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

[BUG] Fix HNSW resize conditions to exclude deleted items #3424

Merged
merged 1 commit into from
Jan 7, 2025

Conversation

eculver
Copy link
Contributor

@eculver eculver commented Jan 7, 2025

Description of changes

We were seeing errors in compaction that looked like:

WriteSegments(ApplyMaterializatedLogsError(HnswIndex(FFIException(\"The number of elements exceeds the specified limit\"))))

This is because the resize logic uses len which removes deleted elements which then causes the conditional for resizing to be skipped. This adds a new method len_with_deleted to return the length of the index including the elements that have been deleted so that the resize is not skipped. See:

let mut index = self.index.inner.upgradable_read();
let index_len = index.len();
let index_capacity = index.capacity();
if index_len + 1 > index_capacity {
index.with_upgraded(|index| {
// Bump allocation by 2x
index
.resize(index_capacity * 2)
.map(|_| ApplyMaterializedLogError::Allocation)
})?;
}

Test plan

A test was added. It will fail when using the existing len method, but passes when using the newly added len_with_deleted method.

  • Tests pass locally with pytest for python, yarn test for js, cargo test for rust

Documentation Changes

N/A

Thanks to @sanketkedia for helping root cause this.

Fixes https://github.com/chroma-core/hosted-chroma/issues/855

@eculver eculver requested review from HammadB and sanketkedia January 7, 2025 22:14
Copy link

github-actions bot commented Jan 7, 2025

Reviewer Checklist

Please leverage this checklist to ensure your code review is thorough before approving

Testing, Bugs, Errors, Logs, Documentation

  • Can you think of any use case in which the code does not behave as intended? Have they been tested?
  • Can you think of any inputs or external events that could break the code? Is user input validated and safe? Have they been tested?
  • If appropriate, are there adequate property based tests?
  • If appropriate, are there adequate unit tests?
  • Should any logging, debugging, tracing information be added or removed?
  • Are error messages user-friendly?
  • Have all documentation changes needed been made?
  • Have all non-obvious changes been commented?

System Compatibility

  • Are there any potential impacts on other parts of the system or backward compatibility?
  • Does this change intersect with any items on our roadmap, and if so, is there a plan for fitting them together?

Quality

  • Is this code of a unexpectedly high quality (Readability, Modularity, Intuitiveness)

@eculver
Copy link
Contributor Author

eculver commented Jan 7, 2025

Going to merge since Sanket has already looked over it all.

@eculver eculver merged commit 6c614ec into main Jan 7, 2025
79 checks passed
@eculver eculver deleted the eculver/hnsw-resize-bug branch January 7, 2025 22:52
chroma-droid pushed a commit that referenced this pull request Jan 7, 2025
## Description of changes

We were seeing errors in compaction that looked like:

```
WriteSegments(ApplyMaterializatedLogsError(HnswIndex(FFIException(\"The number of elements exceeds the specified limit\"))))
```
This is because the resize logic uses `len` which removes deleted
elements which then causes the conditional for resizing to be skipped.
This adds a new method `len_with_deleted` to return the length of the
index including the elements that have been deleted so that the resize
is not skipped. See:
https://github.com/chroma-core/chroma/blob/27564afa0e82d184a0fa853a7384d00861d8e9dc/rust/worker/src/segment/distributed_hnsw_segment.rs#L212-L222

## Test plan
A test was added. It will fail when using the existing `len` method, but
passes when using the newly added `len_with_deleted` method.

- [x] Tests pass locally with `pytest` for python, `yarn test` for js,
`cargo test` for rust

## Documentation Changes
N/A

Thanks to @sanketkedia for helping root cause this. 

Fixes chroma-core/hosted-chroma#855
chroma-droid pushed a commit that referenced this pull request Jan 7, 2025
## Description of changes

We were seeing errors in compaction that looked like:

```
WriteSegments(ApplyMaterializatedLogsError(HnswIndex(FFIException(\"The number of elements exceeds the specified limit\"))))
```
This is because the resize logic uses `len` which removes deleted
elements which then causes the conditional for resizing to be skipped.
This adds a new method `len_with_deleted` to return the length of the
index including the elements that have been deleted so that the resize
is not skipped. See:
https://github.com/chroma-core/chroma/blob/27564afa0e82d184a0fa853a7384d00861d8e9dc/rust/worker/src/segment/distributed_hnsw_segment.rs#L212-L222

## Test plan
A test was added. It will fail when using the existing `len` method, but
passes when using the newly added `len_with_deleted` method.

- [x] Tests pass locally with `pytest` for python, `yarn test` for js,
`cargo test` for rust

## Documentation Changes
N/A

Thanks to @sanketkedia for helping root cause this. 

Fixes chroma-core/hosted-chroma#855
eculver added a commit that referenced this pull request Jan 7, 2025
This PR cherry-picks the commit 6c614ec
onto release/2025-01-03. If there are unresolved conflicts, please
resolve them manually.

Co-authored-by: Evan Culver <[email protected]>
eculver added a commit that referenced this pull request Jan 8, 2025
## Description of changes

This is a follow-up from #3424 - where I had added code to fix the bug
in compaction but then forgot to replace the method at the actual
callsite with the new method 🤦🏼 .

This updates the problem code to use `len_with_deleted` instead of just
`len` so that the resize actually gets triggered.

Some additional context: we decided not to change the behavior of the
existing `len` method in case other callers were depending on existing
behavior. In other words, we didn't want to introduce accidentally
introduce bugs that were relying on the "buggy" code.
chroma-droid pushed a commit that referenced this pull request Jan 8, 2025
## Description of changes

This is a follow-up from #3424 - where I had added code to fix the bug
in compaction but then forgot to replace the method at the actual
callsite with the new method 🤦🏼 .

This updates the problem code to use `len_with_deleted` instead of just
`len` so that the resize actually gets triggered.

Some additional context: we decided not to change the behavior of the
existing `len` method in case other callers were depending on existing
behavior. In other words, we didn't want to introduce accidentally
introduce bugs that were relying on the "buggy" code.
chroma-droid pushed a commit that referenced this pull request Jan 8, 2025
## Description of changes

This is a follow-up from #3424 - where I had added code to fix the bug
in compaction but then forgot to replace the method at the actual
callsite with the new method 🤦🏼 .

This updates the problem code to use `len_with_deleted` instead of just
`len` so that the resize actually gets triggered.

Some additional context: we decided not to change the behavior of the
existing `len` method in case other callers were depending on existing
behavior. In other words, we didn't want to introduce accidentally
introduce bugs that were relying on the "buggy" code.
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.

1 participant