At the end we had an OOM in the compute node and the error log said:
2024-06-05 11:36:07.198 [12358.306313] Out of memory: Killed process 168 (sql_exporter) total-vm:1268868kB, anon-rss:12712kB, file-rss:8672kB, shmem-rss:0kB, UID:65534 pgtables:200kB oom_score_adj:0
1.2 GB RAM for a component just doing some SQL queries to collect metrics seems expensive.
Probably we have a memory leak in the sql exporter or it can not get rid of collected metrics quickly enough.
Steps to reproduce
I ran multiple tests (pgvector indexing, index from subselect) etc., see https://neondb.slack.com/archives/C0732L0A4AH/p1717575836689089?thread_ts=1717499447.241659&cid=C0732L0A4AH
At the end we had an OOM in the compute node and the error log said:
2024-06-05 11:36:07.198 [12358.306313] Out of memory: Killed process 168 (sql_exporter) total-vm:1268868kB, anon-rss:12712kB, file-rss:8672kB, shmem-rss:0kB, UID:65534 pgtables:200kB oom_score_adj:0
1.2 GB RAM for a component just doing some SQL queries to collect metrics seems expensive. Probably we have a memory leak in the sql exporter or it can not get rid of collected metrics quickly enough.
Expected result
Actual result
Environment
Logs, links
https://neonprod.grafana.net/goto/gT_JXPyIR?orgId=1