fix(test): handle NDJSON chunking in log_stream tests#4358
fix(test): handle NDJSON chunking in log_stream tests#4358unleashed merged 1 commit intolinkerd:mainfrom
Conversation
The log endpoint streams NDJSON (newline-delimited JSON) generated by `tracing-subscriber`. However, the log stream tests assume each HTTP body chunk contains a complete JSON object, which isn't necessarily the case. HTTP streaming doesn't guarantee chunk boundaries align with JSON boundaries because TCP can split data anywhere. This causes occasional test failures when chunks split mid-JSON-object in my local machine. Fix by buffering incoming chunks and extracting complete lines (by newline delimiter) before parsing as JSON. We could arguably set the content type header for this endpoint to `application/x-ndjson` to be fully compliant with the NDJSON specs, but that that may be risking a breaking change for little value. Signed-off-by: Alejandro Martinez Ruiz <amr@buoyant.io>
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #4358 +/- ##
==========================================
+ Coverage 67.68% 71.53% +3.84%
==========================================
Files 332 394 +62
Lines 15158 18546 +3388
==========================================
+ Hits 10259 13266 +3007
- Misses 4899 5280 +381 see 299 files with indirect coverage changes Continue to review full report in Codecov by Sentry.
🚀 New features to boost your workflow:
|
|
circling back on this, this seems like a change we can merge now even if it was spurred by some other ongoing work. |
|
@cratelyn yes, it was extracted and meant to be an independent fix to be merged separately. |
|
Can we have this merged? |
oh! conventionally the author merges the pull request, are you able to do that? |
Yes, thanks! |
2 participants
The log endpoint streams NDJSON (newline-delimited JSON) generated by
tracing-subscriber. However, the log stream tests assume each HTTP body chunk contains a complete JSON object, which isn't necessarily the case.HTTP streaming doesn't guarantee chunk boundaries align with JSON boundaries because TCP can split data anywhere. This causes occasional test failures when chunks split mid-JSON-object in my local machine.
Fix by buffering incoming chunks and extracting complete lines (by newline delimiter) before parsing as JSON.
We could arguably set the content type header for this endpoint to
application/x-ndjsonto be fully compliant with the NDJSON specs, but that that may be risking a breaking change for little value.