RD_KAFKA_V_TOPIC("mytopic"),
+                                        RD_KAFKA_V_VALUE("hi", 2),
+                                        RD_KAFKA_V_END);
+                TEST_ASSERT(!err, "produce failed: %s", rd_kafka_err2str(err));
+
+
+                /*
+                 * Send offsets to transaction
+                 */
+
+                offsets = rd_kafka_topic_partition_list_new(1);
+                rd_kafka_topic_partition_list_add(offsets, "srctopic4", 3)
+                    ->offset = 12;
+
+                cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+                error = rd_kafka_send_offsets_to_transaction(rk, offsets,
+                                                             cgmetadata, -1);
+
+                TEST_SAY("send_offsets_to_transaction() #%d: %s\n", i,
+                         rd_kafka_error_string(error));
+
+                /* As we can't control the exact timing and sequence
+                 * of requests this sometimes fails and sometimes succeeds,
+                 * but we run the test enough times to trigger at least
+                 * one failure. */
+                if (error) {
+                        TEST_SAY(
+                            "send_offsets_to_transaction() #%d "
+                            "failed (expectedly): %s\n",
+                            i, rd_kafka_error_string(error));
+                        TEST_ASSERT(rd_kafka_error_txn_requires_abort(error),
+                                    "Expected abortable error for #%d", i);
+                        rd_kafka_error_destroy(error);
+                        errcnt++;
+                }
+
+                rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+                rd_kafka_topic_partition_list_destroy(offsets);
+
+                /* Allow time for internal retries */
+                rd_sleep(2);
+
+                TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, 5000));
+        }
+
+        TEST_SAY("Got %d errors\n", errcnt);
+        TEST_ASSERT(errcnt > 0,
+                    "Expected at least one send_offets_to_transaction() "
+                    "failure");
+
+        /* All done */
+
+        rd_kafka_destroy(rk);
+}
+
+
+static rd_atomic32_t multi_find_req_cnt;
+
+static rd_kafka_resp_err_t
+multi_find_on_response_received_cb(rd_kafka_t *rk,
+                                   int sockfd,
+                                   const char *brokername,
+                                   int32_t brokerid,
+                                   int16_t ApiKey,
+                                   int16_t ApiVersion,
+                                   int32_t CorrId,
+                                   size_t size,
+                                   int64_t rtt,
+                                   rd_kafka_resp_err_t err,
+                                   void *ic_opaque) {
+        rd_kafka_mock_cluster_t *mcluster = rd_kafka_handle_mock_cluster(rk);
+        rd_bool_t done = rd_atomic32_get(&multi_find_req_cnt) > 10000;
+
+        if (ApiKey != RD_KAFKAP_AddOffsetsToTxn || done)
+                return RD_KAFKA_RESP_ERR_NO_ERROR;
+
+        TEST_SAY("on_response_received_cb: %s: %s: brokerid %" PRId32
+                 ", ApiKey %hd, CorrId %d, rtt %.2fms, %s: %s\n",
+                 rd_kafka_name(rk), brokername, brokerid, ApiKey, CorrId,
+                 rtt != -1 ? (float)rtt / 1000.0 : 0.0,
+                 done ? "already done" : "not done yet",
+                 rd_kafka_err2name(err));
+
+
+        if (rd_atomic32_add(&multi_find_req_cnt, 1) == 1) {
+                /* Trigger a broker down/up event, which in turns
+                 * triggers the coord_req_fsm(). */
+                rd_kafka_mock_broker_set_down(mcluster, 2);
+                rd_kafka_mock_broker_set_up(mcluster, 2);
+                return RD_KAFKA_RESP_ERR_NO_ERROR;
+        }
+
+        /* Trigger a broker down/up event, which in turns
+         * triggers the coord_req_fsm(). */
+        rd_kafka_mock_broker_set_down(mcluster, 3);
+        rd_kafka_mock_broker_set_up(mcluster, 3);
+
+        /* Clear the downed broker's latency so that it reconnects
+         * quickly, otherwise the ApiVersionRequest will be delayed and
+         * this will in turn delay the -> UP transition that we need to
+         * trigger the coord_reqs. */
+        rd_kafka_mock_broker_set_rtt(mcluster, 3, 0);
+
+        /* Only do this down/up once */
+        rd_atomic32_add(&multi_find_req_cnt, 10000);
+
+        return RD_KAFKA_RESP_ERR_NO_ERROR;
+}
+
+
+/**
+ * @brief ESC-4444: multiple FindCoordinatorRequests are sent referencing
+ *        the same coord_req_t, but the first one received will destroy
+ *        the coord_req_t object and make the subsequent FindCoordingResponses
+ *        reference a freed object.
+ *
+ * What we want to achieve is this sequence:
+ *  1. AddOffsetsToTxnRequest + Response which..
+ *  2. Triggers TxnOffsetCommitRequest, but the coordinator is not known, so..
+ *  3. Triggers a FindCoordinatorRequest
+ *  4. FindCoordinatorResponse from 3 is received ..
+ *  5. A TxnOffsetCommitRequest is sent from coord_req_fsm().
+ *  6. Another broker changing state to Up triggers coord reqs again, which..
+ *  7. Triggers a second TxnOffsetCommitRequest from coord_req_fsm().
+ *  7. FindCoordinatorResponse from 5 is received, references the destroyed rko
+ *     and crashes.
+ */
+static void do_test_txn_coord_req_multi_find(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_error_t *error;
+        rd_kafka_resp_err_t err;
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+        const char *txnid = "txnid", *groupid = "mygroupid", *topic = "mytopic";
+        int i;
+
+        SUB_TEST();
+
+        rd_atomic32_init(&multi_find_req_cnt, 0);
+
+        on_response_received_cb = multi_find_on_response_received_cb;
+        rk                      = create_txn_producer(&mcluster, txnid, 3,
+                                                      /* Need connections to all brokers so we
+                                                       * can trigger coord_req_fsm events
+                                                       * by toggling connections. */
+                                                      "enable.sparse.connections", "false",
+                                                      /* Set up on_response_received interceptor */
+                                                      "on_response_received", "", NULL);
+
+        /* Let broker 1 be both txn and group coordinator
+         * so that the group coordinator connection is up when it is time
+         * send the TxnOffsetCommitRequest. */
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", txnid, 1);
+        rd_kafka_mock_coordinator_set(mcluster, "group", groupid, 1);
+
+        /* Set broker 1, 2, and 3 as leaders for a partition each and
+         * later produce to both partitions so we know there's a connection
+         * to all brokers. */
+        rd_kafka_mock_topic_create(mcluster, topic, 3, 1);
+        rd_kafka_mock_partition_set_leader(mcluster, topic, 0, 1);
+        rd_kafka_mock_partition_set_leader(mcluster, topic, 1, 2);
+        rd_kafka_mock_partition_set_leader(mcluster, topic, 2, 3);
+
+        /* Broker down is not a test-failing error */
+        allowed_error          = RD_KAFKA_RESP_ERR__TRANSPORT;
+        test_curr->is_fatal_cb = error_is_fatal_cb;
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        for (i = 0; i < 3; i++) {
+                err = rd_kafka_producev(
+                    rk, RD_KAFKA_V_TOPIC(topic), RD_KAFKA_V_PARTITION(i),
+                    RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END);
+                TEST_ASSERT(!err, "produce failed: %s", rd_kafka_err2str(err));
+        }
+
+        test_flush(rk, 5000);
+
+        /*
+         * send_offsets_to_transaction() will query for the group coordinator,
+         * we need to make those requests slow so that multiple requests are
+         * sent.
+         */
+        for (i = 1; i <= 3; i++)
+                rd_kafka_mock_broker_set_rtt(mcluster, (int32_t)i, 4000);
+
+        /*
+         * Send offsets to transaction
+         */
+
+        offsets = rd_kafka_topic_partition_list_new(1);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 3)->offset = 12;
+
+        cgmetadata = rd_kafka_consumer_group_metadata_new(groupid);
+
+        error =
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1);
+
+        TEST_SAY("send_offsets_to_transaction() %s\n",
+                 rd_kafka_error_string(error));
+        TEST_ASSERT(!error, "send_offsets_to_transaction() failed: %s",
+                    rd_kafka_error_string(error));
+
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+        /* Clear delay */
+        for (i = 1; i <= 3; i++)
+                rd_kafka_mock_broker_set_rtt(mcluster, (int32_t)i, 0);
+
+        rd_sleep(5);
+
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, 5000));
+
+        /* All done */
+
+        TEST_ASSERT(rd_atomic32_get(&multi_find_req_cnt) > 10000,
+                    "on_request_sent interceptor did not trigger properly");
+
+        rd_kafka_destroy(rk);
+
+        on_response_received_cb = NULL;
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief ESC-4410: adding producer partitions gradually will trigger multiple
+ *        AddPartitionsToTxn requests. Due to a bug the third partition to be
+ *        registered would hang in PEND_TXN state.
+ *
+ * Trigger this behaviour by having two outstanding AddPartitionsToTxn requests
+ * at the same time, followed by a need for a third:
+ *
+ * 1. Set coordinator broker rtt high (to give us time to produce).
+ * 2. Produce to partition 0, will trigger first AddPartitionsToTxn.
+ * 3. Produce to partition 1, will trigger second AddPartitionsToTxn.
+ * 4. Wait for second AddPartitionsToTxn response.
+ * 5. Produce to partition 2, should trigger AddPartitionsToTxn, but bug
+ *    causes it to be stale in pending state.
+ */
+
+static rd_atomic32_t multi_addparts_resp_cnt;
+static rd_kafka_resp_err_t
+multi_addparts_response_received_cb(rd_kafka_t *rk,
+                                    int sockfd,
+                                    const char *brokername,
+                                    int32_t brokerid,
+                                    int16_t ApiKey,
+                                    int16_t ApiVersion,
+                                    int32_t CorrId,
+                                    size_t size,
+                                    int64_t rtt,
+                                    rd_kafka_resp_err_t err,
+                                    void *ic_opaque) {
+
+        if (ApiKey == RD_KAFKAP_AddPartitionsToTxn) {
+                TEST_SAY("on_response_received_cb: %s: %s: brokerid %" PRId32
+                         ", ApiKey %hd, CorrId %d, rtt %.2fms, count %" PRId32
+                         ": %s\n",
+                         rd_kafka_name(rk), brokername, brokerid, ApiKey,
+                         CorrId, rtt != -1 ? (float)rtt / 1000.0 : 0.0,
+                         rd_atomic32_get(&multi_addparts_resp_cnt),
+                         rd_kafka_err2name(err));
+
+                rd_atomic32_add(&multi_addparts_resp_cnt, 1);
+        }
+
+        return RD_KAFKA_RESP_ERR_NO_ERROR;
+}
+
+
+static void do_test_txn_addparts_req_multi(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        const char *txnid = "txnid", *topic = "mytopic";
+        int32_t txn_coord = 2;
+
+        SUB_TEST();
+
+        rd_atomic32_init(&multi_addparts_resp_cnt, 0);
+
+        on_response_received_cb = multi_addparts_response_received_cb;
+        rk = create_txn_producer(&mcluster, txnid, 3, "linger.ms", "0",
+                                 "message.timeout.ms", "9000",
+                                 /* Set up on_response_received interceptor */
+                                 "on_response_received", "", NULL);
+
+        /* Let broker 1 be txn coordinator. */
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", txnid,
+                                      txn_coord);
+
+        rd_kafka_mock_topic_create(mcluster, topic, 3, 1);
+
+        /* Set partition leaders to non-txn-coord broker so they wont
+         * be affected by rtt delay */
+        rd_kafka_mock_partition_set_leader(mcluster, topic, 0, 1);
+        rd_kafka_mock_partition_set_leader(mcluster, topic, 1, 1);
+        rd_kafka_mock_partition_set_leader(mcluster, topic, 2, 1);
+
+
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        /*
+         * Run one transaction first to let the client familiarize with
+         * the topic, this avoids metadata lookups, etc, when the real
+         * test is run.
+         */
+        TEST_SAY("Running seed transaction\n");
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+        TEST_CALL_ERR__(rd_kafka_producev(rk, RD_KAFKA_V_TOPIC(topic),
+                                          RD_KAFKA_V_VALUE("seed", 4),
+                                          RD_KAFKA_V_END));
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, 5000));
+
+
+        /*
+         * Now perform test transaction with rtt delays
+         */
+        TEST_SAY("Running test transaction\n");
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        /* Reset counter */
+        rd_atomic32_set(&multi_addparts_resp_cnt, 0);
+
+        /* Add latency to txn coordinator so we can pace our produce() calls */
+        rd_kafka_mock_broker_set_rtt(mcluster, txn_coord, 1000);
+
+        /* Produce to partition 0 */
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC(topic), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        rd_usleep(500 * 1000, NULL);
+
+        /* Produce to partition 1 */
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC(topic), RD_KAFKA_V_PARTITION(1),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        TEST_SAY("Waiting for two AddPartitionsToTxnResponse\n");
+        while (rd_atomic32_get(&multi_addparts_resp_cnt) < 2)
+                rd_usleep(10 * 1000, NULL);
+
+        TEST_SAY("%" PRId32 " AddPartitionsToTxnResponses seen\n",
+                 rd_atomic32_get(&multi_addparts_resp_cnt));
+
+        /* Produce to partition 2, this message will hang in
+         * queue if the bug is not fixed. */
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC(topic), RD_KAFKA_V_PARTITION(2),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        /* Allow some extra time for things to settle before committing
+         * transaction. */
+        rd_usleep(1000 * 1000, NULL);
+
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, 10 * 1000));
+
+        /* All done */
+        rd_kafka_destroy(rk);
+
+        on_response_received_cb = NULL;
+
+        SUB_TEST_PASS();
+}
+
+
+
+/**
+ * @brief Test handling of OffsetFetchRequest returning UNSTABLE_OFFSET_COMMIT.
+ *
+ * There are two things to test;
+ *  - OffsetFetch triggered by committed() (and similar code paths)
+ *  - OffsetFetch triggered by assign()
+ */
+static void do_test_unstable_offset_commit(void) {
+        rd_kafka_t *rk, *c;
+        rd_kafka_conf_t *c_conf;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_topic_partition_list_t *offsets;
+        const char *topic              = "srctopic4";
+        const int msgcnt               = 100;
+        const int64_t offset_to_commit = msgcnt / 2;
+        int i;
+        int remains = 0;
+
+        SUB_TEST_QUICK();
+
+        rk = create_txn_producer(&mcluster, "txnid", 3, NULL);
+
+        test_conf_init(&c_conf, NULL, 0);
+        test_conf_set(c_conf, "security.protocol", "PLAINTEXT");
+        test_conf_set(c_conf, "bootstrap.servers",
+                      rd_kafka_mock_cluster_bootstraps(mcluster));
+        test_conf_set(c_conf, "enable.partition.eof", "true");
+        test_conf_set(c_conf, "auto.offset.reset", "error");
+        c = test_create_consumer("mygroup", NULL, c_conf, NULL);
+
+        rd_kafka_mock_topic_create(mcluster, topic, 2, 3);
+
+        /* Produce some messages to the topic so that the consumer has
+         * something to read. */
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, -1));
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+        test_produce_msgs2_nowait(rk, topic, 0, 0, 0, msgcnt, NULL, 0,
+                                  &remains);
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+
+
+        /* Commit offset */
+        offsets = rd_kafka_topic_partition_list_new(1);
+        rd_kafka_topic_partition_list_add(offsets, topic, 0)->offset =
+            offset_to_commit;
+        TEST_CALL_ERR__(rd_kafka_commit(c, offsets, 0 /*sync*/));
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+        /* Retrieve offsets by calling committed().
+         *
+         * Have OffsetFetch fail and retry, on the first iteration
+         * the API timeout is higher than the amount of time the retries will
+         * take and thus succeed, and on the second iteration the timeout
+         * will be lower and thus fail. */
+        for (i = 0; i < 2; i++) {
+                rd_kafka_resp_err_t err;
+                rd_kafka_resp_err_t exp_err =
+                    i == 0 ? RD_KAFKA_RESP_ERR_NO_ERROR
+                           : RD_KAFKA_RESP_ERR__TIMED_OUT;
+                int timeout_ms = exp_err ? 200 : 5 * 1000;
+
+                rd_kafka_mock_push_request_errors(
+                    mcluster, RD_KAFKAP_OffsetFetch,
+                    1 + 5, /* first request + some retries */
+                    RD_KAFKA_RESP_ERR_UNSTABLE_OFFSET_COMMIT,
+                    RD_KAFKA_RESP_ERR_UNSTABLE_OFFSET_COMMIT,
+                    RD_KAFKA_RESP_ERR_UNSTABLE_OFFSET_COMMIT,
+                    RD_KAFKA_RESP_ERR_UNSTABLE_OFFSET_COMMIT,
+                    RD_KAFKA_RESP_ERR_UNSTABLE_OFFSET_COMMIT,
+                    RD_KAFKA_RESP_ERR_UNSTABLE_OFFSET_COMMIT);
+
+                offsets = rd_kafka_topic_partition_list_new(1);
+                rd_kafka_topic_partition_list_add(offsets, topic, 0);
+
+                err = rd_kafka_committed(c, offsets, timeout_ms);
+
+                TEST_SAY("#%d: committed() returned %s (expected %s)\n", i,
+                         rd_kafka_err2name(err), rd_kafka_err2name(exp_err));
+
+                TEST_ASSERT(err == exp_err,
+                            "#%d: Expected committed() to return %s, not %s", i,
+                            rd_kafka_err2name(exp_err), rd_kafka_err2name(err));
+                TEST_ASSERT(offsets->cnt == 1,
+                            "Expected 1 committed offset, not %d",
+                            offsets->cnt);
+                if (!exp_err)
+                        TEST_ASSERT(offsets->elems[0].offset ==
+                                        offset_to_commit,
+                                    "Expected committed offset %" PRId64
+                                    ", "
+                                    "not %" PRId64,
+                                    offset_to_commit, offsets->elems[0].offset);
+                else
+                        TEST_ASSERT(offsets->elems[0].offset < 0,
+                                    "Expected no committed offset, "
+                                    "not %" PRId64,
+                                    offsets->elems[0].offset);
+
+                rd_kafka_topic_partition_list_destroy(offsets);
+        }
+
+        TEST_SAY("Phase 2: OffsetFetch lookup through assignment\n");
+        offsets = rd_kafka_topic_partition_list_new(1);
+        rd_kafka_topic_partition_list_add(offsets, topic, 0)->offset =
+            RD_KAFKA_OFFSET_STORED;
+
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_OffsetFetch,
+            1 + 5, /* first request + some retries */
+            RD_KAFKA_RESP_ERR_UNSTABLE_OFFSET_COMMIT,
+            RD_KAFKA_RESP_ERR_UNSTABLE_OFFSET_COMMIT,
+            RD_KAFKA_RESP_ERR_UNSTABLE_OFFSET_COMMIT,
+            RD_KAFKA_RESP_ERR_UNSTABLE_OFFSET_COMMIT,
+            RD_KAFKA_RESP_ERR_UNSTABLE_OFFSET_COMMIT,
+            RD_KAFKA_RESP_ERR_UNSTABLE_OFFSET_COMMIT);
+
+        test_consumer_incremental_assign("assign", c, offsets);
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+        test_consumer_poll_exact("consume", c, 0, 1 /*eof*/, 0, msgcnt / 2,
+                                 rd_true /*exact counts*/, NULL);
+
+        /* All done */
+        rd_kafka_destroy(c);
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief If a message times out locally before being attempted to send
+ *        and commit_transaction() is called, the transaction must not succeed.
+ *        https://github.com/confluentinc/confluent-kafka-dotnet/issues/1568
+ */
+static void do_test_commit_after_msg_timeout(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        int32_t coord_id, leader_id;
+        rd_kafka_resp_err_t err;
+        rd_kafka_error_t *error;
+        const char *topic            = "test";
+        const char *transactional_id = "txnid";
+        int remains                  = 0;
+
+        SUB_TEST_QUICK();
+
+        /* Assign coordinator and leader to two different brokers */
+        coord_id  = 1;
+        leader_id = 2;
+
+        rk = create_txn_producer(&mcluster, transactional_id, 3,
+                                 "message.timeout.ms", "5000",
+                                 "transaction.timeout.ms", "10000", NULL);
+
+        /* Broker down is not a test-failing error */
+        allowed_error          = RD_KAFKA_RESP_ERR__TRANSPORT;
+        test_curr->is_fatal_cb = error_is_fatal_cb;
+        test_curr->exp_dr_err  = RD_KAFKA_RESP_ERR__MSG_TIMED_OUT;
+
+        err = rd_kafka_mock_topic_create(mcluster, topic, 1, 3);
+        TEST_ASSERT(!err, "Failed to create topic: %s", rd_kafka_err2str(err));
+
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", transactional_id,
+                                      coord_id);
+        rd_kafka_mock_partition_set_leader(mcluster, topic, 0, leader_id);
+
+        /* Start transactioning */
+        TEST_SAY("Starting transaction\n");
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, -1));
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        TEST_SAY("Bringing down %" PRId32 "\n", leader_id);
+        rd_kafka_mock_broker_set_down(mcluster, leader_id);
+        rd_kafka_mock_broker_set_down(mcluster, coord_id);
+
+        test_produce_msgs2_nowait(rk, topic, 0, 0, 0, 1, NULL, 0, &remains);
+
+        error = rd_kafka_commit_transaction(rk, -1);
+        TEST_ASSERT(error != NULL, "expected commit_transaciton() to fail");
+        TEST_SAY_ERROR(error, "commit_transaction() failed (as expected): ");
+        TEST_ASSERT(rd_kafka_error_txn_requires_abort(error),
+                    "Expected txn_requires_abort error");
+        rd_kafka_error_destroy(error);
+
+        /* Bring the brokers up so the abort can complete */
+        rd_kafka_mock_broker_set_up(mcluster, coord_id);
+        rd_kafka_mock_broker_set_up(mcluster, leader_id);
+
+        TEST_SAY("Aborting transaction\n");
+        TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, -1));
+
+        TEST_ASSERT(remains == 0, "%d message(s) were not flushed\n", remains);
+
+        TEST_SAY("Attempting second transaction, which should succeed\n");
+        test_curr->is_fatal_cb = error_is_fatal_cb;
+        test_curr->exp_dr_err  = RD_KAFKA_RESP_ERR_NO_ERROR;
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+        test_produce_msgs2_nowait(rk, topic, 0, 0, 0, 1, NULL, 0, &remains);
+
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+
+        TEST_ASSERT(remains == 0, "%d message(s) were not produced\n", remains);
+
+        rd_kafka_destroy(rk);
+
+        allowed_error          = RD_KAFKA_RESP_ERR_NO_ERROR;
+        test_curr->is_fatal_cb = NULL;
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief #3575: Verify that OUT_OF_ORDER_SEQ does not trigger an epoch bump
+ *        during an ongoing transaction.
+ *        The transaction should instead enter the abortable state.
+ */
+static void do_test_out_of_order_seq(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_error_t *error;
+        int32_t txn_coord = 1, leader = 2;
+        const char *txnid = "myTxnId";
+        test_timing_t timing;
+        rd_kafka_resp_err_t err;
+
+        SUB_TEST_QUICK();
+
+        rk = create_txn_producer(&mcluster, txnid, 3, "batch.num.messages", "1",
+                                 NULL);
+
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", txnid,
+                                      txn_coord);
+
+        rd_kafka_mock_partition_set_leader(mcluster, "mytopic", 0, leader);
+
+        test_curr->ignore_dr_err = rd_true;
+        test_curr->is_fatal_cb   = NULL;
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, -1));
+
+        /*
+         * Start a transaction
+         */
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+
+
+        /* Produce one seeding message first to get the leader up and running */
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+        test_flush(rk, -1);
+
+        /* Let partition leader have a latency of 2 seconds
+         * so that we can have multiple messages in-flight. */
+        rd_kafka_mock_broker_set_rtt(mcluster, leader, 2 * 1000);
+
+        /* Produce a message, let it fail with with different errors,
+         * ending with OUT_OF_ORDER which previously triggered an
+         * Epoch bump. */
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_Produce, 3,
+            RD_KAFKA_RESP_ERR_NOT_LEADER_FOR_PARTITION,
+            RD_KAFKA_RESP_ERR_NOT_LEADER_FOR_PARTITION,
+            RD_KAFKA_RESP_ERR_OUT_OF_ORDER_SEQUENCE_NUMBER);
+
+        /* Produce three messages that will be delayed
+         * and have errors injected.*/
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        /* Now sleep a short while so that the messages are processed
+         * by the broker and errors are returned. */
+        TEST_SAY("Sleeping..\n");
+        rd_sleep(5);
+
+        rd_kafka_mock_broker_set_rtt(mcluster, leader, 0);
+
+        /* Produce a fifth message, should fail with ERR__STATE since
+         * the transaction should have entered the abortable state. */
+        err = rd_kafka_producev(rk, RD_KAFKA_V_TOPIC("mytopic"),
+                                RD_KAFKA_V_PARTITION(0),
+                                RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END);
+        TEST_ASSERT(err == RD_KAFKA_RESP_ERR__STATE,
+                    "Expected produce() to fail with ERR__STATE, not %s",
+                    rd_kafka_err2name(err));
+        TEST_SAY("produce() failed as expected: %s\n", rd_kafka_err2str(err));
+
+        /* Commit the transaction, should fail with abortable error. */
+        TIMING_START(&timing, "commit_transaction(-1)");
+        error = rd_kafka_commit_transaction(rk, -1);
+        TIMING_STOP(&timing);
+        TEST_ASSERT(error != NULL, "Expected commit_transaction() to fail");
+
+        TEST_SAY("commit_transaction() failed (expectedly): %s\n",
+                 rd_kafka_error_string(error));
+
+        TEST_ASSERT(!rd_kafka_error_is_fatal(error),
+                    "Did not expect fatal error");
+        TEST_ASSERT(rd_kafka_error_txn_requires_abort(error),
+                    "Expected abortable error");
+        rd_kafka_error_destroy(error);
+
+        /* Abort the transaction */
+        TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, -1));
+
+        /* Run a new transaction without errors to verify that the
+         * producer can recover. */
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Verify lossless delivery if topic disappears from Metadata for awhile.
+ *
+ * If a topic is removed from metadata inbetween transactions, the producer
+ * will remove its partition state for the topic's partitions.
+ * If later the same topic comes back (same topic instance, not a new creation)
+ * then the producer must restore the previously used msgid/BaseSequence
+ * in case the same Epoch is still used, or messages will be silently lost
+ * as they would seem like legit duplicates to the broker.
+ *
+ * Reproduction:
+ *   1. produce msgs to topic, commit transaction.
+ *   2. remove topic from metadata
+ *   3. make sure client updates its metadata, which removes the partition
+ *      objects.
+ *   4. restore the topic in metadata
+ *   5. produce new msgs to topic, commit transaction.
+ *   6. consume topic. All messages should be accounted for.
+ */
+static void do_test_topic_disappears_for_awhile(void) {
+        rd_kafka_t *rk, *c;
+        rd_kafka_conf_t *c_conf;
+        rd_kafka_mock_cluster_t *mcluster;
+        const char *topic = "mytopic";
+        const char *txnid = "myTxnId";
+        test_timing_t timing;
+        int i;
+        int msgcnt              = 0;
+        const int partition_cnt = 10;
+
+        SUB_TEST_QUICK();
+
+        rk = create_txn_producer(
+            &mcluster, txnid, 1, "batch.num.messages", "3", "linger.ms", "100",
+            "topic.metadata.refresh.interval.ms", "2000", NULL);
+
+        rd_kafka_mock_topic_create(mcluster, topic, partition_cnt, 1);
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, -1));
+
+        for (i = 0; i < 2; i++) {
+                int cnt                = 3 * 2 * partition_cnt;
+                rd_bool_t remove_topic = (i % 2) == 0;
+                /*
+                 * Start a transaction
+                 */
+                TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+
+                while (cnt-- >= 0) {
+                        TEST_CALL_ERR__(rd_kafka_producev(
+                            rk, RD_KAFKA_V_TOPIC(topic),
+                            RD_KAFKA_V_PARTITION(cnt % partition_cnt),
+                            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+                        msgcnt++;
+                }
+
+                /* Commit the transaction */
+                TIMING_START(&timing, "commit_transaction(-1)");
+                TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+                TIMING_STOP(&timing);
+
+
+
+                if (remove_topic) {
+                        /* Make it seem the topic is removed, refresh metadata,
+                         * and then make the topic available again. */
+                        const rd_kafka_metadata_t *md;
+
+                        TEST_SAY("Marking topic as non-existent\n");
+
+                        rd_kafka_mock_topic_set_error(
+                            mcluster, topic,
+                            RD_KAFKA_RESP_ERR_UNKNOWN_TOPIC_OR_PART);
+
+                        TEST_CALL_ERR__(rd_kafka_metadata(rk, 0, NULL, &md,
+                                                          tmout_multip(5000)));
+
+                        rd_kafka_metadata_destroy(md);
+
+                        rd_sleep(2);
+
+                        TEST_SAY("Bringing topic back to life\n");
+                        rd_kafka_mock_topic_set_error(
+                            mcluster, topic, RD_KAFKA_RESP_ERR_NO_ERROR);
+                }
+        }
+
+        TEST_SAY("Verifying messages by consumtion\n");
+        test_conf_init(&c_conf, NULL, 0);
+        test_conf_set(c_conf, "security.protocol", "PLAINTEXT");
+        test_conf_set(c_conf, "bootstrap.servers",
+                      rd_kafka_mock_cluster_bootstraps(mcluster));
+        test_conf_set(c_conf, "enable.partition.eof", "true");
+        test_conf_set(c_conf, "auto.offset.reset", "earliest");
+        c = test_create_consumer("mygroup", NULL, c_conf, NULL);
+
+        test_consumer_subscribe(c, topic);
+        test_consumer_poll_exact("consume", c, 0, partition_cnt, 0, msgcnt,
+                                 rd_true /*exact*/, NULL);
+        rd_kafka_destroy(c);
+
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Test that group coordinator requests can handle an
+ *        untimely disconnect.
+ *
+ * The transaction manager makes use of librdkafka coord_req to commit
+ * transaction offsets to the group coordinator.
+ * If the connection to the given group coordinator is not up the
+ * coord_req code will request a connection once, but if this connection fails
+ * there will be no new attempts and the coord_req will idle until either
+ * destroyed or the connection is retried for other reasons.
+ * This in turn stalls the send_offsets_to_transaction() call until the
+ * transaction times out.
+ *
+ * There are two variants to this test based on switch_coord:
+ *  - True - Switches the coordinator during the downtime.
+ *           The client should detect this and send the request to the
+ *           new coordinator.
+ *  - False - The coordinator remains on the down broker. Client will reconnect
+ *            when down broker comes up again.
+ */
+struct some_state {
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_bool_t switch_coord;
+        int32_t broker_id;
+        const char *grpid;
+};
+
+static int delayed_up_cb(void *arg) {
+        struct some_state *state = arg;
+        rd_sleep(3);
+        if (state->switch_coord) {
+                TEST_SAY("Switching group coordinator to %" PRId32 "\n",
+                         state->broker_id);
+                rd_kafka_mock_coordinator_set(state->mcluster, "group",
+                                              state->grpid, state->broker_id);
+        } else {
+                TEST_SAY("Bringing up group coordinator %" PRId32 "..\n",
+                         state->broker_id);
+                rd_kafka_mock_broker_set_up(state->mcluster, state->broker_id);
+        }
+        return 0;
+}
+
+static void do_test_disconnected_group_coord(rd_bool_t switch_coord) {
+        const char *topic       = "mytopic";
+        const char *txnid       = "myTxnId";
+        const char *grpid       = "myGrpId";
+        const int partition_cnt = 1;
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+        struct some_state state = RD_ZERO_INIT;
+        test_timing_t timing;
+        thrd_t thrd;
+        int ret;
+
+        SUB_TEST_QUICK("switch_coord=%s", RD_STR_ToF(switch_coord));
+
+        test_curr->is_fatal_cb = error_is_fatal_cb;
+        allowed_error          = RD_KAFKA_RESP_ERR__TRANSPORT;
+
+        rk = create_txn_producer(&mcluster, txnid, 3, NULL);
+
+        rd_kafka_mock_topic_create(mcluster, topic, partition_cnt, 1);
+
+        /* Broker 1: txn coordinator
+         * Broker 2: group coordinator
+         * Broker 3: partition leader & backup coord if switch_coord=true */
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", txnid, 1);
+        rd_kafka_mock_coordinator_set(mcluster, "group", grpid, 2);
+        rd_kafka_mock_partition_set_leader(mcluster, topic, 0, 3);
+
+        /* Bring down group coordinator so there are no undesired
+         * connections to it. */
+        rd_kafka_mock_broker_set_down(mcluster, 2);
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, -1));
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC(topic), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+        test_flush(rk, -1);
+
+        rd_sleep(1);
+
+        /* Run a background thread that after 3s, which should be enough
+         * to perform the first failed connection attempt, makes the
+         * group coordinator available again. */
+        state.switch_coord = switch_coord;
+        state.mcluster     = mcluster;
+        state.grpid        = grpid;
+        state.broker_id    = switch_coord ? 3 : 2;
+        if (thrd_create(&thrd, delayed_up_cb, &state) != thrd_success)
+                TEST_FAIL("Failed to create thread");
+
+        TEST_SAY("Calling send_offsets_to_transaction()\n");
+        offsets = rd_kafka_topic_partition_list_new(1);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 0)->offset = 1;
+        cgmetadata = rd_kafka_consumer_group_metadata_new(grpid);
+
+        TIMING_START(&timing, "send_offsets_to_transaction(-1)");
+        TEST_CALL_ERROR__(
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1));
+        TIMING_STOP(&timing);
+        TIMING_ASSERT(&timing, 0, 10 * 1000 /*10s*/);
+
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+        thrd_join(thrd, &ret);
+
+        /* Commit the transaction */
+        TIMING_START(&timing, "commit_transaction(-1)");
+        TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+        TIMING_STOP(&timing);
+
+        rd_kafka_destroy(rk);
+
+        allowed_error          = RD_KAFKA_RESP_ERR_NO_ERROR;
+        test_curr->is_fatal_cb = NULL;
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Test that a NULL coordinator is not fatal when
+ * the transactional producer reconnects to the txn coordinator
+ * and the first thing it does is a FindCoordinatorRequest that
+ * fails with COORDINATOR_NOT_AVAILABLE, setting coordinator to NULL.
+ */
+static void do_test_txn_coordinator_null_not_fatal(void) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_error_t *error;
+        rd_kafka_resp_err_t err;
+        int32_t coord_id             = 1;
+        const char *topic            = "test";
+        const char *transactional_id = "txnid";
+        int msgcnt                   = 1;
+        int remains                  = 0;
+
+        SUB_TEST_QUICK();
+
+        /* Broker down is not a test-failing error */
+        allowed_error          = RD_KAFKA_RESP_ERR__TRANSPORT;
+        allowed_error_2        = RD_KAFKA_RESP_ERR__TIMED_OUT;
+        test_curr->is_fatal_cb = error_is_fatal_cb;
+        test_curr->exp_dr_err  = RD_KAFKA_RESP_ERR__MSG_TIMED_OUT;
+
+        /* One second is the minimum transaction timeout */
+        rk = create_txn_producer(&mcluster, transactional_id, 1,
+                                 "transaction.timeout.ms", "1000", NULL);
+
+        err = rd_kafka_mock_topic_create(mcluster, topic, 1, 1);
+        TEST_ASSERT(!err, "Failed to create topic: %s", rd_kafka_err2str(err));
+
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", transactional_id,
+                                      coord_id);
+        rd_kafka_mock_partition_set_leader(mcluster, topic, 0, coord_id);
+
+        /* Start transactioning */
+        TEST_SAY("Starting transaction\n");
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        /* Makes the produce request timeout. */
+        rd_kafka_mock_broker_push_request_error_rtts(
+            mcluster, coord_id, RD_KAFKAP_Produce, 1,
+            RD_KAFKA_RESP_ERR_NO_ERROR, 3000);
+
+        test_produce_msgs2_nowait(rk, topic, 0, RD_KAFKA_PARTITION_UA, 0,
+                                  msgcnt, NULL, 0, &remains);
+
+        /* This value is linked to transaction.timeout.ms, needs enough time
+         * so the message times out and a DrainBump sequence is started. */
+        rd_kafka_flush(rk, 1000);
+
+        /* To trigger the error the COORDINATOR_NOT_AVAILABLE response
+         * must come AFTER idempotent state has changed to WaitTransport
+         * but BEFORE it changes to WaitPID. To make it more likely
+         * rd_kafka_txn_coord_timer_start timeout can be changed to 5 ms
+         * in rd_kafka_txn_coord_query, when unable to query for
+         * transaction coordinator.
+         */
+        rd_kafka_mock_broker_push_request_error_rtts(
+            mcluster, coord_id, RD_KAFKAP_FindCoordinator, 1,
+            RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE, 10);
+
+        /* Coordinator down starts the FindCoordinatorRequest loop. */
+        TEST_SAY("Bringing down coordinator %" PRId32 "\n", coord_id);
+        rd_kafka_mock_broker_set_down(mcluster, coord_id);
+
+        /* Coordinator down for some time. */
+        rd_usleep(100 * 1000, NULL);
+
+        /* When it comes up, the error is triggered, if the preconditions
+         * happen. */
+        TEST_SAY("Bringing up coordinator %" PRId32 "\n", coord_id);
+        rd_kafka_mock_broker_set_up(mcluster, coord_id);
+
+        /* Make sure DRs are received */
+        rd_kafka_flush(rk, 1000);
+
+        error = rd_kafka_commit_transaction(rk, -1);
+
+        TEST_ASSERT(remains == 0, "%d message(s) were not produced\n", remains);
+        TEST_ASSERT(error != NULL, "Expected commit_transaction() to fail");
+        TEST_SAY("commit_transaction() failed (expectedly): %s\n",
+                 rd_kafka_error_string(error));
+        rd_kafka_error_destroy(error);
+
+        /* Needs to wait some time before closing to make sure it doesn't go
+         * into TERMINATING state before error is triggered. */
+        rd_usleep(1000 * 1000, NULL);
+        rd_kafka_destroy(rk);
+
+        allowed_error          = RD_KAFKA_RESP_ERR_NO_ERROR;
+        allowed_error_2        = RD_KAFKA_RESP_ERR_NO_ERROR;
+        test_curr->exp_dr_err  = RD_KAFKA_RESP_ERR_NO_ERROR;
+        test_curr->is_fatal_cb = NULL;
+
+        SUB_TEST_PASS();
+}
+
+
+
+/**
+ * @brief Simple test to make sure the init_transactions() timeout is honoured
+ *        and also not infinite.
+ */
+static void do_test_txn_resumable_init(void) {
+        rd_kafka_t *rk;
+        const char *transactional_id = "txnid";
+        rd_kafka_error_t *error;
+        test_timing_t duration;
+
+        SUB_TEST();
+
+        rd_kafka_conf_t *conf;
+
+        test_conf_init(&conf, NULL, 20);
+        test_conf_set(conf, "bootstrap.servers", "");
+        test_conf_set(conf, "transactional.id", transactional_id);
+        test_conf_set(conf, "transaction.timeout.ms", "4000");
+
+        rk = test_create_handle(RD_KAFKA_PRODUCER, conf);
+
+        /* First make sure a lower timeout is honoured. */
+        TIMING_START(&duration, "init_transactions(1000)");
+        error = rd_kafka_init_transactions(rk, 1000);
+        TIMING_STOP(&duration);
+
+        if (error)
+                TEST_SAY("First init_transactions failed (as expected): %s\n",
+                         rd_kafka_error_string(error));
+        TEST_ASSERT(rd_kafka_error_code(error) == RD_KAFKA_RESP_ERR__TIMED_OUT,
+                    "Expected _TIMED_OUT, not %s",
+                    error ? rd_kafka_error_string(error) : "success");
+        rd_kafka_error_destroy(error);
+
+        TIMING_ASSERT(&duration, 900, 1500);
+
+        TEST_SAY(
+            "Performing second init_transactions() call now with an "
+            "infinite timeout: "
+            "should time out in 2 x transaction.timeout.ms\n");
+
+        TIMING_START(&duration, "init_transactions(infinite)");
+        error = rd_kafka_init_transactions(rk, -1);
+        TIMING_STOP(&duration);
+
+        if (error)
+                TEST_SAY("Second init_transactions failed (as expected): %s\n",
+                         rd_kafka_error_string(error));
+        TEST_ASSERT(rd_kafka_error_code(error) == RD_KAFKA_RESP_ERR__TIMED_OUT,
+                    "Expected _TIMED_OUT, not %s",
+                    error ? rd_kafka_error_string(error) : "success");
+        rd_kafka_error_destroy(error);
+
+        TIMING_ASSERT(&duration, 2 * 4000 - 500, 2 * 4000 + 500);
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Retries a transaction call until it succeeds or returns a
+ *        non-retriable error - which will cause the test to fail.
+ *
+ * @param intermed_calls Is a block of code that will be called after each
+ *                       retriable failure of \p call.
+ */
+#define RETRY_TXN_CALL__(call, intermed_calls)                                 \
+        do {                                                                   \
+                rd_kafka_error_t *_error = call;                               \
+                if (!_error)                                                   \
+                        break;                                                 \
+                TEST_SAY_ERROR(_error, "%s: ", "" #call);                      \
+                TEST_ASSERT(rd_kafka_error_is_retriable(_error),               \
+                            "Expected retriable error");                       \
+                TEST_SAY("%s failed, retrying in 1 second\n", "" #call);       \
+                rd_kafka_error_destroy(_error);                                \
+                intermed_calls;                                                \
+                rd_sleep(1);                                                   \
+        } while (1)
+
+/**
+ * @brief Call \p call and expect it to fail with \p exp_err_code.
+ */
+#define TXN_CALL_EXPECT_ERROR__(call, exp_err_code)                            \
+        do {                                                                   \
+                rd_kafka_error_t *_error = call;                               \
+                TEST_ASSERT(_error != NULL,                                    \
+                            "%s: Expected %s error, got success", "" #call,    \
+                            rd_kafka_err2name(exp_err_code));                  \
+                TEST_SAY_ERROR(_error, "%s: ", "" #call);                      \
+                TEST_ASSERT(rd_kafka_error_code(_error) == exp_err_code,       \
+                            "%s: Expected %s error, got %s", "" #call,         \
+                            rd_kafka_err2name(exp_err_code),                   \
+                            rd_kafka_error_name(_error));                      \
+                rd_kafka_error_destroy(_error);                                \
+        } while (0)
+
+
+/**
+ * @brief Simple test to make sure short API timeouts can be safely resumed
+ *        by calling the same API again.
+ *
+ * @param do_commit Commit transaction if true, else abort transaction.
+ */
+static void do_test_txn_resumable_calls_timeout(rd_bool_t do_commit) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_resp_err_t err;
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+        int32_t coord_id             = 1;
+        const char *topic            = "test";
+        const char *transactional_id = "txnid";
+        int msgcnt                   = 1;
+        int remains                  = 0;
+
+        SUB_TEST("%s_transaction", do_commit ? "commit" : "abort");
+
+        rk = create_txn_producer(&mcluster, transactional_id, 1, NULL);
+
+        err = rd_kafka_mock_topic_create(mcluster, topic, 1, 1);
+        TEST_ASSERT(!err, "Failed to create topic: %s", rd_kafka_err2str(err));
+
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", transactional_id,
+                                      coord_id);
+        rd_kafka_mock_partition_set_leader(mcluster, topic, 0, coord_id);
+
+        TEST_SAY("Starting transaction\n");
+        TEST_SAY("Delaying first two InitProducerIdRequests by 500ms\n");
+        rd_kafka_mock_broker_push_request_error_rtts(
+            mcluster, coord_id, RD_KAFKAP_InitProducerId, 2,
+            RD_KAFKA_RESP_ERR_NO_ERROR, 500, RD_KAFKA_RESP_ERR_NO_ERROR, 500);
+
+        RETRY_TXN_CALL__(
+            rd_kafka_init_transactions(rk, 100),
+            TXN_CALL_EXPECT_ERROR__(rd_kafka_abort_transaction(rk, -1),
+                                    RD_KAFKA_RESP_ERR__CONFLICT));
+
+        RETRY_TXN_CALL__(rd_kafka_begin_transaction(rk), /*none*/);
+
+
+        TEST_SAY("Delaying ProduceRequests by 3000ms\n");
+        rd_kafka_mock_broker_push_request_error_rtts(
+            mcluster, coord_id, RD_KAFKAP_Produce, 1,
+            RD_KAFKA_RESP_ERR_NO_ERROR, 3000);
+
+        test_produce_msgs2_nowait(rk, topic, 0, RD_KAFKA_PARTITION_UA, 0,
+                                  msgcnt, NULL, 0, &remains);
+
+
+        TEST_SAY("Delaying SendOffsetsToTransaction by 400ms\n");
+        rd_kafka_mock_broker_push_request_error_rtts(
+            mcluster, coord_id, RD_KAFKAP_AddOffsetsToTxn, 1,
+            RD_KAFKA_RESP_ERR_NO_ERROR, 400);
+        offsets = rd_kafka_topic_partition_list_new(1);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 0)->offset = 12;
+        cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+        /* This is not a resumable call on timeout */
+        TEST_CALL_ERROR__(
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1));
+
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+
+        TEST_SAY("Delaying EndTxnRequests by 1200ms\n");
+        rd_kafka_mock_broker_push_request_error_rtts(
+            mcluster, coord_id, RD_KAFKAP_EndTxn, 1, RD_KAFKA_RESP_ERR_NO_ERROR,
+            1200);
+
+        /* Committing/aborting the transaction will also be delayed by the
+         * previous accumulated remaining delays. */
+
+        if (do_commit) {
+                TEST_SAY("Committing transaction\n");
+
+                RETRY_TXN_CALL__(
+                    rd_kafka_commit_transaction(rk, 100),
+                    TXN_CALL_EXPECT_ERROR__(rd_kafka_abort_transaction(rk, -1),
+                                            RD_KAFKA_RESP_ERR__CONFLICT));
+        } else {
+                TEST_SAY("Aborting transaction\n");
+
+                RETRY_TXN_CALL__(
+                    rd_kafka_abort_transaction(rk, 100),
+                    TXN_CALL_EXPECT_ERROR__(rd_kafka_commit_transaction(rk, -1),
+                                            RD_KAFKA_RESP_ERR__CONFLICT));
+        }
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Verify that resuming timed out calls that after the timeout, but
+ *        before the resuming call, would error out.
+ */
+static void do_test_txn_resumable_calls_timeout_error(rd_bool_t do_commit) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_resp_err_t err;
+        int32_t coord_id             = 1;
+        const char *topic            = "test";
+        const char *transactional_id = "txnid";
+        int msgcnt                   = 1;
+        int remains                  = 0;
+        rd_kafka_error_t *error;
+
+        SUB_TEST_QUICK("%s_transaction", do_commit ? "commit" : "abort");
+
+        rk = create_txn_producer(&mcluster, transactional_id, 1, NULL);
+
+        err = rd_kafka_mock_topic_create(mcluster, topic, 1, 1);
+        TEST_ASSERT(!err, "Failed to create topic: %s", rd_kafka_err2str(err));
+
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", transactional_id,
+                                      coord_id);
+        rd_kafka_mock_partition_set_leader(mcluster, topic, 0, coord_id);
+
+        TEST_SAY("Starting transaction\n");
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, -1));
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        test_produce_msgs2_nowait(rk, topic, 0, RD_KAFKA_PARTITION_UA, 0,
+                                  msgcnt, NULL, 0, &remains);
+
+
+        TEST_SAY("Fail EndTxn fatally after 2000ms\n");
+        rd_kafka_mock_broker_push_request_error_rtts(
+            mcluster, coord_id, RD_KAFKAP_EndTxn, 1,
+            RD_KAFKA_RESP_ERR_INVALID_TXN_STATE, 2000);
+
+        if (do_commit) {
+                TEST_SAY("Committing transaction\n");
+
+                TXN_CALL_EXPECT_ERROR__(rd_kafka_commit_transaction(rk, 500),
+                                        RD_KAFKA_RESP_ERR__TIMED_OUT);
+
+                /* Sleep so that the background EndTxn fails locally and sets
+                 * an error result. */
+                rd_sleep(3);
+
+                error = rd_kafka_commit_transaction(rk, -1);
+
+        } else {
+                TEST_SAY("Aborting transaction\n");
+
+                TXN_CALL_EXPECT_ERROR__(rd_kafka_commit_transaction(rk, 500),
+                                        RD_KAFKA_RESP_ERR__TIMED_OUT);
+
+                /* Sleep so that the background EndTxn fails locally and sets
+                 * an error result. */
+                rd_sleep(3);
+
+                error = rd_kafka_commit_transaction(rk, -1);
+        }
+
+        TEST_ASSERT(error != NULL && rd_kafka_error_is_fatal(error),
+                    "Expected fatal error, not %s",
+                    rd_kafka_error_string(error));
+        TEST_ASSERT(rd_kafka_error_code(error) ==
+                        RD_KAFKA_RESP_ERR_INVALID_TXN_STATE,
+                    "Expected error INVALID_TXN_STATE, got %s",
+                    rd_kafka_error_name(error));
+        rd_kafka_error_destroy(error);
+
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Concurrent transaction API calls are not permitted.
+ *        This test makes sure they're properly enforced.
+ *
+ * For each transactional API, call it with a 5s timeout, and during that time
+ * from another thread call transactional APIs, one by one, and verify that
+ * we get an ERR__CONFLICT error back in the second thread.
+ *
+ * We use a mutex for synchronization, the main thread will hold the lock
+ * when not calling an API but release it just prior to calling.
+ * The other thread will acquire the lock, sleep, and hold the lock while
+ * calling the concurrent API that should fail immediately, releasing the lock
+ * when done.
+ *
+ */
+
+struct _txn_concurrent_state {
+        const char *api;
+        mtx_t lock;
+        rd_kafka_t *rk;
+        struct test *test;
+};
+
+static int txn_concurrent_thread_main(void *arg) {
+        struct _txn_concurrent_state *state = arg;
+        static const char *apis[]           = {
+            "init_transactions",           "begin_transaction",
+            "send_offsets_to_transaction", "commit_transaction",
+            "abort_transaction",           NULL};
+        rd_kafka_t *rk       = state->rk;
+        const char *main_api = NULL;
+        int i;
+
+        /* Update TLS variable so TEST_..() macros work */
+        test_curr = state->test;
+
+        while (1) {
+                const char *api         = NULL;
+                const int timeout_ms    = 10000;
+                rd_kafka_error_t *error = NULL;
+                rd_kafka_resp_err_t exp_err;
+                test_timing_t duration;
+
+                /* Wait for other thread's txn call to start, then sleep a bit
+                 * to increase the chance of that call has really begun. */
+                mtx_lock(&state->lock);
+
+                if (state->api && state->api == main_api) {
+                        /* Main thread is still blocking on the last API call */
+                        TEST_SAY("Waiting for main thread to finish %s()\n",
+                                 main_api);
+                        mtx_unlock(&state->lock);
+                        rd_sleep(1);
+                        continue;
+                } else if (!(main_api = state->api)) {
+                        mtx_unlock(&state->lock);
+                        break;
+                }
+
+                rd_sleep(1);
+
+                for (i = 0; (api = apis[i]) != NULL; i++) {
+                        TEST_SAY(
+                            "Triggering concurrent %s() call while "
+                            "main is in %s() call\n",
+                            api, main_api);
+                        TIMING_START(&duration, "%s", api);
+
+                        if (!strcmp(api, "init_transactions"))
+                                error =
+                                    rd_kafka_init_transactions(rk, timeout_ms);
+                        else if (!strcmp(api, "begin_transaction"))
+                                error = rd_kafka_begin_transaction(rk);
+                        else if (!strcmp(api, "send_offsets_to_transaction")) {
+                                rd_kafka_topic_partition_list_t *offsets =
+                                    rd_kafka_topic_partition_list_new(1);
+                                rd_kafka_consumer_group_metadata_t *cgmetadata =
+                                    rd_kafka_consumer_group_metadata_new(
+                                        "mygroupid");
+                                rd_kafka_topic_partition_list_add(
+                                    offsets, "srctopic4", 0)
+                                    ->offset = 12;
+
+                                error = rd_kafka_send_offsets_to_transaction(
+                                    rk, offsets, cgmetadata, -1);
+                                rd_kafka_consumer_group_metadata_destroy(
+                                    cgmetadata);
+                                rd_kafka_topic_partition_list_destroy(offsets);
+                        } else if (!strcmp(api, "commit_transaction"))
+                                error =
+                                    rd_kafka_commit_transaction(rk, timeout_ms);
+                        else if (!strcmp(api, "abort_transaction"))
+                                error =
+                                    rd_kafka_abort_transaction(rk, timeout_ms);
+                        else
+                                TEST_FAIL("Unknown API: %s", api);
+
+                        TIMING_STOP(&duration);
+
+                        TEST_SAY_ERROR(error, "Conflicting %s() call: ", api);
+                        TEST_ASSERT(error,
+                                    "Expected conflicting %s() call to fail",
+                                    api);
+
+                        exp_err = !strcmp(api, main_api)
+                                      ? RD_KAFKA_RESP_ERR__PREV_IN_PROGRESS
+                                      : RD_KAFKA_RESP_ERR__CONFLICT;
+
+                        TEST_ASSERT(rd_kafka_error_code(error) == exp_err,
+
+                                    "Conflicting %s(): Expected %s, not %s",
+                                    api, rd_kafka_err2str(exp_err),
+                                    rd_kafka_error_name(error));
+                        TEST_ASSERT(
+                            rd_kafka_error_is_retriable(error),
+                            "Conflicting %s(): Expected retriable error", api);
+                        rd_kafka_error_destroy(error);
+                        /* These calls should fail immediately */
+                        TIMING_ASSERT(&duration, 0, 100);
+                }
+
+                mtx_unlock(&state->lock);
+        }
+
+        return 0;
+}
+
+static void do_test_txn_concurrent_operations(rd_bool_t do_commit) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        int32_t coord_id = 1;
+        rd_kafka_resp_err_t err;
+        const char *topic            = "test";
+        const char *transactional_id = "txnid";
+        int remains                  = 0;
+        thrd_t thrd;
+        struct _txn_concurrent_state state = RD_ZERO_INIT;
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+
+        SUB_TEST("%s", do_commit ? "commit" : "abort");
+
+        test_timeout_set(90);
+
+        /* We need to override the value of socket.connection.setup.timeout.ms
+         * to be at least 2*RTT of the mock broker. This is because the first
+         * ApiVersion request will fail, since we make the request with v3, and
+         * the mock broker's MaxVersion is 2, so the request is retried with v0.
+         * We use the value 3*RTT to add some buffer.
+         */
+        rk = create_txn_producer(&mcluster, transactional_id, 1,
+                                 "socket.connection.setup.timeout.ms", "15000",
+                                 NULL);
+
+        /* Set broker RTT to 3.5s so that the background thread has ample
+         * time to call its conflicting APIs.
+         * This value must be less than socket.connection.setup.timeout.ms/2. */
+        rd_kafka_mock_broker_set_rtt(mcluster, coord_id, 3500);
+
+        err = rd_kafka_mock_topic_create(mcluster, topic, 1, 1);
+        TEST_ASSERT(!err, "Failed to create topic: %s", rd_kafka_err2str(err));
+
+        /* Set up shared state between us and the concurrent thread */
+        mtx_init(&state.lock, mtx_plain);
+        state.test = test_curr;
+        state.rk   = rk;
+
+        /* We release the lock only while calling the TXN API */
+        mtx_lock(&state.lock);
+
+        /* Spin up concurrent thread */
+        if (thrd_create(&thrd, txn_concurrent_thread_main, (void *)&state) !=
+            thrd_success)
+                TEST_FAIL("Failed to create thread");
+
+#define _start_call(callname)                                                  \
+        do {                                                                   \
+                state.api = callname;                                          \
+                mtx_unlock(&state.lock);                                       \
+        } while (0)
+#define _end_call() mtx_lock(&state.lock)
+
+        _start_call("init_transactions");
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, -1));
+        _end_call();
+
+        /* This call doesn't block, so can't really be tested concurrently. */
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        test_produce_msgs2_nowait(rk, topic, 0, RD_KAFKA_PARTITION_UA, 0, 10,
+                                  NULL, 0, &remains);
+
+        _start_call("send_offsets_to_transaction");
+        offsets = rd_kafka_topic_partition_list_new(1);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 0)->offset = 12;
+        cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+        TEST_CALL_ERROR__(
+            rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata, -1));
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+        _end_call();
+
+        if (do_commit) {
+                _start_call("commit_transaction");
+                TEST_CALL_ERROR__(rd_kafka_commit_transaction(rk, -1));
+                _end_call();
+        } else {
+                _start_call("abort_transaction");
+                TEST_CALL_ERROR__(rd_kafka_abort_transaction(rk, -1));
+                _end_call();
+        }
+
+        /* Signal completion to background thread */
+        state.api = NULL;
+
+        mtx_unlock(&state.lock);
+
+        thrd_join(thrd, NULL);
+
+        rd_kafka_destroy(rk);
+
+        mtx_destroy(&state.lock);
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief KIP-360: Test that fatal idempotence errors triggers abortable
+ *        transaction errors, but let the broker-side abort of the
+ *        transaction fail with a fencing error.
+ *        Should raise a fatal error.
+ *
+ * @param error_code Which error code EndTxn should fail with.
+ *                   Either RD_KAFKA_RESP_ERR_INVALID_PRODUCER_EPOCH (older)
+ *                   or RD_KAFKA_RESP_ERR_PRODUCER_FENCED (newer).
+ */
+static void do_test_txn_fenced_abort(rd_kafka_resp_err_t error_code) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_error_t *error;
+        int32_t txn_coord = 2;
+        const char *txnid = "myTxnId";
+        char errstr[512];
+        rd_kafka_resp_err_t fatal_err;
+        size_t errors_cnt;
+
+        SUB_TEST_QUICK("With error %s", rd_kafka_err2name(error_code));
+
+        rk = create_txn_producer(&mcluster, txnid, 3, "batch.num.messages", "1",
+                                 NULL);
+
+        rd_kafka_mock_coordinator_set(mcluster, "transaction", txnid,
+                                      txn_coord);
+
+        test_curr->ignore_dr_err = rd_true;
+        test_curr->is_fatal_cb   = error_is_fatal_cb;
+        allowed_error            = RD_KAFKA_RESP_ERR__FENCED;
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, -1));
+
+        /*
+         * Start a transaction
+         */
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+
+        /* Produce a message without error first */
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        test_flush(rk, -1);
+
+        /* Fail abort transaction  */
+        rd_kafka_mock_broker_push_request_error_rtts(
+            mcluster, txn_coord, RD_KAFKAP_EndTxn, 1, error_code, 0);
+
+        /* Fail the PID reinit */
+        rd_kafka_mock_broker_push_request_error_rtts(
+            mcluster, txn_coord, RD_KAFKAP_InitProducerId, 1, error_code, 0);
+
+        /* Produce a message, let it fail with a fatal idempo error. */
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_Produce, 1,
+            RD_KAFKA_RESP_ERR_UNKNOWN_PRODUCER_ID);
+
+        TEST_CALL_ERR__(rd_kafka_producev(
+            rk, RD_KAFKA_V_TOPIC("mytopic"), RD_KAFKA_V_PARTITION(0),
+            RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END));
+
+        test_flush(rk, -1);
+
+        /* Abort the transaction, should fail with a fatal error */
+        error = rd_kafka_abort_transaction(rk, -1);
+        TEST_ASSERT(error != NULL, "Expected abort_transaction() to fail");
+
+        TEST_SAY_ERROR(error, "abort_transaction() failed: ");
+        TEST_ASSERT(rd_kafka_error_is_fatal(error), "Expected a fatal error");
+        rd_kafka_error_destroy(error);
+
+        fatal_err = rd_kafka_fatal_error(rk, errstr, sizeof(errstr));
+        TEST_ASSERT(fatal_err, "Expected a fatal error to have been raised");
+        TEST_SAY("Fatal error: %s: %s\n", rd_kafka_err2name(fatal_err), errstr);
+
+        /* Verify that the producer sent the expected number of EndTxn requests
+         * by inspecting the mock broker error stack,
+         * which should now be empty. */
+        if (rd_kafka_mock_broker_error_stack_cnt(
+                mcluster, txn_coord, RD_KAFKAP_EndTxn, &errors_cnt)) {
+                TEST_FAIL(
+                    "Broker error count should succeed for API %s"
+                    " on broker %" PRId32,
+                    rd_kafka_ApiKey2str(RD_KAFKAP_EndTxn), txn_coord);
+        }
+        /* Checks all the  RD_KAFKAP_EndTxn responses have been consumed */
+        TEST_ASSERT(errors_cnt == 0,
+                    "Expected error count 0 for API %s, found %zu",
+                    rd_kafka_ApiKey2str(RD_KAFKAP_EndTxn), errors_cnt);
+
+        if (rd_kafka_mock_broker_error_stack_cnt(
+                mcluster, txn_coord, RD_KAFKAP_InitProducerId, &errors_cnt)) {
+                TEST_FAIL(
+                    "Broker error count should succeed for API %s"
+                    " on broker %" PRId32,
+                    rd_kafka_ApiKey2str(RD_KAFKAP_InitProducerId), txn_coord);
+        }
+        /* Checks none of the RD_KAFKAP_InitProducerId responses have been
+         * consumed
+         */
+        TEST_ASSERT(errors_cnt == 1,
+                    "Expected error count 1 for API %s, found %zu",
+                    rd_kafka_ApiKey2str(RD_KAFKAP_InitProducerId), errors_cnt);
+
+        /* All done */
+        rd_kafka_destroy(rk);
+
+        allowed_error = RD_KAFKA_RESP_ERR_NO_ERROR;
+
+        SUB_TEST_PASS();
+}
+
+
+/**
+ * @brief Test that the TxnOffsetCommit op doesn't retry without waiting
+ * if the coordinator is found but not available, causing too frequent retries.
+ */
+static void
+do_test_txn_offset_commit_doesnt_retry_too_quickly(rd_bool_t times_out) {
+        rd_kafka_t *rk;
+        rd_kafka_mock_cluster_t *mcluster;
+        rd_kafka_resp_err_t err;
+        rd_kafka_topic_partition_list_t *offsets;
+        rd_kafka_consumer_group_metadata_t *cgmetadata;
+        rd_kafka_error_t *error;
+        int timeout;
+
+        SUB_TEST_QUICK("times_out=%s", RD_STR_ToF(times_out));
+
+        rk = create_txn_producer(&mcluster, "txnid", 3, NULL);
+
+        test_curr->ignore_dr_err = rd_true;
+
+        TEST_CALL_ERROR__(rd_kafka_init_transactions(rk, 5000));
+
+        TEST_CALL_ERROR__(rd_kafka_begin_transaction(rk));
+
+        err = rd_kafka_producev(rk, RD_KAFKA_V_TOPIC("mytopic"),
+                                RD_KAFKA_V_VALUE("hi", 2), RD_KAFKA_V_END);
+        TEST_ASSERT(!err, "produce failed: %s", rd_kafka_err2str(err));
+
+        /* Wait for messages to be delivered */
+        test_flush(rk, 5000);
+
+        /*
+         * Fail TxnOffsetCommit with COORDINATOR_NOT_AVAILABLE
+         * repeatedly.
+         */
+        rd_kafka_mock_push_request_errors(
+            mcluster, RD_KAFKAP_TxnOffsetCommit, 4,
+            RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE,
+            RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE,
+            RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE,
+            RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE);
+
+        offsets = rd_kafka_topic_partition_list_new(1);
+        rd_kafka_topic_partition_list_add(offsets, "srctopic4", 3)->offset = 1;
+
+        cgmetadata = rd_kafka_consumer_group_metadata_new("mygroupid");
+
+        /* The retry delay is 500ms, with 4 retries it should take at least
+         * 2000ms for this call to succeed. */
+        timeout = times_out ? 500 : 4000;
+        error   = rd_kafka_send_offsets_to_transaction(rk, offsets, cgmetadata,
+                                                       timeout);
+        rd_kafka_consumer_group_metadata_destroy(cgmetadata);
+        rd_kafka_topic_partition_list_destroy(offsets);
+
+        if (times_out) {
+                TEST_ASSERT(rd_kafka_error_code(error) ==
+                                RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE,
+                            "expected %s, got: %s",
+                            rd_kafka_err2name(
+                                RD_KAFKA_RESP_ERR_COORDINATOR_NOT_AVAILABLE),
+                            rd_kafka_err2str(rd_kafka_error_code(error)));
+        } else {
+                TEST_ASSERT(rd_kafka_error_code(error) ==
+                                RD_KAFKA_RESP_ERR_NO_ERROR,
+                            "expected \"Success\", found: %s",
+                            rd_kafka_err2str(rd_kafka_error_code(error)));
+        }
+        rd_kafka_error_destroy(error);
+
+        /* All done */
+        rd_kafka_destroy(rk);
+
+        SUB_TEST_PASS();
+}
+
+
+int main_0105_transactions_mock(int argc, char **argv) {
+        TEST_SKIP_MOCK_CLUSTER(0);
+
+        do_test_txn_recoverable_errors();
+
+        do_test_txn_fatal_idempo_errors();
+
+        do_test_txn_fenced_reinit(RD_KAFKA_RESP_ERR_INVALID_PRODUCER_EPOCH);
+        do_test_txn_fenced_reinit(RD_KAFKA_RESP_ERR_PRODUCER_FENCED);
+
+        do_test_txn_req_cnt();
+
+        do_test_txn_requires_abort_errors();
+
+        do_test_txn_slow_reinit(rd_false);
+        do_test_txn_slow_reinit(rd_true);
+
+        /* Just do a subset of tests in quick mode */
+        if (test_quick)
+                return 0;
+
+        do_test_txn_endtxn_errors();
+
+        do_test_txn_endtxn_infinite();
+
+        do_test_txn_endtxn_timeout();
+
+        do_test_txn_endtxn_timeout_inflight();
+
+        /* Bring down the coordinator */
+        do_test_txn_broker_down_in_txn(rd_true);
+
+        /* Bring down partition leader */
+        do_test_txn_broker_down_in_txn(rd_false);
+
+        do_test_txns_not_supported();
+
+        do_test_txns_send_offsets_concurrent_is_retried();
+
+        do_test_txns_send_offsets_non_eligible();
+
+        do_test_txn_coord_req_destroy();
+
+        do_test_txn_coord_req_multi_find();
+
+        do_test_txn_addparts_req_multi();
+
+        do_test_txns_no_timeout_crash();
+
+        do_test_txn_auth_failure(
+            RD_KAFKAP_InitProducerId,
+            RD_KAFKA_RESP_ERR_CLUSTER_AUTHORIZATION_FAILED);
+
+        do_test_txn_auth_failure(
+            RD_KAFKAP_FindCoordinator,
+            RD_KAFKA_RESP_ERR_CLUSTER_AUTHORIZATION_FAILED);
+
+        do_test_txn_flush_timeout();
+
+        do_test_unstable_offset_commit();
+
+        do_test_commit_after_msg_timeout();
+
+        do_test_txn_switch_coordinator();
+
+        do_test_txn_switch_coordinator_refresh();
+
+        do_test_out_of_order_seq();
+
+        do_test_topic_disappears_for_awhile();
+
+        do_test_disconnected_group_coord(rd_false);
+
+        do_test_disconnected_group_coord(rd_true);
+
+        do_test_txn_coordinator_null_not_fatal();
+
+        do_test_txn_resumable_calls_timeout(rd_true);
+
+        do_test_txn_resumable_calls_timeout(rd_false);
+
+        do_test_txn_resumable_calls_timeout_error(rd_true);
+
+        do_test_txn_resumable_calls_timeout_error(rd_false);
+        do_test_txn_resumable_init();
+
+        do_test_txn_concurrent_operations(rd_true /*commit*/);
+
+        do_test_txn_concurrent_operations(rd_false /*abort*/);
+
+        do_test_txn_fenced_abort(RD_KAFKA_RESP_ERR_INVALID_PRODUCER_EPOCH);
+
+        do_test_txn_fenced_abort(RD_KAFKA_RESP_ERR_PRODUCER_FENCED);
+
+        do_test_txn_offset_commit_doesnt_retry_too_quickly(rd_true);
+
+        do_test_txn_offset_commit_doesnt_retry_too_quickly(rd_false);
+
+        return 0;
+}