kafka-python 2.1.4__tar.gz → 2.2.0__tar.gz

{kafka_python-2.1.4 → kafka_python-2.2.0}/CHANGES.md

@@ -1,3 +1,61 @@
+# 2.2.0 (Apr 28, 2025)
+
+KafkaProducer
+* KIP-98: Add idempotent producer support (#2569)
+* KIP-98: Transactional Producer (#2587)
+* KIP-98: Add offsets support to transactional KafkaProducer (#2590)
+* Prefix producer logs w/ client id and transactional id (#2591)
+* KAFKA-5429: Ignore produce response if batch was previously aborted
+* KIP-91: KafkaProducer `delivery_timeout_ms`
+* Default retries -> infinite
+* Expand KafkaProducer docstring w/ idempotent and transactional notes
+* RecordAccumulator: Use helper method to get/set `_tp_locks`; get dq with lock in reenqueue()
+
+KafkaConsumer
+* KIP-98: Add Consumer support for `READ_COMMITTED` (#2582)
+* KIP-394: handle `MEMBER_ID_REQUIRED` error w/ second join group request (#2598)
+* KAFKA-5078: Defer fetch record exception if iterator has already moved across a valid record
+* KAFKA-5075: Defer consumer fetcher exception if fetch position has already increased
+* KAFKA-4937: Batch offset fetches in the Consumer
+* KAFKA-4547: Avoid resetting paused partitions to committed offsets
+* KAFKA-6397: Consumer should not block setting positions of unavailable partitions (#2593)
+
+Potentially Breaking Changes (internal)
+* Rename CorruptRecordException -> CorruptRecordError
+* Rename Coordinator errors to generic not group (#2585)
+* Rename `ClusterMetadata.add_group_coordinator` -> `add_coordinator` + support txn type
+* Use SaslAuthenticationFailedError in kafka.conn connection failure; Drop unused AuthenticationFailedError
+* Remove old/unused errors; reorder; KafkaTimeout -> retriable
+* Drop `log_start_offset` from producer RecordMetadata
+
+Internal
+* MemoryRecords iterator; MemoryRecordsBuilder records() helper
+* Convert `DefaultRecordsBuilder.size_in_bytes` to classmethod
+
+Fixes
+* Resolve datetime deprecation warnings (#2589)
+* Avoid self refcount in log messages; test thread close on all pythons
+* Fix client.wakeup() race from producer/sender close
+* Fix ElectionNotNeededError handling in admin client
+
+Tests
+* Move integration tests and fixtures to test/integration/; simplify unit fixtures (#2588)
+* Expand Sender test coverage (#2586)
+* py2 test fixups
+* Drop unused KafkaClient import from `test_fetcher`
+
+# 2.1.5 (Apr 4, 2025)
+
+Fixes
+* Fix python2.7 errors (#2578)
+
+Improvements
+* Move benchmark scripts to kafka.benchmarks module (#2584)
+* Use __slots__ for metrics (#2583)
+* Pass `metrics_enabled=False` to disable metrics (#2581)
+* Drop unused kafka.producer.buffer / SimpleBufferPool (#2580)
+* Raise UnsupportedVersionError from coordinator (#2579)
+
 # 2.1.4 (Mar 28, 2025)
 
 Fixes

{kafka_python-2.1.4 → kafka_python-2.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: kafka-python
-Version: 2.1.4
+Version: 2.2.0
 Summary: Pure Python client for Apache Kafka
 Author-email: Dana Powers <dana.powers@gmail.com>
 Project-URL: Homepage, https://github.com/dpkp/kafka-python
@@ -39,6 +39,8 @@ Requires-Dist: pytest; extra == "testing"
 Requires-Dist: mock; python_version < "3.3" and extra == "testing"
 Requires-Dist: pytest-mock; extra == "testing"
 Requires-Dist: pytest-timeout; extra == "testing"
+Provides-Extra: benchmarks
+Requires-Dist: pyperf; extra == "benchmarks"
 
 Kafka Python client
 ------------------------

{kafka_python-2.1.4 → kafka_python-2.2.0}/kafka/admin/client.py

@@ -15,7 +15,7 @@ from kafka.client_async import KafkaClient, selectors
 from kafka.coordinator.protocol import ConsumerProtocolMemberMetadata, ConsumerProtocolMemberAssignment, ConsumerProtocol
 import kafka.errors as Errors
 from kafka.errors import (
-    IncompatibleBrokerVersion, KafkaConfigurationError, NotControllerError, UnknownTopicOrPartitionError,
+    IncompatibleBrokerVersion, KafkaConfigurationError, UnknownTopicOrPartitionError,
     UnrecognizedBrokerVersion, IllegalArgumentError)
 from kafka.metrics import MetricConfig, Metrics
 from kafka.protocol.admin import (
@@ -411,7 +411,7 @@ class KafkaAdminClient(object):
         # extra values (usually the error_message)
         for topic, error_code in map(lambda e: e[:2], topic_error_tuples):
             error_type = Errors.for_code(error_code)
-            if tries and error_type is NotControllerError:
+            if tries and error_type is Errors.NotControllerError:
                 # No need to inspect the rest of the errors for
                 # non-retriable errors because NotControllerError should
                 # either be thrown for all errors or no errors.
@@ -431,13 +431,13 @@ class KafkaAdminClient(object):
         for topic, partition_results in response.replication_election_results:
             for partition_id, error_code in map(lambda e: e[:2], partition_results):
                 error_type = Errors.for_code(error_code)
-                if tries and error_type is NotControllerError:
+                if tries and error_type is Errors.NotControllerError:
                     # No need to inspect the rest of the errors for
                     # non-retriable errors because NotControllerError should
                     # either be thrown for all errors or no errors.
                     self._refresh_controller_id()
                     return False
-                elif error_type not in [Errors.NoError, Errors.ElectionNotNeeded]:
+                elif error_type not in (Errors.NoError, Errors.ElectionNotNeededError):
                     raise error_type(
                         "Request '{}' failed with response '{}'."
                         .format(request, response))
@@ -1460,9 +1460,9 @@ class KafkaAdminClient(object):
             list: List of tuples of Consumer Groups.
 
         Raises:
-            GroupCoordinatorNotAvailableError: The coordinator is not
+            CoordinatorNotAvailableError: The coordinator is not
                 available, so cannot process requests.
-            GroupLoadInProgressError: The coordinator is loading and
+            CoordinatorLoadInProgressError: The coordinator is loading and
                 hence can't process requests.
         """
         # While we return a list, internally use a set to prevent duplicates

kafka_python-2.2.0/kafka/benchmarks/consumer_performance.py

@@ -0,0 +1,142 @@
+#!/usr/bin/env python
+# Adapted from https://github.com/mrafayaleem/kafka-jython
+
+from __future__ import absolute_import, print_function
+
+import argparse
+import pprint
+import sys
+import threading
+import time
+import traceback
+
+from kafka import KafkaConsumer
+
+
+class ConsumerPerformance(object):
+    @staticmethod
+    def run(args):
+        try:
+            props = {}
+            for prop in args.consumer_config:
+                k, v = prop.split('=')
+                try:
+                    v = int(v)
+                except ValueError:
+                    pass
+                if v == 'None':
+                    v = None
+                elif v == 'False':
+                    v = False
+                elif v == 'True':
+                    v = True
+                props[k] = v
+
+            print('Initializing Consumer...')
+            props['bootstrap_servers'] = args.bootstrap_servers
+            props['auto_offset_reset'] = 'earliest'
+            if 'group_id' not in props:
+                props['group_id'] = 'kafka-consumer-benchmark'
+            if 'consumer_timeout_ms' not in props:
+                props['consumer_timeout_ms'] = 10000
+            props['metrics_sample_window_ms'] = args.stats_interval * 1000
+            for k, v in props.items():
+                print('---> {0}={1}'.format(k, v))
+            consumer = KafkaConsumer(args.topic, **props)
+            print('---> group_id={0}'.format(consumer.config['group_id']))
+            print('---> report stats every {0} secs'.format(args.stats_interval))
+            print('---> raw metrics? {0}'.format(args.raw_metrics))
+            timer_stop = threading.Event()
+            timer = StatsReporter(args.stats_interval, consumer,
+                                  event=timer_stop,
+                                  raw_metrics=args.raw_metrics)
+            timer.start()
+            print('-> OK!')
+            print()
+
+            start_time = time.time()
+            records = 0
+            for msg in consumer:
+                records += 1
+                if records >= args.num_records:
+                    break
+
+            end_time = time.time()
+            timer_stop.set()
+            timer.join()
+            print('Consumed {0} records'.format(records))
+            print('Execution time:', end_time - start_time, 'secs')
+
+        except Exception:
+            exc_info = sys.exc_info()
+            traceback.print_exception(*exc_info)
+            sys.exit(1)
+
+
+class StatsReporter(threading.Thread):
+    def __init__(self, interval, consumer, event=None, raw_metrics=False):
+        super(StatsReporter, self).__init__()
+        self.interval = interval
+        self.consumer = consumer
+        self.event = event
+        self.raw_metrics = raw_metrics
+
+    def print_stats(self):
+        metrics = self.consumer.metrics()
+        if self.raw_metrics:
+            pprint.pprint(metrics)
+        else:
+            print('{records-consumed-rate} records/sec ({bytes-consumed-rate} B/sec),'
+                  ' {fetch-latency-avg} latency,'
+                  ' {fetch-rate} fetch/s,'
+                  ' {fetch-size-avg} fetch size,'
+                  ' {records-lag-max} max record lag,'
+                  ' {records-per-request-avg} records/req'
+                  .format(**metrics['consumer-fetch-manager-metrics']))
+
+
+    def print_final(self):
+        self.print_stats()
+
+    def run(self):
+        while self.event and not self.event.wait(self.interval):
+            self.print_stats()
+        else:
+            self.print_final()
+
+
+def get_args_parser():
+    parser = argparse.ArgumentParser(
+        description='This tool is used to verify the consumer performance.')
+
+    parser.add_argument(
+        '--bootstrap-servers', type=str, nargs='+', default=(),
+        help='host:port for cluster bootstrap servers')
+    parser.add_argument(
+        '--topic', type=str,
+        help='Topic for consumer test (default: kafka-python-benchmark-test)',
+        default='kafka-python-benchmark-test')
+    parser.add_argument(
+        '--num-records', type=int,
+        help='number of messages to consume (default: 1000000)',
+        default=1000000)
+    parser.add_argument(
+        '--consumer-config', type=str, nargs='+', default=(),
+        help='kafka consumer related configuration properties like '
+             'bootstrap_servers,client_id etc..')
+    parser.add_argument(
+        '--fixture-compression', type=str,
+        help='specify a compression type for use with broker fixtures / producer')
+    parser.add_argument(
+        '--stats-interval', type=int,
+        help='Interval in seconds for stats reporting to console (default: 5)',
+        default=5)
+    parser.add_argument(
+        '--raw-metrics', action='store_true',
+        help='Enable this flag to print full metrics dict on each interval')
+    return parser
+
+
+if __name__ == '__main__':
+    args = get_args_parser().parse_args()
+    ConsumerPerformance.run(args)

kafka_python-2.2.0/kafka/benchmarks/load_example.py

@@ -0,0 +1,110 @@
+#!/usr/bin/env python
+from __future__ import print_function
+
+import argparse
+import logging
+import threading
+import time
+
+from kafka import KafkaConsumer, KafkaProducer
+
+
+class Producer(threading.Thread):
+
+    def __init__(self, bootstrap_servers, topic, stop_event, msg_size):
+        super(Producer, self).__init__()
+        self.bootstrap_servers = bootstrap_servers
+        self.topic = topic
+        self.stop_event = stop_event
+        self.big_msg = b'1' * msg_size
+
+    def run(self):
+        producer = KafkaProducer(bootstrap_servers=self.bootstrap_servers)
+        self.sent = 0
+
+        while not self.stop_event.is_set():
+            producer.send(self.topic, self.big_msg)
+            self.sent += 1
+        producer.flush()
+        producer.close()
+
+
+class Consumer(threading.Thread):
+    def __init__(self, bootstrap_servers, topic, stop_event, msg_size):
+        super(Consumer, self).__init__()
+        self.bootstrap_servers = bootstrap_servers
+        self.topic = topic
+        self.stop_event = stop_event
+        self.msg_size = msg_size
+
+    def run(self):
+        consumer = KafkaConsumer(bootstrap_servers=self.bootstrap_servers,
+                                 auto_offset_reset='earliest')
+        consumer.subscribe([self.topic])
+        self.valid = 0
+        self.invalid = 0
+
+        for message in consumer:
+            if len(message.value) == self.msg_size:
+                self.valid += 1
+            else:
+                print('Invalid message:', len(message.value), self.msg_size)
+                self.invalid += 1
+
+            if self.stop_event.is_set():
+                break
+        consumer.close()
+
+
+def get_args_parser():
+    parser = argparse.ArgumentParser(
+        description='This tool is used to demonstrate consumer and producer load.')
+
+    parser.add_argument(
+        '--bootstrap-servers', type=str, nargs='+', default=('localhost:9092'),
+        help='host:port for cluster bootstrap servers (default: localhost:9092)')
+    parser.add_argument(
+        '--topic', type=str,
+        help='Topic for load test (default: kafka-python-benchmark-load-example)',
+        default='kafka-python-benchmark-load-example')
+    parser.add_argument(
+        '--msg-size', type=int,
+        help='Message size, in bytes, for load test (default: 524288)',
+        default=524288)
+    parser.add_argument(
+        '--load-time', type=int,
+        help='number of seconds to run load test (default: 10)',
+        default=10)
+    parser.add_argument(
+        '--log-level', type=str,
+        help='Optional logging level for load test: ERROR|INFO|DEBUG etc',
+        default=None)
+    return parser
+
+
+def main(args):
+    if args.log_level:
+        logging.basicConfig(
+            format='%(asctime)s.%(msecs)s:%(name)s:%(thread)d:%(levelname)s:%(process)d:%(message)s',
+            level=getattr(logging, args.log_level))
+    producer_stop = threading.Event()
+    consumer_stop = threading.Event()
+    threads = [
+        Producer(args.bootstrap_servers, args.topic, producer_stop, args.msg_size),
+        Consumer(args.bootstrap_servers, args.topic, consumer_stop, args.msg_size)
+    ]
+
+    for t in threads:
+        t.start()
+
+    time.sleep(args.load_time)
+    producer_stop.set()
+    consumer_stop.set()
+    print('Messages sent: %d' % threads[0].sent)
+    print('Messages recvd: %d' % threads[1].valid)
+    print('Messages invalid: %d' % threads[1].invalid)
+
+
+if __name__ == "__main__":
+    args = get_args_parser().parse_args()
+    main(args)

kafka_python-2.2.0/kafka/benchmarks/producer_performance.py

@@ -0,0 +1,153 @@
+#!/usr/bin/env python
+# Adapted from https://github.com/mrafayaleem/kafka-jython
+
+from __future__ import absolute_import, print_function
+
+import argparse
+import pprint
+import sys
+import threading
+import time
+import traceback
+
+from kafka.vendor.six.moves import range
+
+from kafka import KafkaProducer
+
+
+class ProducerPerformance(object):
+    @staticmethod
+    def run(args):
+        try:
+            props = {}
+            for prop in args.producer_config:
+                k, v = prop.split('=')
+                try:
+                    v = int(v)
+                except ValueError:
+                    pass
+                if v == 'None':
+                    v = None
+                elif v == 'False':
+                    v = False
+                elif v == 'True':
+                    v = True
+                props[k] = v
+
+            print('Initializing producer...')
+            props['bootstrap_servers'] = args.bootstrap_servers
+            record = bytes(bytearray(args.record_size))
+            props['metrics_sample_window_ms'] = args.stats_interval * 1000
+
+            producer = KafkaProducer(**props)
+            for k, v in props.items():
+                print('---> {0}={1}'.format(k, v))
+            print('---> send {0} byte records'.format(args.record_size))
+            print('---> report stats every {0} secs'.format(args.stats_interval))
+            print('---> raw metrics? {0}'.format(args.raw_metrics))
+            timer_stop = threading.Event()
+            timer = StatsReporter(args.stats_interval, producer,
+                                  event=timer_stop,
+                                  raw_metrics=args.raw_metrics)
+            timer.start()
+            print('-> OK!')
+            print()
+
+            def _benchmark():
+                results = []
+                for i in range(args.num_records):
+                    results.append(producer.send(topic=args.topic, value=record))
+                print("Send complete...")
+                producer.flush()
+                producer.close()
+                count_success, count_failure = 0, 0
+                for r in results:
+                    if r.succeeded():
+                        count_success += 1
+                    elif r.failed():
+                        count_failure += 1
+                    else:
+                        raise ValueError(r)
+                print("%d suceeded, %d failed" % (count_success, count_failure))
+
+            start_time = time.time()
+            _benchmark()
+            end_time = time.time()
+            timer_stop.set()
+            timer.join()
+            print('Execution time:', end_time - start_time, 'secs')
+
+        except Exception:
+            exc_info = sys.exc_info()
+            traceback.print_exception(*exc_info)
+            sys.exit(1)
+
+
+class StatsReporter(threading.Thread):
+    def __init__(self, interval, producer, event=None, raw_metrics=False):
+        super(StatsReporter, self).__init__()
+        self.interval = interval
+        self.producer = producer
+        self.event = event
+        self.raw_metrics = raw_metrics
+
+    def print_stats(self):
+        metrics = self.producer.metrics()
+        if not metrics:
+            return
+        if self.raw_metrics:
+            pprint.pprint(metrics)
+        else:
+            print('{record-send-rate} records/sec ({byte-rate} B/sec),'
+                  ' {request-latency-avg} latency,'
+                  ' {record-size-avg} record size,'
+                  ' {batch-size-avg} batch size,'
+                  ' {records-per-request-avg} records/req'
+                  .format(**metrics['producer-metrics']))
+
+    def print_final(self):
+        self.print_stats()
+
+    def run(self):
+        while self.event and not self.event.wait(self.interval):
+            self.print_stats()
+        else:
+            self.print_final()
+
+
+def get_args_parser():
+    parser = argparse.ArgumentParser(
+        description='This tool is used to verify the producer performance.')
+
+    parser.add_argument(
+        '--bootstrap-servers', type=str, nargs='+', default=(),
+        help='host:port for cluster bootstrap server')
+    parser.add_argument(
+        '--topic', type=str,
+        help='Topic name for test (default: kafka-python-benchmark-test)',
+        default='kafka-python-benchmark-test')
+    parser.add_argument(
+        '--num-records', type=int,
+        help='number of messages to produce (default: 1000000)',
+        default=1000000)
+    parser.add_argument(
+        '--record-size', type=int,
+        help='message size in bytes (default: 100)',
+        default=100)
+    parser.add_argument(
+        '--producer-config', type=str, nargs='+', default=(),
+        help='kafka producer related configuaration properties like '
+             'bootstrap_servers,client_id etc..')
+    parser.add_argument(
+        '--stats-interval', type=int,
+        help='Interval in seconds for stats reporting to console (default: 5)',
+        default=5)
+    parser.add_argument(
+        '--raw-metrics', action='store_true',
+        help='Enable this flag to print full metrics dict on each interval')
+    return parser
+
+
+if __name__ == '__main__':
+    args = get_args_parser().parse_args()
+    ProducerPerformance.run(args)

kafka_python-2.2.0/kafka/benchmarks/record_batch_compose.py

@@ -0,0 +1,78 @@
+#!/usr/bin/env python3
+from __future__ import print_function
+import hashlib
+import itertools
+import os
+import random
+
+import pyperf
+
+from kafka.record.memory_records import MemoryRecordsBuilder
+
+
+DEFAULT_BATCH_SIZE = 1600 * 1024
+KEY_SIZE = 6
+VALUE_SIZE = 60
+TIMESTAMP_RANGE = [1505824130000, 1505824140000]
+
+# With values above v1 record is 100 bytes, so 10 000 bytes for 100 messages
+MESSAGES_PER_BATCH = 100
+
+
+def random_bytes(length):
+    buffer = bytearray(length)
+    for i in range(length):
+        buffer[i] = random.randint(0, 255)
+    return bytes(buffer)
+
+
+def prepare():
+    return iter(itertools.cycle([
+        (random_bytes(KEY_SIZE),
+         random_bytes(VALUE_SIZE),
+         random.randint(*TIMESTAMP_RANGE)
+         )
+        for _ in range(int(MESSAGES_PER_BATCH * 1.94))
+    ]))
+
+
+def finalize(results):
+    # Just some strange code to make sure PyPy does execute the main code
+    # properly, without optimizing it away
+    hash_val = hashlib.md5()
+    for buf in results:
+        hash_val.update(buf)
+    print(hash_val, file=open(os.devnull, "w"))
+
+
+def func(loops, magic):
+    # Jit can optimize out the whole function if the result is the same each
+    # time, so we need some randomized input data )
+    precomputed_samples = prepare()
+    results = []
+
+    # Main benchmark code.
+    t0 = pyperf.perf_counter()
+    for _ in range(loops):
+        batch = MemoryRecordsBuilder(
+            magic, batch_size=DEFAULT_BATCH_SIZE, compression_type=0)
+        for _ in range(MESSAGES_PER_BATCH):
+            key, value, timestamp = next(precomputed_samples)
+            size = batch.append(
+                timestamp=timestamp, key=key, value=value)
+            assert size
+        batch.close()
+        results.append(batch.buffer())
+
+    res = pyperf.perf_counter() - t0
+
+    finalize(results)
+
+    return res
+
+
+if __name__ == '__main__':
+    runner = pyperf.Runner()
+    runner.bench_time_func('batch_append_v0', func, 0)
+    runner.bench_time_func('batch_append_v1', func, 1)
+    runner.bench_time_func('batch_append_v2', func, 2)

kafka_python-2.2.0/kafka/benchmarks/record_batch_read.py

@@ -0,0 +1,83 @@
+#!/usr/bin/env python
+from __future__ import print_function
+import hashlib
+import itertools
+import os
+import random
+
+import pyperf
+
+from kafka.record.memory_records import MemoryRecords, MemoryRecordsBuilder
+
+
+DEFAULT_BATCH_SIZE = 1600 * 1024
+KEY_SIZE = 6
+VALUE_SIZE = 60
+TIMESTAMP_RANGE = [1505824130000, 1505824140000]
+
+BATCH_SAMPLES = 5
+MESSAGES_PER_BATCH = 100
+
+
+def random_bytes(length):
+    buffer = bytearray(length)
+    for i in range(length):
+        buffer[i] = random.randint(0, 255)
+    return bytes(buffer)
+
+
+def prepare(magic):
+    samples = []
+    for _ in range(BATCH_SAMPLES):
+        batch = MemoryRecordsBuilder(
+            magic, batch_size=DEFAULT_BATCH_SIZE, compression_type=0)
+        for _ in range(MESSAGES_PER_BATCH):
+            size = batch.append(
+                random.randint(*TIMESTAMP_RANGE),
+                random_bytes(KEY_SIZE),
+                random_bytes(VALUE_SIZE),
+                headers=[])
+            assert size
+        batch.close()
+        samples.append(bytes(batch.buffer()))
+
+    return iter(itertools.cycle(samples))
+
+
+def finalize(results):
+    # Just some strange code to make sure PyPy does execute the code above
+    # properly
+    hash_val = hashlib.md5()
+    for buf in results:
+        hash_val.update(buf)
+    print(hash_val, file=open(os.devnull, "w"))
+
+
+def func(loops, magic):
+    # Jit can optimize out the whole function if the result is the same each
+    # time, so we need some randomized input data )
+    precomputed_samples = prepare(magic)
+    results = []
+
+    # Main benchmark code.
+    batch_data = next(precomputed_samples)
+    t0 = pyperf.perf_counter()
+    for _ in range(loops):
+        records = MemoryRecords(batch_data)
+        while records.has_next():
+            batch = records.next_batch()
+            batch.validate_crc()
+            for record in batch:
+                results.append(record.value)
+
+    res = pyperf.perf_counter() - t0
+    finalize(results)
+
+    return res
+
+
+if __name__ == '__main__':
+    runner = pyperf.Runner()
+    runner.bench_time_func('batch_read_v0', func, 0)
+    runner.bench_time_func('batch_read_v1', func, 1)
+    runner.bench_time_func('batch_read_v2', func, 2)