We have an odd scenario on one of our OLTP systems, which behaves the same way on a streamer, of a 700-1000ms planning time for a query like this:
SELECT *
FROM table1
WHERE source_id IN (SELECT id FROM table2 WHERE customer_id = $1);
The actual execution time is sub-ms.
We initially thought maybe catalog bloat? But we were able to reindex all of the highly churned catalog tables, and I even did VACUUM FULL on pg_attribute and pg_statistic, to no avail.
There are no custom settings for pg_attribute for the given tables either.
Interestingly, the problem goes away on a SAN snapshot of the target system.
Any ideas of what else we could try? A PL function that caches the query plan works, but that is just a workaround.
more times I seen similar issue based on bloated indexes on table - pg in planning time detect min max from possible indexes