Fix reference count bug in partition batcher

[aditya-systems-hub]

---

Summary

This PR fixes an off-by-one error in the partition batcher’s reference counting logic that could cause exporter errors to be silently dropped under specific error conditions.

When a batch is split into multiple requests and MergeSplit() returns an error, the reference counter was initialized with an incorrect value due to a copy-paste mistake. This could lead to the done callback firing too early, before all flush operations completed.

---

Problem Description

In Consume, the number of references (numRefs) is intentionally incremented when mergeSplitErr is non-nil to account for the additional error callback. However, the reference counter was initialized using len(reqList) instead of numRefs.

As a result, the reference count could be lower than the actual number of callbacks that would be invoked.

Buggy behavior (simplified)

numRefs := len(reqList)
if mergeSplitErr != nil {
    numRefs++
}

done = newRefCountDone(done, int64(len(reqList))) // incorrect

This mismatch causes the underlying done callback to be triggered prematurely.

---

Impact

Before this fix

• Exporter errors from final flush operations could be silently lost
• done could be invoked before all export operations completed
• Error aggregation reported back to callers could be incomplete
• In queue-based exporters using waitForResult, callers could observe success even when exports failed
• Silent data loss in production telemetry pipelines was possible

After this fix

• All flush operations are correctly tracked
• Errors from all callbacks are properly aggregated
• done is invoked only after all operations complete
• Completion signaling is consistent and reliable

---

Steps to Reproduce

This issue occurs when all of the following conditions are met:

1. A batch is split into multiple requests (len(reqList) >= 2)
2. MergeSplit() returns a non-nil error
3. The partition batcher initializes the reference counter

In this case, the extra error callback increases the true number of references, but the counter was initialized with a lower value, causing premature completion.

This is an edge case and does not crash or panic, which makes it difficult to detect without careful inspection or targeted testing.

---

Fix

The fix ensures the reference counter is initialized with the correct number of references (numRefs) so that all callbacks are properly accounted for.

Correct behavior

numRefs := len(reqList)
if mergeSplitErr != nil {
    numRefs++
}

done = newRefCountDone(done, int64(numRefs)) // correct

This aligns the logic with the already-correct implementation used earlier in the same file and restores correct lifecycle handling.

---

Why This Is Important

This bug is particularly hard to detect because:

• No crash or panic occurs
• The failure mode is silent
• Most exports still succeed
• The issue only affects specific edge cases involving split batches and merge errors

However, when it does occur, it can lead to silent telemetry data loss and misleading success signals in production systems.

---

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: aditya4044656 / name: aditya kuchekar (6d459ee, 6e48cb4, 8026535)

[aditya-systems-hub]

Hi @dmitryax , @bogdandrutu
I’ve addressed the issue and added a fix for the reference counting bug in the partition batcher. I’ve also included a detailed explanation in the PR description covering the impact, reproduction scenario, and rationale for the change.

When you have time, I’d appreciate a review or any feedback you may have. Please let me know if you’d like additional tests or adjustments.

Thank you for your time and guidance.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.80%. Comparing base (7c31dd5) to head (6d459ee).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #14444      +/-   ##
==========================================
- Coverage   91.81%   91.80%   -0.02%     
==========================================
  Files         677      677              
  Lines       42677    42677              
==========================================
- Hits        39184    39179       -5     
- Misses       2433     2436       +3     
- Partials     1060     1062       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[bogdandrutu]

Please add a changelog entry

[aditya-systems-hub]

Thank you sir @bogdandrutu for pointing this out. I will add an appropriate changelog entry under /.chloggen to document this change.

[otelbot]

Thank you for your contribution @aditya4044656! 🎉 We would like to hear from you about your experience contributing to OpenTelemetry by taking a few minutes to fill out this survey.