Produce is slow (SQL0038)

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

The produce call exceeded ProducerConfig.SlowProduceThreshold.

duration includes waiting in the producer’s batch queue and writing to the database. For callback methods, it also includes your application’s work to build the message.

Identify where the call stalled

Run during the slowdown:

SELECT pid, application_name,
    statement_timestamp() - query_start AS query_age,
    wait_event_type, wait_event, pg_blocking_pids(pid) AS blocking_pids,
    query
FROM pg_stat_activity
WHERE datname = current_database()
    AND pid <> pg_backend_pid()
    AND state <> 'idle'
ORDER BY query_start;

Result	Action
SQLStreams query has blocking pids	Find those sessions in the results. Have their owning application finish or roll back the blocking transactions.
Partition creation coincides with `SQL0057` or `SQL0033`	Follow step 1 below.
Queries run without blockers, or no relevant query appears	Check connection waiting and application callback work in steps 2–3. Local queue time is not visible in this query.

Fix the recurring delay

If partition creation is on the insert path, read its accompanying error. SQL0057 means the insert must create a covering partition. SQL0033’s error explains why create-ahead failed. Resolve that database error first.

If SQL0056 continues during bursts, reduce producer pressure.
If calls wait without database activity, inspect the application’s pool. From the process that owns pool, read:
```
stats := pool.Stat()
fmt.Println(stats.AcquiredConns(), stats.TotalConns(), stats.MaxConns())
```
If all allowed connections stay acquired, shorten transactions and release connections held by other work before increasing producer concurrency. Each active batch transaction needs a connection.
If the pool has room, check work inside ProduceFunc or an outer transaction. Inspect database and service calls made by the callback that builds the message. Move work that does not need the transaction outside it. If another transaction holds the same compacted message key, finish that transaction promptly to release the waiting produce call.

Producer logs — Timing and partition warning fields.
Producer batching settings — Batch size and flush timing.
Produce within a transaction — Transaction ownership and timing boundaries.

Identify where the call stalled

Fix the recurring delay

Related topics