fix(graphcache): Prevent refetch cycles when handling cache misses

[kitten]

NOTE: This fix isn't fully verified yet and the code needs to be verified to not break any existing behaviour.

Summary

See prior approach in #2731.

When an operation is read from the cache, if we encounter a cache miss then a network request is sent, which subsequently updates the cache.

However, suppose an operation has several dependencies on the cache-missed operation. If any of the dependent operations then also experience a cache-miss, this will then subsequently retrigger the original operation as well. This is because any dependencies between operations will always be bi-directional by definition.

The easiest approach to solve this issue is to prevent direct cycles, meaning, if an operation triggers another, then we disallow the subsequent operation to trigger the first again, hence breaking infinite cycles.

As per the note at the top of this PR, while this is the minimum solution, we haven't verified two things, due to this being hard for us to test for currently:

• Are cache-miss cycles always transitive? In other words, can a cycle take a longer path than two operations?
• Are there any valid cases where two operations are now unable to updated in sequence? This must at all costs be prevented. There could be sequences of two operations updating in sequence that must then re-trigger a cache read for the first

At least the second statement above is guaranteed and verified by only applying this restriction to cacheMissOps$, i.e. the stream of operations after it has tried to read from the cache, experienced a cache miss, and are about to be forwarded.

Set of changes

• Introduce double-set tracking to prevent operation re-execution cycles

[changeset-bot]

🦋 Changeset detected

Latest commit: f5a38e0

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@urql/exchange-graphcache	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR