Improve cache flush handling for large multisite database upgrades

[xParham]

We run a large WP multisite (200 sub-sites) and with recent WP major version upgrades, the upgrade process has caused a significant performance drop while the database upgrade was in progress, causing downtime and 503 server errors. This seems to be mainly due to multiple back-to-back flushing of the object cache, network-wide, while the sites are under traffic.

In wp_upgrade() I see two calls to wp_cache_flush() which would happen for each of the subsite DB upgrades and I was wondering if things can be improved there.

• Could we make the cache flush to be more selective and drop specific groups rather than dropping the whole network cache?
• Could we do only one cache flush at the end of the DB upgrade for all sites instead of dropping the cache for each of the subsites' DB upgrades?
• For object caching implementations that support subsite cache flushing (e.g. OCP) could we maybe utilize that feature?

I have attached some logs and reports from our APM. This was for upgrading from WP 6.3.x to 6.4.x, using Redis with Object Cache Pro, and 4GB of cached items.