[Do not Merge] POC for ACO#13158
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request implements a background metadata fetching mechanism to enrich OpenTelemetry spans with bucket-specific resource IDs and locations. It introduces a BucketMetadataCache and decorators for Span and SpanBuilder to handle attribute extraction and application across both gRPC and HTTP transports. Feedback highlights several critical areas for improvement: the use of fragile reflection to access internal gRPC channels, the risks associated with an unbounded thread pool for background fetches, and the omission of certain SpanBuilder method overrides which could bypass the metadata logic. Additionally, it is recommended to improve error handling for background tasks and replace brittle sleep statements in tests with robust polling.
| } catch (Exception e) { | ||
| } |
There was a problem hiding this comment.
Exceptions during the background metadata fetch are swallowed without logging. This makes it difficult to diagnose why bucket attributes might be missing or stuck on default values. At a minimum, these exceptions should be logged.
| storage.getOptions().toBuilder().setOpenTelemetry(openTelemetrySdk).build(); | ||
| try (Storage storage = storageOptions.getService()) { | ||
| storage.create(BlobInfo.newBuilder(bucket, generator.randomObjectName()).build()); | ||
| Thread.sleep(800); |
There was a problem hiding this comment.
Using Thread.sleep() in integration tests is brittle and can lead to flaky tests or unnecessary delays in the CI pipeline. It is better to use a polling mechanism like Awaitility to wait for the background metadata fetch to complete and the cache to be updated.
…flection-based gRPC channel extraction
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request introduces a bucket metadata caching mechanism to enrich OpenTelemetry spans with resource attributes, specifically 'gcp.resource.destination.id' and 'gcp.resource.destination.location'. It adds new classes like AcoSpan, AcoSpanBuilder, and BucketMetadataCache, and implements the GetStorageLayout RPC for both gRPC and HTTP transports. Review feedback identifies several areas for improvement: the cache invalidation logic for 404 errors is currently too aggressive and may trigger redundant fetches; the use of an unbounded cached thread pool should be replaced with a bounded one for better resource management; and exceptions during background fetches are being swallowed without logging. Additionally, it is recommended to make the hardcoded cache capacity configurable and to use a more graceful shutdown sequence for the executor service.
| if (exception instanceof StorageException | ||
| && parent != null | ||
| && parent.delegate instanceof StorageInternal) { | ||
| StorageException se = (StorageException) exception; | ||
| if (se.getCode() == 404) { | ||
| ((StorageInternal) parent.delegate).getBucketMetadataCache().remove(bucketName); | ||
| } | ||
| } |
There was a problem hiding this comment.
The logic to remove the bucket from the metadata cache on any 404 error is too aggressive. A 404 status code can be returned for object-level operations (e.g., getObject) when the object is missing but the bucket still exists. This results in unnecessary cache invalidation and redundant background fetches. Consider checking if the error specifically indicates that the bucket itself is not found before invalidating the cache.
nidhiii-27
added
the
kokoro:force-run
yoshi-kokoro
removed
the
kokoro:force-run
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request introduces an asynchronous bucket metadata enrichment feature for OpenTelemetry spans in the Google Cloud Storage client. It adds AcoSpan, AcoSpanBuilder, and BucketMetadataCache to cache and inject bucket location and resource ID attributes into spans. The review identified several opportunities for improvement, including refactoring duplicated exception handling in AcoSpan, using a bounded queue for the cache executor to prevent memory exhaustion, replacing the synchronized keyword with explicit locks for better performance, fixing naming conventions, reducing the shutdown timeout for the cache executor, and defining constants for location types.
| try { | ||
| bucketMetadataCache.clear(); | ||
| cacheExecutor.shutdownNow(); | ||
| cacheExecutor.awaitTermination(5, TimeUnit.MINUTES); |
There was a problem hiding this comment.
Waiting for 5 minutes for the cache executor to terminate is excessive for non-critical telemetry enrichment tasks. A shorter timeout (e.g., 10 seconds) would be more appropriate to avoid delaying application shutdown.
| cacheExecutor.awaitTermination(5, TimeUnit.MINUTES); | |
| cacheExecutor.awaitTermination(10, TimeUnit.SECONDS); |
2 participants
No description provided.