Indexing one kafka source leads to postgresql pool timed out errors after some time

I saw this error several times after running an indexing pipeline during at least 30-45 minutes and with 140 partitions (indexing throughput of 40MB/s).
It seems to happen randomly on any metastore endpoint. After this error, the pipeline restarts normally.

I will continue to investigate on this, I don't understand yet why we have this.

My node config

#
# -------------------------------- Indexer settings ---------------------------------
#
indexer:
  split_store_max_num_bytes: 300G
  split_store_max_num_splits: 10000

The error message

2022-10-13T21:47:01.191Z ERROR quickwit_actors::spawn_builder: actor-failure cause=Failed to publish splits.

Caused by:
    Database error: `pool timed out while waiting for an open connection`. exit_status=Failure(Failed to publish splits.

Caused by:
    Database error: `pool timed out while waiting for an open connection`.)
2022-10-13T21:47:01.195Z  INFO quickwit_actors::spawn_builder: actor-exit actor_id=MergePublisher-lively-SDmB exit_status=Failure(cause=Failed to publish splits.

No source file, test, or entry point is identified. Start by reproducing the indexing pipeline with one Kafka source, 140 partitions, and the shown indexer settings for 30–45 minutes, then trace the metastore PostgreSQL pool timeout during split publication. Done means the cause is identified and the failure is prevented or documented with a regression test.