Repository navigation
[storage] Check the cache behavior with very wide records #7554
Description
It seems that the caching layer refuses to cache in memory very wide records.
This can lead to performance anomalies for relatively small collections with very wide records.
E.g. for a program that uses ARRAY_AGG() - most of the time is spent in the UPSERT node of the aggregate, which never hits in the cache.
Activity
ryzhyk commentedon Oct 20, 2026
Note that refusing to cache maybe the optimal behavior at least in some cases. It's not obvious that forcing such records to be cached is always a good idea.