Indexing subscription facet pools generates sql query per consumer
- get consumer ids belonging to this pool from candlepin (around 15s)
- loop through the consumer ids and create an entry in SubscriptionFacetPool if it does not exist
- delete bad entries in SubscriptionFacetPool (i.e. delete where facet ids no longer exists)
step 2 generates a query for every consumer id in the pool to verify it exists. In some high volume customer cases that could mean 50K sql queries (all doing a where clause to verify the existence of a subscription facet pool) on every registration. This could also result in edge cases and pain points related errors like
PG::Error: ERROR: insert or update on table \"katello_subscription_facet_pools\" violates foreign key constraint \"katello_sub_facet_pools_sf_id_fk\"\nDETAIL: Key (subscription_facet_id)=(1000) is not present in table \"katello_subscription_facets\".\n: INSERT INTO \"katello_subscription_facet_pools\" (\"pool_id\", \"subscription_facet_id\") VALUES ($1, $2) RETURNING \"id\"
where the slow response causes the registering client to time out and retry with --force causing race conditions.
Fixes #23096 - Improve SubFacetPool indexing perf
This commit intends to improve the performance of the import hosts
method used while indexing Pools. It reduces the number of queries made
on SubscriptionFacetPool to determine what needs to be added or removed
from it. In some high volume consumers this could mean reductions from
50K sql queries to 4 (depending on the number of consumers in a pool).