smengcl opened a new pull request, #11836: URL: https://github.com/apache/ozone/pull/11836
Generated-by: Claude Code (Opus 5.5) ## What changes were proposed in this pull request? `ChunkInputStream.read(byte[], int, int)` and `read(ByteBuffer)` can first copy bytes still buffered from the previous ReadChunk and then issue another ReadChunk for the rest. If that ReadChunk fails, the copied byte count is lost with the exception but the chunk position stays advanced. `BlockInputStream.readWithStrategy` then refreshes the pipeline and token and retries from the advanced position: - `byte[]`: the `ByteArrayReader` offset never advanced, so the retry writes the following bytes over the start of the caller's buffer and reports success. The read silently returns wrong data and the stream ends up that many bytes ahead. - `ByteBuffer`: the data lands in the right place but the count is short, so the read fails with "Inconsistent read" although the retry itself succeeded. Triggers include a block token expiring on a long open stream (`BLOCK_TOKEN_VERIFICATION_FAILED`, which the client is designed to refresh and retry) or a transient datanode or network failure, combined with a sequential read that starts inside a buffered checksum range (any read size or seek offset not aligned to bytes per checksum). Fix: when `prepareRead` throws after the call has already copied bytes, seek the chunk stream back by that many bytes (and reset the `ByteBuffer` position) before rethrowing, so a failed call has no effect and the retry resumes from where the call started. Successful reads and reads that fail before copying anything are unchanged. When retries are exhausted, the stream is now left at the start of the failed call instead of past the copied bytes. Positioned reads and the streaming read path are not affected. ## What is the link to the Apache JIRA https://issues.apache.org/jira/browse/HDDS-17000 ## How was this patch tested? Added `retriedReadReturnsContiguousData` and `failedReadLeavesPositionUnchanged` to `TestBlockInputStream`, using a real `ChunkInputStream` with ReadChunk stubbed to fail with `BLOCK_TOKEN_VERIFICATION_FAILED`, for both `byte[]` and `ByteBuffer`, after an unaligned and an aligned first read. Without the fix, the four unaligned/exhausted cases fail (wrong bytes, "Inconsistent read", wrong position); with the fix all pass. Also verified locally: the reproduction attached to the JIRA shows the defect on master and no longer does with the fix; `hdds-client` and `ozone-client` unit tests, `TestChunkInputStream` and `TestKeyInputStream` pass; checkstyle is clean. 🤖 Generated with [Claude Code](https://claude.com/claude-code) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
