plan queries in random order during warm up #3769
Assignees
Comments
This was referenced Sep 6, 2023
Geal
added a commit
that referenced
this issue
Sep 18, 2023
Fix #3769 When using query plan cache warm up, we take the list of most used queries from the in memory cache, and plan them. The list will come in order, from the most used to the least used. If multiple router instances are using cache warm up, and are also configured to use the distributed cache for query plans, then they would all plan the list of queries in the same order, since they will have roughly the same distribution in the in memory cache. By randomizing the list of queries, we increase the chances that other instances can get plans from the distributed cache during warm up instead of calculating them, because other instances might have already done the work.
Merged
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Right now, queries are extracted from the cache and ordered according to their frequency by the in memory cache. That means all the router instances will warm up the same query roughly at the same time. If they are using distributed caching, they are racing to do the same work and cannot share much.
If we reorder the queries randomly (after selecting the most used), then there's a higher chance that each router would plan a different query, stores the plan in Redis, and as warm up progress, they will get more plans directly from Redis instead of recalculating them
The text was updated successfully, but these errors were encountered: