fix(ui): perf improvements

[psychedelicious]

What type of PR is this? (check all applicable)

• Refactor
• Feature
• Bug Fix
• Optimization
• Documentation Update
• Community Node Submission

Description

Investigation of performance issues indicates the main factor is the change of reselect's default memoization function from lruMemoize to weakMapMemoize. This changed in the RTK v2 release (v2 depends on the reselect v5, which includes this change).

lruMemoize provides a less sophisticated memoize strategy, with a configurable, but finite, cache size. We've never changed it from the default cache size of 1. Instead, we wrap parameterized selectors in useMemo and rely on closures to provide the selector parameters. The cache size never grows.

weakMapMemoize provides an effectively unbounded cache size out of the box. When selectors are reused with different parameters, the cache grows.

For our particular use patterns, with many parameterized selectors, this substantially increases memory allocations and causes a frequent major GCs. This is my best estimation of the situation and there may be more nuance of course.

Previously (between v3.4.0post2/v3.5.1 and now), I had configured most of the app to use lruMemoize, but there were two missing spots:

• createEntityAdapter's getSelectors() defaults to weakMapMemoize, changed to lruMemoize in this PR.
• RTK Query uses weakMapMemoize internally, but there's no way to change. Pending a change upstream to allow customizing the memoize function, I've directly patched the builds so that lruMemoize is always used.

In my profiling, this has smoothed out the remaining bumps, but it's possible that other changes introduced after v3.4.0post2 were also contributing and we still have a few places to fix up.

Related Tickets & Documents

• Related https://discord.com/channels/1020123559063990373/1187836949524906014

QA Instructions, Screenshots, Recordings

Test canvas and workflow editor. Should be smooth. If there are still performance issues I'll need to continue investigating.

Merge Plan

This PR can be merged when approved.

[hipsterusername]

@psychedelicious @maryhipp - only thing blocking atm is lint checks are failing.

[psychedelicious]

I've also reviewed every single redux selector in the app and optimized all but a couple of them.

The useNextPrevImage hook needs to be refactored to avoid using a state => state input selector, and there's a canvas selector I ran out of steam on. Neither are causing issues though (I commented them out and nothing changed), so I will tackle those in the future.

Our redux implementation is better now than it ever has been. Sweet!