Producer logs

SQL0038  Slow produce
SQL0033  Partition not created ahead
SQL0057  Partition created during insert

These events report production latency and partition creation.

Slow produce (SQL0038)

level=WARN msg="produce exceeded the duration threshold" code=SQL0038 stream={stream} duration={duration} threshold={threshold}

Emitted after a produce call whose elapsed duration exceeds the producer’s positive SlowProduceThreshold, regardless of success or failure. The setting defaults to 0, which disables this warning.

duration measures the total time spent in the produce call:

• Batch queue: time the message waits in the producer’s local queue before its batch is written, when automatic batching applies.
• Database operations: time spent writing the message, including waits for database connections and locks.
• Application callback: for ProduceFunc and ProduceFuncInTx, time spent running the function you supplied to build the message.

For ProduceInTx and ProduceFuncInTx, timing stops when the method returns. Your application still owns the transaction. Work you perform afterward, including committing that transaction, is outside the logged duration.

Configure ProducerConfig.SlowProduceThreshold. For investigation, see Produce is slow.

Partition not created ahead (SQL0033)

level=WARN msg="could not create partition ahead -- the first insert past the boundary will create it" code=SQL0033 stream_id={stream_id} error={error}

The asynchronous attempt to create the next partition failed. A later insert can create the partition when it reaches that boundary.

Read the nested error to distinguish lock contention from other database failures. SQL0018 identifies a partition-creation lock timeout.

Partition created during insert (SQL0057)

level=WARN msg="no partition covers the next message id -- the insert creates it and pays the creation latency; run a consumer for the stream's upkeep or raise PartitionSize" code=SQL0057 stream_id={stream_id} message_id={message_id}

An insert encountered an uncovered message id and entered partition creation. The warning is emitted before creation finishes, so it is not proof that the retry succeeded.

If repeated creation cannot catch up with allocated ids, the Produce call can return ErrPartitionCreationBehind (SQL0056). See Produce is slow for the investigation path.

Suppression

Producer-instance Warn/Error repeats share the one-minute suppression rule. The next emitted repeat can include suppressed_count. Datastore events use their owning logger’s configuration, so suppression is not guaranteed for every partition event.

Related topics

• Register a producer — Duration and batching settings.
• Produce a message — Returns, errors, and cancellation.
• Errors and diagnostics — Structured errors and offline code lookup.
• Produce is slow — Investigate duration warnings and partition creation delays.