Memory leak in Aurora Postgres since 3.9.1

[al-kudryavtsev]

Version

3.9.1, 4.5.7

Context

Our app has been on vertx-pg-client 3.9.0 for a while until recently when we migrated to version 4.5.7.

Shortly after the upgrade we noticed our AWS Aurora PG cluster is running out of freeable memory and rebooting. It is directly correlated to the load on the service - during peak times DB can run out of memory (60GB) in 45 minutes. We employed the maxLifetime parameter for the pool, which helps but is not ideal as it causes latency spikes. We prewarm the pool on deployment so all the connections expire at about the same time.

On 3.9.0, there was no issue with DB memory consumption. We rolled back the upgrade to confirm this. On 3.9.1, confirmed the issue is present. This thread has a report of the same issue on 3.9.13 as well as 4.x branch.

Do you have a reproducer?

I don't have a reproducer; this might be AWS/aurora-specific. I can test potential solutions if that can help.

Extra

Pretty much all our queries are preparedQuery + execute or executeBatch. We also use standalone prepare with a cursor for paginated reads. Pool configuration: https://gist.github.com/al-kudryavtsev/3e6eeb3cfd200afc66df5a12932bba25