Send queries to fetch schema cache to replicas
What does this MR do and why?
We have a CPU load issue when pods are created. Many pods all request schema information for tables either on boot, or on the first requests. All of these queries go to the primary, and are then cached in the schema cache.
It turns out that the schema queries always use the primary connection pool. That is the default behaviour, see ConnectionProxy#method_missing.
To alleviate this load issue, we configure schema_cache to use replica connection pools when possible.
Related issue: #398101 (closed)
Is it safe to use replicas to query schema information?:
Well, we happily rely on replicas for other queries - which means we implicitly trust the schema on replicas already :)
Screenshots or screen recordings
| Before | After |
|---|---|
 |
 |
How to set up and validate locally
-
Set up DB load balancing in
config/database.yml.load_balancing: hosts: - <same host as main:> - <same host as main:> -
gdk restart rails-web -
Navigate to
127.0.0.1:3000/ -
Open the performance bar. Observe whether the
SELECT a.attname,...queries is sent to primary, or replica.
MR acceptance checklist
This checklist encourages us to confirm any changes have been analyzed to reduce risks in quality, performance, reliability, security, and maintainability.
-
I have evaluated the MR acceptance checklist for this MR.