Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix fetch group processing handling of dependencies #2622

Merged
merged 3 commits into from
Jun 14, 2023

Conversation

pcmanus
Copy link
Contributor

@pcmanus pcmanus commented Jun 9, 2023

The code that process FetchGroup in dependency order has a bug that, for some types of dependencies, can finish with some group not handled, even though they could and should have been.

In practice, this shows itself by failing the post-processing assertion that double-check that no groups are left unhandled.

The error message when this triggered looks like Root groups X should have no remaining groups unhandled (...) (where X is a dump of the fetch groups involved and so this can be large and hard to parse, but basically the assertion error starts with "Root groups").

@pcmanus pcmanus requested a review from a team as a code owner June 9, 2023 13:19
@changeset-bot
Copy link

changeset-bot bot commented Jun 9, 2023

🦋 Changeset detected

Latest commit: 847e95a

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

This PR includes changesets to release 7 packages
Name Type
@apollo/query-planner Patch
@apollo/gateway Patch
@apollo/federation-internals Patch
@apollo/composition Patch
@apollo/query-graphs Patch
@apollo/subgraph Patch
apollo-federation-integration-testsuite 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

@netlify
Copy link

netlify bot commented Jun 9, 2023

👷 Deploy request for apollo-federation-docs pending review.

Visit the deploys page to approve it

Name Link
🔨 Latest commit 847e95a

@pcmanus pcmanus force-pushed the fix-fetch-group-processing branch from 87292ae to 4b4e0a4 Compare June 9, 2023 13:20
@codesandbox-ci
Copy link

codesandbox-ci bot commented Jun 9, 2023

This pull request is automatically built and testable in CodeSandbox.

To see build info of the built libraries, click here or the icon next to each commit SHA.

@pcmanus pcmanus force-pushed the fix-fetch-group-processing branch from 4b4e0a4 to 359f410 Compare June 9, 2023 14:30
for (const [g, edges] of r1) {
const newEdges = this.mergeRemaingsAndRemoveIfFound(g, edges, r2);
if (newEdges.length == 0) {
toHandle.push(g);
readyToHandle.push(g);
} else {
unhandled.push([g, newEdges])
}
}
unhandled.push(...r2);
Copy link
Contributor

@sachindshinde sachindshinde Jun 10, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Bit confused here, is there a reason this

unhandled.push(...r2);

isn't

for (const [g, edges] of r2) {
  if (edges.length == 0) {
    readyToHandle.push(g);
  } else {
    unhandled.push([g, edges])
  }
}

instead?

It would seem that if processGroup() returned an unhandled where some of its groups had no edges, then those groups are "ready to handle". Or at least with the current code, since processGroups() loops over multiple/n groups, the last group's zero-edge unhandleds won't be considered "ready to handle", while the previous n-1 groups' will.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You're right this was fishy here. The intend though was that UnhandledGroups would never contain group whose "edges" are empty (both methods that modify it make sure to remove groups as soon as their edges are all empty). But the "bug" was indeed that processGroup could "generate" an UnhandledGroups with "zero-edges" groups and that's where the code was confused. I've addressed in the small refactor I did (it now does genuinely avoid created zero-edge entries in UnhandledGroups, well, at least unless I brain-farted again obviously).

const [canHandle, newRemaining] = this.mergeRemainings(remainingNext, unhandled);
remainingNext = newRemaining;
nextGroups = canHandle.concat(next);
remainingNext = this.mergeRemainings(remainingNext, unhandled, next);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm confused about the purpose of this call to mergeRemainings().

My understanding is:

  1. processGroups() is passed remainingNext.
  2. Along the way, processGroups() repeatedly merges its remainingNext with the unhandleds it receives from processGroup() (moving a group to next whenever it gets down to zero edges). It doesn't modify the original remainingNext, and makes copies instead.
  3. processGroups() returns the new remainingNext along with next.
  4. Then on this line, we merge the old remainingNext with the new remainingNext.

It seems like this merge might add back groups to remainingNext that were transferred over to next in processGroups(), which feels like a bug.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You're right, this was unnecessary (and probably dangerous).

The overall code around this was admittedly a bit confusing, so I did a small refactor to hopefully clear things up a bit. It should be somewhat equivalent to what we have before (save the real bug you rightfully pointed in the previous comment) but I think encapsulate things a tad better/is more explicit. It's probably not perfect, I think this processing code could maybe be cleaned up som more, but I'll leave that for later (started a slightly bigger refactor and messed it up, so figured I should leave that separate from this bug-fixing).

readyToHandle: FetchGroup[],
): UnhandledGroups {
const unhandled: UnhandledGroups = [];
for (const current of currentUnhandled) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

minor nit: To be consistent with mergeRemainings()/make code more clear, could you change current to e.g. [g, edges]? (edges is more clear than current[1])

@pcmanus pcmanus force-pushed the fix-fetch-group-processing branch from f975062 to ba2af08 Compare June 13, 2023 08:17
@pcmanus pcmanus requested a review from sachindshinde June 13, 2023 08:17
pcmanus added 2 commits June 13, 2023 17:06
The code that process `FetchGroup` in dependency order has a bug that,
for some kind of dependencies, can finish with some group not handled,
even though they could and should have been.

In practice, this shows itself by failing the post-processing assertion
that double-check that no groups are left unhandled.
@pcmanus pcmanus force-pushed the fix-fetch-group-processing branch from ba2af08 to 3c5719c Compare June 13, 2023 15:07
Copy link
Contributor

@sachindshinde sachindshinde left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for making the changes! There's some comments below, but nothing really blocking this PR from merging.

const groupIsOnlyParentOfAllChildren = children.every(g => g.parents().length === 1);
if (groupIsOnlyParentOfAllChildren) {
const state = ProcessingState.forChildrenOfProcessedGroup(group, children);
if (state.next.length > 0) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note that there's a logic change here.

Before this change, this if block executes only if all children have exactly one parent. After this change, this if block executes if at least one child has exactly one parent.

This might be fine/intentional, given that you've also swapped out children down below for state.next. Though in that case, you should update the comment below to say e.g. ready children instead of just children.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, sorry for not pointing out that change myself. That was intentional: the old code was more restrictive than made sense and I think I made that initially out of convenience, but it wasn't hard to "fix" that as I refactored, so did it. I did updated the comment though, good point.

@@ -2473,86 +2582,55 @@ class FetchDependencyGraph {
deferredGroups: allDeferredGroups,
} = this.processRootMainGroups({
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note that because processRootMainGroups() doesn't take the old state/return a new state, presumably when it eventually calls updateForProcessedGroups() for the state it internally creates, it won't alter the separate state in the upstream processGroups() call. This was the case before this PR too though, so it's probably fine to punt that to another PR if you'd like. (It might also be the case this was intentional.)

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You know what, you're right. I hesitating changing it and for some reason didn't do it, but there is probably case where that could be a genuine issue, so took the liberty to fix this quickly (it's cleaner if anything anyway).

@pcmanus pcmanus force-pushed the fix-fetch-group-processing branch 3 times, most recently from 7848ca0 to e24b631 Compare June 14, 2023 07:10
@pcmanus pcmanus force-pushed the fix-fetch-group-processing branch from e24b631 to 847e95a Compare June 14, 2023 07:14
@pcmanus pcmanus merged commit 1293034 into apollographql:main Jun 14, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants