Details
-
Bug
-
Resolution: Unresolved
-
Critical
-
1.6.2.Final
-
None
-
False
-
False
Description
Running Debezium with:
- debezium/connect:1.6
- Running in AWS ECS
- PostgreSQL 12.7 on AWS RDS (db.m5.4xlarge / 16 vCPU / 64 GB RAM)
- Kafka 2.7.0 via AWS MSK
Within a few minutes of running, the load on RDS increases such that our DB becomes unusable and fails with a high number of errors (like pq: the database system is in recovery mode), forcing us to stop it to reduce load.
RDS Perf insights lists the top wait offender as:
SELECT COUNT(1) FROM pg_publication WHERE pubname = 'debezium_pub'
This has been consistent with the past 2 versions we've tried, 1.5 and 1.6.
Some logs, connector config, and RDS perf info is included as attachments.
How can we reduce the perf impact?