sentry-arroyo 2.32.2__py3-none-any.whl → 2.32.3__py3-none-any.whl

arroyo/backends/kafka/consumer.py

@@ -157,21 +157,32 @@ class KafkaConsumer(Consumer[KafkaPayload]):
         self,
         configuration: Mapping[str, Any],
     ) -> None:
+        configuration = dict(configuration)
+
+        # Feature flag to enable retrying on `Broker handle destroyed` errors
+        # which can occur if we attempt to commit during a rebalance when
+        # the consumer group coordinator changed
+        self.__retry_handle_destroyed = as_kafka_configuration_bool(
+            configuration.pop("arroyo.retry.broker.handle.destroyed", False)
+        )
+
+        retryable_errors: Tuple[int, ...] = (
+            KafkaError.REQUEST_TIMED_OUT,
+            KafkaError.NOT_COORDINATOR,
+            KafkaError._WAIT_COORD,
+            KafkaError.STALE_MEMBER_EPOCH,  # kip-848
+            KafkaError.COORDINATOR_LOAD_IN_PROGRESS,
+        )
+        if self.__retry_handle_destroyed:
+            retryable_errors += (KafkaError._DESTROY,)
+
         commit_retry_policy = BasicRetryPolicy(
             3,
             1,
             lambda e: isinstance(e, KafkaException)
-            and e.args[0].code()
-            in (
-                KafkaError.REQUEST_TIMED_OUT,
-                KafkaError.NOT_COORDINATOR,
-                KafkaError._WAIT_COORD,
-                KafkaError.STALE_MEMBER_EPOCH,  # kip-848
-                KafkaError.COORDINATOR_LOAD_IN_PROGRESS,
-            ),
+            and e.args[0].code() in retryable_errors,
         )
 
-        configuration = dict(configuration)
         self.__is_cooperative_sticky = (
             configuration.get("partition.assignment.strategy") == "cooperative-sticky"
         )
@@ -253,6 +264,9 @@ class KafkaConsumer(Consumer[KafkaPayload]):
 
         self.__state = KafkaConsumerState.CONSUMING
 
+        self.__metrics = get_metrics()
+        self.__group_id = configuration.get("group.id")
+
     def __on_commit_callback(
         self,
         error: Optional[KafkaException],
@@ -265,6 +279,23 @@ class KafkaConsumer(Consumer[KafkaPayload]):
                 error,
                 partition_info,
             )
+            tags = {"status": "error"}
+            if self.__group_id:
+                tags["group_id"] = self.__group_id
+            self.__metrics.increment(
+                name="arroyo.consumer.commit_status",
+                value=1,
+                tags=tags,
+            )
+        else:
+            tags = {"status": "success"}
+            if self.__group_id:
+                tags["group_id"] = self.__group_id
+            self.__metrics.increment(
+                name="arroyo.consumer.commit_status",
+                value=1,
+                tags=tags,
+            )
 
     def __resolve_partition_offset_earliest(
         self, partition: ConfluentTopicPartition

arroyo/utils/metricDefs.json

@@ -1 +1 @@
-{"arroyo.strategies.run_task_with_multiprocessing.batch.size.msg": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.size.msg", "type": "Time", "description": "Number of messages in a multiprocessing batch"}, "arroyo.strategies.run_task_with_multiprocessing.batch.size.bytes": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.size.bytes", "type": "Time", "description": "Number of bytes in a multiprocessing batch"}, "arroyo.strategies.run_task_with_multiprocessing.batch.submit.time": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.submit.time", "type": "Time", "description": "How long it took to submit a batch to multiprocessing"}, "arroyo.strategies.run_task_with_multiprocessing.output_batch.size.msg": {"name": "arroyo.strategies.run_task_with_multiprocessing.output_batch.size.msg", "type": "Time", "description": "Number of messages in a multiprocessing batch after the message transformation"}, "arroyo.strategies.run_task_with_multiprocessing.output_batch.size.bytes": {"name": "arroyo.strategies.run_task_with_multiprocessing.output_batch.size.bytes", "type": "Time", "description": "Number of bytes in a multiprocessing batch after the message transformation"}, "arroyo.consumer.run.count": {"name": "arroyo.consumer.run.count", "type": "Counter", "description": "Number of times the consumer is spinning"}, "arroyo.consumer.invalid_message.count": {"name": "arroyo.consumer.invalid_message.count", "type": "Counter", "description": "Number of times the consumer encountered an invalid message."}, "arroyo.strategies.reduce.batch_time": {"name": "arroyo.strategies.reduce.batch_time", "type": "Time", "description": "How long it took the Reduce step to fill up a batch"}, "arroyo.strategies.run_task_with_multiprocessing.batch.backpressure": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.backpressure", "type": "Counter", "description": "Incremented when a strategy after multiprocessing applies\nbackpressure to multiprocessing. May be a reason why CPU cannot be\nsaturated."}, "arroyo.strategies.run_task_with_multiprocessing.batch.input.overflow": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.input.overflow", "type": "Counter", "description": "Incremented when multiprocessing cannot fill the input batch\nbecause not enough memory was allocated. This results in batches smaller\nthan configured. Increase `input_block_size` to fix."}, "arroyo.strategies.run_task_with_multiprocessing.batch.output.overflow": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.output.overflow", "type": "Counter", "description": "Incremented when multiprocessing cannot pull results in batches\nequal to the input batch size, because not enough memory was allocated.\nThis can be devastating for throughput. Increase `output_block_size` to\nfix."}, "arroyo.strategies.run_task_with_multiprocessing.batch.input.resize": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.input.resize", "type": "Counter", "description": "Arroyo has decided to re-allocate a block in order to combat input\nbuffer overflow. This behavior can be disabled by explicitly setting\n`input_block_size` to a not-None value in `RunTaskWithMultiprocessing`."}, "arroyo.strategies.run_task_with_multiprocessing.batch.output.resize": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.output.resize", "type": "Counter", "description": "Arroyo has decided to re-allocate a block in order to combat output\nbuffer overflow. This behavior can be disabled by explicitly setting\n`output_block_size` to a not-None value in `RunTaskWithMultiprocessing`."}, "arroyo.strategies.run_task_with_multiprocessing.batches_in_progress": {"name": "arroyo.strategies.run_task_with_multiprocessing.batches_in_progress", "type": "Gauge", "description": "How many batches are being processed in parallel by multiprocessing."}, "arroyo.strategies.run_task_with_multiprocessing.processes": {"name": "arroyo.strategies.run_task_with_multiprocessing.processes", "type": "Gauge", "description": "Shows the total number of available processes in the pool."}, "arroyo.strategies.run_task_with_multiprocessing.pool.create": {"name": "arroyo.strategies.run_task_with_multiprocessing.pool.create", "type": "Counter", "description": "A subprocess by multiprocessing unexpectedly died.\n\"sigchld.detected\",\nCounter: Incremented when the multiprocessing pool is created (or re-created)."}, "arroyo.consumer.poll.time": {"name": "arroyo.consumer.poll.time", "type": "Time", "description": "(unitless) spent polling librdkafka for new messages."}, "arroyo.consumer.processing.time": {"name": "arroyo.consumer.processing.time", "type": "Time", "description": "(unitless) spent in strategies (blocking in strategy.submit or\nstrategy.poll)"}, "arroyo.consumer.backpressure.time": {"name": "arroyo.consumer.backpressure.time", "type": "Time", "description": "(unitless) spent pausing the consumer due to backpressure (MessageRejected)"}, "arroyo.consumer.dlq.time": {"name": "arroyo.consumer.dlq.time", "type": "Time", "description": "(unitless) spent in handling `InvalidMessage` exceptions and sending\nmessages to the the DLQ."}, "arroyo.consumer.join.time": {"name": "arroyo.consumer.join.time", "type": "Time", "description": "(unitless) spent in waiting for the strategy to exit, such as during\nshutdown or rebalancing."}, "arroyo.consumer.callback.time": {"name": "arroyo.consumer.callback.time", "type": "Time", "description": "(unitless) spent in librdkafka callbacks. This metric's timings\noverlap other timings, and might spike at the same time."}, "arroyo.consumer.shutdown.time": {"name": "arroyo.consumer.shutdown.time", "type": "Time", "description": "(unitless) spent in shutting down the consumer. This metric's\ntimings overlap other timings, and might spike at the same time."}, "arroyo.consumer.run.callback": {"name": "arroyo.consumer.run.callback", "type": "Time", "description": "A regular duration metric where each datapoint is measuring the time it\ntook to execute a single callback. This metric is distinct from the\narroyo.consumer.*.time metrics as it does not attempt to accumulate time\nspent per second in an attempt to keep monitoring overhead low.\nThe metric is tagged by the name of the internal callback function being\nexecuted, as 'callback_name'. Possible values are on_partitions_assigned\nand on_partitions_revoked."}, "arroyo.consumer.run.close_strategy": {"name": "arroyo.consumer.run.close_strategy", "type": "Time", "description": "Duration metric measuring the time it took to flush in-flight messages\nand shut down the strategies."}, "arroyo.consumer.run.create_strategy": {"name": "arroyo.consumer.run.create_strategy", "type": "Time", "description": "Duration metric measuring the time it took to create the processing strategy."}, "arroyo.consumer.partitions_revoked.count": {"name": "arroyo.consumer.partitions_revoked.count", "type": "Counter", "description": "How many partitions have been revoked just now."}, "arroyo.consumer.partitions_assigned.count": {"name": "arroyo.consumer.partitions_assigned.count", "type": "Counter", "description": "How many partitions have been assigned just now."}, "arroyo.consumer.latency": {"name": "arroyo.consumer.latency", "type": "Time", "description": "Consumer latency in seconds. Recorded by the commit offsets strategy."}, "arroyo.consumer.pause": {"name": "arroyo.consumer.pause", "type": "Counter", "description": "Metric for when the underlying rdkafka consumer is being paused.\nThis flushes internal prefetch buffers."}, "arroyo.consumer.resume": {"name": "arroyo.consumer.resume", "type": "Counter", "description": "Metric for when the underlying rdkafka consumer is being resumed.\nThis might cause increased network usage as messages are being re-fetched."}, "arroyo.consumer.librdkafka.total_queue_size": {"name": "arroyo.consumer.librdkafka.total_queue_size", "type": "Gauge", "description": "Queue size of background queue that librdkafka uses to prefetch messages."}, "arroyo.processing.strategies.healthcheck.touch": {"name": "arroyo.processing.strategies.healthcheck.touch", "type": "Counter", "description": "Counter metric to measure how often the healthcheck file has been touched."}, "arroyo.strategies.filter.dropped_messages": {"name": "arroyo.strategies.filter.dropped_messages", "type": "Counter", "description": "Number of messages dropped in the FilterStep strategy"}, "arroyo.consumer.dlq.dropped_messages": {"name": "arroyo.consumer.dlq.dropped_messages", "type": "Counter", "description": "how many messages are dropped due to errors producing to the dlq"}, "arroyo.consumer.dlq_buffer.len": {"name": "arroyo.consumer.dlq_buffer.len", "type": "Gauge", "description": "Current length of the DLQ buffer deque"}, "arroyo.consumer.dlq_buffer.exceeded": {"name": "arroyo.consumer.dlq_buffer.exceeded", "type": "Counter", "description": "Number of times the DLQ buffer size has been exceeded, causing messages to be dropped"}, "arroyo.consumer.dlq_buffer.assigned_partitions": {"name": "arroyo.consumer.dlq_buffer.assigned_partitions", "type": "Gauge", "description": "Number of partitions being tracked in the DLQ buffer"}, "arroyo.producer.librdkafka.p99_int_latency": {"name": "arroyo.producer.librdkafka.p99_int_latency", "type": "Time", "description": "Internal producer queue latency from librdkafka statistics.\nTagged by broker_id."}, "arroyo.producer.librdkafka.p99_outbuf_latency": {"name": "arroyo.producer.librdkafka.p99_outbuf_latency", "type": "Time", "description": "Output buffer latency from librdkafka statistics.\nTagged by broker_id."}, "arroyo.producer.librdkafka.p99_rtt": {"name": "arroyo.producer.librdkafka.p99_rtt", "type": "Time", "description": "Round-trip time to brokers from librdkafka statistics.\nTagged by broker_id."}, "arroyo.producer.librdkafka.avg_int_latency": {"name": "arroyo.producer.librdkafka.avg_int_latency", "type": "Time", "description": "Average internal producer queue latency from librdkafka statistics.\nTagged by broker_id."}, "arroyo.producer.librdkafka.avg_outbuf_latency": {"name": "arroyo.producer.librdkafka.avg_outbuf_latency", "type": "Time", "description": "Average output buffer latency from librdkafka statistics.\nTagged by broker_id."}, "arroyo.producer.librdkafka.avg_rtt": {"name": "arroyo.producer.librdkafka.avg_rtt", "type": "Time", "description": "Average round-trip time to brokers from librdkafka statistics.\nTagged by broker_id."}, "arroyo.producer.produce_status": {"name": "arroyo.producer.produce_status", "type": "Counter", "description": "Number of times the produce strategy failed to produce a message"}, "arroyo.producer.librdkafka.message_count": {"name": "arroyo.producer.librdkafka.message_count", "type": "Gauge", "description": "Producer message count metric from librdkafka statistics\nTagged by producer_name"}, "arroyo.producer.librdkafka.message_count_max": {"name": "arroyo.producer.librdkafka.message_count_max", "type": "Gauge", "description": "Maximum producer message count from librdkafka statistics\nTagged by producer_name"}, "arroyo.producer.librdkafka.message_size": {"name": "arroyo.producer.librdkafka.message_size", "type": "Gauge", "description": "Producer message size from librdkafka statistics\nTagged by producer_name"}, "arroyo.producer.librdkafka.message_size_max": {"name": "arroyo.producer.librdkafka.message_size_max", "type": "Gauge", "description": "Maximum producer message size from librdkafka statistics\nTagged by producer_name"}, "arroyo.producer.librdkafka.txmsgs": {"name": "arroyo.producer.librdkafka.txmsgs", "type": "Gauge", "description": "Total number of messages transmitted from librdkafka statistics\nTagged by producer_name"}, "arroyo.producer.librdkafka.broker_tx": {"name": "arroyo.producer.librdkafka.broker_tx", "type": "Gauge", "description": "Total number of transmission requests from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_txbytes": {"name": "arroyo.producer.librdkafka.broker_txbytes", "type": "Gauge", "description": "Total number of bytes transmitted from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_outbuf_requests": {"name": "arroyo.producer.librdkafka.broker_outbuf_requests", "type": "Gauge", "description": "Number of requests awaiting transmission to broker from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_outbuf_messages": {"name": "arroyo.producer.librdkafka.broker_outbuf_messages", "type": "Gauge", "description": "Number of messages awaiting transmission to broker from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_connects": {"name": "arroyo.producer.librdkafka.broker_connects", "type": "Gauge", "description": "Number of connection attempts to broker from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_disconnects": {"name": "arroyo.producer.librdkafka.broker_disconnects", "type": "Gauge", "description": "Number of disconnections from broker from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_txerrs": {"name": "arroyo.producer.librdkafka.broker_txerrs", "type": "Gauge", "description": "Total number of transmission errors from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_txretries": {"name": "arroyo.producer.librdkafka.broker_txretries", "type": "Gauge", "description": "Total number of request retries from librdkafka statistics\nTagged by broker_id, producer_name"}}
+{"arroyo.strategies.run_task_with_multiprocessing.batch.size.msg": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.size.msg", "type": "Time", "description": "Number of messages in a multiprocessing batch"}, "arroyo.strategies.run_task_with_multiprocessing.batch.size.bytes": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.size.bytes", "type": "Time", "description": "Number of bytes in a multiprocessing batch"}, "arroyo.strategies.run_task_with_multiprocessing.batch.submit.time": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.submit.time", "type": "Time", "description": "How long it took to submit a batch to multiprocessing"}, "arroyo.strategies.run_task_with_multiprocessing.output_batch.size.msg": {"name": "arroyo.strategies.run_task_with_multiprocessing.output_batch.size.msg", "type": "Time", "description": "Number of messages in a multiprocessing batch after the message transformation"}, "arroyo.strategies.run_task_with_multiprocessing.output_batch.size.bytes": {"name": "arroyo.strategies.run_task_with_multiprocessing.output_batch.size.bytes", "type": "Time", "description": "Number of bytes in a multiprocessing batch after the message transformation"}, "arroyo.consumer.run.count": {"name": "arroyo.consumer.run.count", "type": "Counter", "description": "Number of times the consumer is spinning"}, "arroyo.consumer.invalid_message.count": {"name": "arroyo.consumer.invalid_message.count", "type": "Counter", "description": "Number of times the consumer encountered an invalid message."}, "arroyo.strategies.reduce.batch_time": {"name": "arroyo.strategies.reduce.batch_time", "type": "Time", "description": "How long it took the Reduce step to fill up a batch"}, "arroyo.strategies.run_task_with_multiprocessing.batch.backpressure": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.backpressure", "type": "Counter", "description": "Incremented when a strategy after multiprocessing applies\nbackpressure to multiprocessing. May be a reason why CPU cannot be\nsaturated."}, "arroyo.strategies.run_task_with_multiprocessing.batch.input.overflow": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.input.overflow", "type": "Counter", "description": "Incremented when multiprocessing cannot fill the input batch\nbecause not enough memory was allocated. This results in batches smaller\nthan configured. Increase `input_block_size` to fix."}, "arroyo.strategies.run_task_with_multiprocessing.batch.output.overflow": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.output.overflow", "type": "Counter", "description": "Incremented when multiprocessing cannot pull results in batches\nequal to the input batch size, because not enough memory was allocated.\nThis can be devastating for throughput. Increase `output_block_size` to\nfix."}, "arroyo.strategies.run_task_with_multiprocessing.batch.input.resize": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.input.resize", "type": "Counter", "description": "Arroyo has decided to re-allocate a block in order to combat input\nbuffer overflow. This behavior can be disabled by explicitly setting\n`input_block_size` to a not-None value in `RunTaskWithMultiprocessing`."}, "arroyo.strategies.run_task_with_multiprocessing.batch.output.resize": {"name": "arroyo.strategies.run_task_with_multiprocessing.batch.output.resize", "type": "Counter", "description": "Arroyo has decided to re-allocate a block in order to combat output\nbuffer overflow. This behavior can be disabled by explicitly setting\n`output_block_size` to a not-None value in `RunTaskWithMultiprocessing`."}, "arroyo.strategies.run_task_with_multiprocessing.batches_in_progress": {"name": "arroyo.strategies.run_task_with_multiprocessing.batches_in_progress", "type": "Gauge", "description": "How many batches are being processed in parallel by multiprocessing."}, "arroyo.strategies.run_task_with_multiprocessing.processes": {"name": "arroyo.strategies.run_task_with_multiprocessing.processes", "type": "Gauge", "description": "Shows the total number of available processes in the pool."}, "arroyo.strategies.run_task_with_multiprocessing.pool.create": {"name": "arroyo.strategies.run_task_with_multiprocessing.pool.create", "type": "Counter", "description": "A subprocess by multiprocessing unexpectedly died.\n\"sigchld.detected\",\nCounter: Incremented when the multiprocessing pool is created (or re-created)."}, "arroyo.consumer.poll.time": {"name": "arroyo.consumer.poll.time", "type": "Time", "description": "(unitless) spent polling librdkafka for new messages."}, "arroyo.consumer.processing.time": {"name": "arroyo.consumer.processing.time", "type": "Time", "description": "(unitless) spent in strategies (blocking in strategy.submit or\nstrategy.poll)"}, "arroyo.consumer.backpressure.time": {"name": "arroyo.consumer.backpressure.time", "type": "Time", "description": "(unitless) spent pausing the consumer due to backpressure (MessageRejected)"}, "arroyo.consumer.dlq.time": {"name": "arroyo.consumer.dlq.time", "type": "Time", "description": "(unitless) spent in handling `InvalidMessage` exceptions and sending\nmessages to the the DLQ."}, "arroyo.consumer.join.time": {"name": "arroyo.consumer.join.time", "type": "Time", "description": "(unitless) spent in waiting for the strategy to exit, such as during\nshutdown or rebalancing."}, "arroyo.consumer.callback.time": {"name": "arroyo.consumer.callback.time", "type": "Time", "description": "(unitless) spent in librdkafka callbacks. This metric's timings\noverlap other timings, and might spike at the same time."}, "arroyo.consumer.shutdown.time": {"name": "arroyo.consumer.shutdown.time", "type": "Time", "description": "(unitless) spent in shutting down the consumer. This metric's\ntimings overlap other timings, and might spike at the same time."}, "arroyo.consumer.run.callback": {"name": "arroyo.consumer.run.callback", "type": "Time", "description": "A regular duration metric where each datapoint is measuring the time it\ntook to execute a single callback. This metric is distinct from the\narroyo.consumer.*.time metrics as it does not attempt to accumulate time\nspent per second in an attempt to keep monitoring overhead low.\nThe metric is tagged by the name of the internal callback function being\nexecuted, as 'callback_name'. Possible values are on_partitions_assigned\nand on_partitions_revoked."}, "arroyo.consumer.run.close_strategy": {"name": "arroyo.consumer.run.close_strategy", "type": "Time", "description": "Duration metric measuring the time it took to flush in-flight messages\nand shut down the strategies."}, "arroyo.consumer.run.create_strategy": {"name": "arroyo.consumer.run.create_strategy", "type": "Time", "description": "Duration metric measuring the time it took to create the processing strategy."}, "arroyo.consumer.partitions_revoked.count": {"name": "arroyo.consumer.partitions_revoked.count", "type": "Counter", "description": "How many partitions have been revoked just now."}, "arroyo.consumer.partitions_assigned.count": {"name": "arroyo.consumer.partitions_assigned.count", "type": "Counter", "description": "How many partitions have been assigned just now."}, "arroyo.consumer.latency": {"name": "arroyo.consumer.latency", "type": "Time", "description": "Consumer latency in seconds. Recorded by the commit offsets strategy."}, "arroyo.consumer.pause": {"name": "arroyo.consumer.pause", "type": "Counter", "description": "Metric for when the underlying rdkafka consumer is being paused.\nThis flushes internal prefetch buffers."}, "arroyo.consumer.resume": {"name": "arroyo.consumer.resume", "type": "Counter", "description": "Metric for when the underlying rdkafka consumer is being resumed.\nThis might cause increased network usage as messages are being re-fetched."}, "arroyo.consumer.librdkafka.total_queue_size": {"name": "arroyo.consumer.librdkafka.total_queue_size", "type": "Gauge", "description": "Queue size of background queue that librdkafka uses to prefetch messages."}, "arroyo.processing.strategies.healthcheck.touch": {"name": "arroyo.processing.strategies.healthcheck.touch", "type": "Counter", "description": "Counter metric to measure how often the healthcheck file has been touched."}, "arroyo.strategies.filter.dropped_messages": {"name": "arroyo.strategies.filter.dropped_messages", "type": "Counter", "description": "Number of messages dropped in the FilterStep strategy"}, "arroyo.consumer.dlq.dropped_messages": {"name": "arroyo.consumer.dlq.dropped_messages", "type": "Counter", "description": "how many messages are dropped due to errors producing to the dlq"}, "arroyo.consumer.commit_status": {"name": "arroyo.consumer.commit_status", "type": "Counter", "description": "Number of times consumer commit succeeds or fails"}, "arroyo.consumer.dlq_buffer.len": {"name": "arroyo.consumer.dlq_buffer.len", "type": "Gauge", "description": "Current length of the DLQ buffer deque"}, "arroyo.consumer.dlq_buffer.exceeded": {"name": "arroyo.consumer.dlq_buffer.exceeded", "type": "Counter", "description": "Number of times the DLQ buffer size has been exceeded, causing messages to be dropped"}, "arroyo.consumer.dlq_buffer.assigned_partitions": {"name": "arroyo.consumer.dlq_buffer.assigned_partitions", "type": "Gauge", "description": "Number of partitions being tracked in the DLQ buffer"}, "arroyo.producer.librdkafka.p99_int_latency": {"name": "arroyo.producer.librdkafka.p99_int_latency", "type": "Time", "description": "Internal producer queue latency from librdkafka statistics.\nTagged by broker_id."}, "arroyo.producer.librdkafka.p99_outbuf_latency": {"name": "arroyo.producer.librdkafka.p99_outbuf_latency", "type": "Time", "description": "Output buffer latency from librdkafka statistics.\nTagged by broker_id."}, "arroyo.producer.librdkafka.p99_rtt": {"name": "arroyo.producer.librdkafka.p99_rtt", "type": "Time", "description": "Round-trip time to brokers from librdkafka statistics.\nTagged by broker_id."}, "arroyo.producer.librdkafka.avg_int_latency": {"name": "arroyo.producer.librdkafka.avg_int_latency", "type": "Time", "description": "Average internal producer queue latency from librdkafka statistics.\nTagged by broker_id."}, "arroyo.producer.librdkafka.avg_outbuf_latency": {"name": "arroyo.producer.librdkafka.avg_outbuf_latency", "type": "Time", "description": "Average output buffer latency from librdkafka statistics.\nTagged by broker_id."}, "arroyo.producer.librdkafka.avg_rtt": {"name": "arroyo.producer.librdkafka.avg_rtt", "type": "Time", "description": "Average round-trip time to brokers from librdkafka statistics.\nTagged by broker_id."}, "arroyo.producer.produce_status": {"name": "arroyo.producer.produce_status", "type": "Counter", "description": "Number of times the produce strategy succeeds or fails"}, "arroyo.producer.librdkafka.message_count": {"name": "arroyo.producer.librdkafka.message_count", "type": "Gauge", "description": "Producer message count metric from librdkafka statistics\nTagged by producer_name"}, "arroyo.producer.librdkafka.message_count_max": {"name": "arroyo.producer.librdkafka.message_count_max", "type": "Gauge", "description": "Maximum producer message count from librdkafka statistics\nTagged by producer_name"}, "arroyo.producer.librdkafka.message_size": {"name": "arroyo.producer.librdkafka.message_size", "type": "Gauge", "description": "Producer message size from librdkafka statistics\nTagged by producer_name"}, "arroyo.producer.librdkafka.message_size_max": {"name": "arroyo.producer.librdkafka.message_size_max", "type": "Gauge", "description": "Maximum producer message size from librdkafka statistics\nTagged by producer_name"}, "arroyo.producer.librdkafka.txmsgs": {"name": "arroyo.producer.librdkafka.txmsgs", "type": "Gauge", "description": "Total number of messages transmitted from librdkafka statistics\nTagged by producer_name"}, "arroyo.producer.librdkafka.broker_tx": {"name": "arroyo.producer.librdkafka.broker_tx", "type": "Gauge", "description": "Total number of transmission requests from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_txbytes": {"name": "arroyo.producer.librdkafka.broker_txbytes", "type": "Gauge", "description": "Total number of bytes transmitted from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_outbuf_requests": {"name": "arroyo.producer.librdkafka.broker_outbuf_requests", "type": "Gauge", "description": "Number of requests awaiting transmission to broker from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_outbuf_messages": {"name": "arroyo.producer.librdkafka.broker_outbuf_messages", "type": "Gauge", "description": "Number of messages awaiting transmission to broker from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_connects": {"name": "arroyo.producer.librdkafka.broker_connects", "type": "Gauge", "description": "Number of connection attempts to broker from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_disconnects": {"name": "arroyo.producer.librdkafka.broker_disconnects", "type": "Gauge", "description": "Number of disconnections from broker from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_txerrs": {"name": "arroyo.producer.librdkafka.broker_txerrs", "type": "Gauge", "description": "Total number of transmission errors from librdkafka statistics\nTagged by broker_id, producer_name"}, "arroyo.producer.librdkafka.broker_txretries": {"name": "arroyo.producer.librdkafka.broker_txretries", "type": "Gauge", "description": "Total number of request retries from librdkafka statistics\nTagged by broker_id, producer_name"}}

arroyo/utils/metric_defs.py

@@ -101,6 +101,8 @@ MetricName = Literal[
     "arroyo.strategies.filter.dropped_messages",
     # Counter: how many messages are dropped due to errors producing to the dlq
     "arroyo.consumer.dlq.dropped_messages",
+    # Counter: Number of times consumer commit succeeds or fails
+    "arroyo.consumer.commit_status",
     # Gauge: Current length of the DLQ buffer deque
     "arroyo.consumer.dlq_buffer.len",
     # Counter: Number of times the DLQ buffer size has been exceeded, causing messages to be dropped
@@ -125,7 +127,7 @@ MetricName = Literal[
     # Time: Average round-trip time to brokers from librdkafka statistics.
     # Tagged by broker_id.
     "arroyo.producer.librdkafka.avg_rtt",
-    # Counter: Number of times the produce strategy failed to produce a message
+    # Counter: Number of times the produce strategy succeeds or fails
     "arroyo.producer.produce_status",
     # Gauge: Producer message count metric from librdkafka statistics
     # Tagged by producer_name

{sentry_arroyo-2.32.2.dist-info → sentry_arroyo-2.32.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sentry-arroyo
-Version: 2.32.2
+Version: 2.32.3
 Summary: Arroyo is a Python library for working with streaming data.
 Home-page: https://github.com/getsentry/arroyo
 Author: Sentry

{sentry_arroyo-2.32.2.dist-info → sentry_arroyo-2.32.3.dist-info}/RECORD

@@ -9,7 +9,7 @@ arroyo/backends/abstract.py,sha256=Wy9xhE1dtFiumG8Cz3JhksJ0rF74uJWZWq10UO1rxOI,9
 arroyo/backends/kafka/__init__.py,sha256=xgf-AqHbQkJsh73YokO2uoyyHfZf8XwUp6BULtM8stI,445
 arroyo/backends/kafka/commit.py,sha256=LPsjvX5PPXR62DT6sa5GuSF78qk9F_L--Fz4kw7-m-s,3060
 arroyo/backends/kafka/configuration.py,sha256=zB54w7qsyVeMVkH5MpV6F8ztXfEzIXrex6aKYX-GcqA,9141
-arroyo/backends/kafka/consumer.py,sha256=zZ2ZoDaurLDBN9l9QR0fFWL16RJcf0D8Apaa3aff22k,33534
+arroyo/backends/kafka/consumer.py,sha256=dzypkibGGieLg819TOoxtbVpVaxKYF9dazorQv5FxPg,34688
 arroyo/backends/local/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 arroyo/backends/local/backend.py,sha256=hUXdCV6B5e7s4mjFC6HnIuUhjENU2tNZt5vuEOJmGZQ,13888
 arroyo/backends/local/storages/__init__.py,sha256=AGYujdAAcn3osoj9jq84IzTywYbkIDv9wRg2rLhLXeg,104
@@ -37,8 +37,8 @@ arroyo/utils/clock.py,sha256=r2EMO4nL5qIb1xnAd1sTAk2yK1UltyUi04lk5BqWKIc,944
 arroyo/utils/codecs.py,sha256=x-8SJK0GLTOH4c_k24K97JPjBckxyQJcSpgoEViGUy0,541
 arroyo/utils/concurrent.py,sha256=dbdPinjqmxCQ7izUGFNbGjB3OxfSIO01bnCSTANaVOE,1187
 arroyo/utils/logging.py,sha256=Y1PnhYcI9XNNEK0H13Ct2xKLr2Niuw0dxayc6sWnui8,606
-arroyo/utils/metricDefs.json,sha256=xsDxmGMLo4nhNqMdVNefOEgEa0zLFQFIS7xQwonb8pI,13674
-arroyo/utils/metric_defs.py,sha256=y36K1VmBj5ZI2wUcQDZiM6g0H56j8pnBjBAxqhG2Ns4,9634
+arroyo/utils/metricDefs.json,sha256=_-l3jg85aC5FcjlJ5ISX1YMhGaLvjDBXiQg6QEvsOlQ,13827
+arroyo/utils/metric_defs.py,sha256=aaiTe1cCTRl6DJocHFd7vZ9Vd7FNZye9a7mq8HhR1QQ,9726
 arroyo/utils/metrics.py,sha256=kcyUR5cacoPMoU80RHSUhTMNzEcMBDpTXzcyW7yWZBk,3308
 arroyo/utils/profiler.py,sha256=aiYy2RRPX_IiDIO7AnFM3hARaHCctS3rqUS5nrHXbSg,2452
 arroyo/utils/retries.py,sha256=4MRhHUR7da9x1ytlo7YETo8S9HEebXmPF2-mKP4xYz0,3445
@@ -46,12 +46,13 @@ examples/transform_and_produce/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5
 examples/transform_and_produce/batched.py,sha256=st2R6qTneAtV0JFbKP30Ti3sJDYj8Jkbmta9JckKdZU,2636
 examples/transform_and_produce/script.py,sha256=8kSMIjQNqGYEVyE0PvrfJh-a_UYCrJSstTp_De7kyyg,2306
 examples/transform_and_produce/simple.py,sha256=H7xqxItjl4tx34wVW5dy6mB9G39QucAtxkJSBzVmjgA,1637
-sentry_arroyo-2.32.2.dist-info/licenses/LICENSE,sha256=0Ng3MFdEcnz0sVD1XvGBBzbavvNp_7OAM5yVObB46jU,10829
+sentry_arroyo-2.32.3.dist-info/licenses/LICENSE,sha256=0Ng3MFdEcnz0sVD1XvGBBzbavvNp_7OAM5yVObB46jU,10829
 tests/backends/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 tests/backends/mixins.py,sha256=sfNyE0VTeiD3GHOnBYl-9urvPuURI2G1BWke0cz7Dvc,20445
 tests/backends/test_commit.py,sha256=iTHfK1qsBxim0XwxgMvNNSMqDUMEHoYkYBDcgxGBFbs,831
 tests/backends/test_confluent_producer.py,sha256=KWqgvjDvqAdd0HxngdWKsUJaV7Hl1L5vAVQhBYlHeHU,3146
 tests/backends/test_kafka.py,sha256=wBFCKEHoP6h0uG1bgDuzk84IZmrV_UVOFCrtbxztmJg,15506
+tests/backends/test_kafka_commit_callback.py,sha256=svpY3T7FvhXvz2jp729e60LLn8MyFa8w88FQ1Y0Ursw,1987
 tests/backends/test_kafka_producer.py,sha256=LpwkqnstcCDxemlKZ0FpzNKrP-1UuXXY15P7P-spjhE,3912
 tests/backends/test_local.py,sha256=Mfd4DFuWVSVtl1GomQ6TIoWuJNcAliKqKU0BShPlEMY,3363
 tests/processing/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -74,7 +75,7 @@ tests/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 tests/utils/test_concurrent.py,sha256=Gwdzym2UZ1HO3rhOSGmzxImWcLFygY8P7MXHT3Q0xTE,455
 tests/utils/test_metrics.py,sha256=bI0EtGgPokMQyEqX58i0-8zvLfxRP2nWaWr2wLMaJ_o,917
 tests/utils/test_retries.py,sha256=AxJLkXWeL9AjHv_p1n0pe8CXXJp24ZQIuYBHfNcmiz4,3075
-sentry_arroyo-2.32.2.dist-info/METADATA,sha256=4VH3vZ14YfL1gCnBIpnf_1EnFCykgRXj6LcrVEsd-sc,2208
-sentry_arroyo-2.32.2.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-sentry_arroyo-2.32.2.dist-info/top_level.txt,sha256=DVdMZKysL_iIxm5aY0sYgZtP5ZXMg9YBaBmGQHVmDXA,22
-sentry_arroyo-2.32.2.dist-info/RECORD,,
+sentry_arroyo-2.32.3.dist-info/METADATA,sha256=EfJLX_0UhKzQRBftakhcDvPKZOqlR-JS17uEN-hNV4w,2208
+sentry_arroyo-2.32.3.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+sentry_arroyo-2.32.3.dist-info/top_level.txt,sha256=DVdMZKysL_iIxm5aY0sYgZtP5ZXMg9YBaBmGQHVmDXA,22
+sentry_arroyo-2.32.3.dist-info/RECORD,,

tests/backends/test_kafka_commit_callback.py

@@ -0,0 +1,69 @@
+from confluent_kafka import KafkaError, KafkaException
+from confluent_kafka import TopicPartition as ConfluentTopicPartition
+
+from arroyo.backends.kafka import KafkaConsumer
+from arroyo.backends.kafka.configuration import build_kafka_configuration
+from tests.metrics import Increment, TestingMetricsBackend
+
+
+def test_commit_callback_success_metric() -> None:
+    configuration = build_kafka_configuration(
+        {
+            "bootstrap.servers": "localhost:9092",
+            "group.id": "test-group",
+            "arroyo.enable.auto.commit": True,
+        }
+    )
+
+    consumer = KafkaConsumer(configuration)
+
+    TestingMetricsBackend.calls.clear()
+
+    partitions = [
+        ConfluentTopicPartition("test-topic", 0, 10),
+        ConfluentTopicPartition("test-topic", 1, 20),
+    ]
+
+    consumer._KafkaConsumer__on_commit_callback(None, partitions)  # type: ignore[attr-defined]
+
+    assert (
+        Increment(
+            "arroyo.consumer.commit_status",
+            1,
+            {"group_id": "test-group", "status": "success"},
+        )
+        in TestingMetricsBackend.calls
+    )
+
+
+def test_commit_callback_error_metric() -> None:
+    """Test that failed commits emit a metric with status=error"""
+    configuration = build_kafka_configuration(
+        {
+            "bootstrap.servers": "localhost:9092",
+            "group.id": "test-group",
+            "arroyo.enable.auto.commit": True,
+        }
+    )
+
+    consumer = KafkaConsumer(configuration)
+
+    TestingMetricsBackend.calls.clear()
+
+    kafka_error = KafkaError(1, "Test commit error")
+    error = KafkaException(kafka_error)
+
+    partitions = [
+        ConfluentTopicPartition("test-topic", 0, 10),
+    ]
+
+    consumer._KafkaConsumer__on_commit_callback(error, partitions)  # type: ignore[attr-defined]
+
+    assert (
+        Increment(
+            "arroyo.consumer.commit_status",
+            1,
+            {"group_id": "test-group", "status": "error"},
+        )
+        in TestingMetricsBackend.calls
+    )