Umesh Kumar Kumawat created HBASE-30320:
-------------------------------------------
Summary: TestMemStoreLAB#testLABChunkQueue OOM on fast hardware
due to unbounded off-pool chunk allocation
Key: HBASE-30320
URL: https://issues.apache.org/jira/browse/HBASE-30320
Project: HBase
Issue Type: Bug
Components: test
Affects Versions: 2.6.7, 2.5.16
Reporter: Umesh Kumar Kumawat
Assignee: Umesh Kumar Kumawat
The test spawns 10 threads that allocate 256KB chunks in a tight loop for a
fixed 1-second window (Thread.sleep(1000)). Once the ChunkCreator pool reaches
its maxCount, subsequent allocations fall through
to unbounded off-pool chunk creation (ChunkCreator#createChunk). All chunks
remain live in mslab.chunks until mslab.close() is called after the threads
stop.
On fast hardware (Apple Silicon, high memory bandwidth), threads complete
~5,000 iterations/thread/sec, creating ~50,000 chunks (~12.5GB) in that
1-second window. On slower x86 CI machines, the same
window produces only ~700 iterations/thread/sec (~1.7GB), fitting within the
default 2.2GB surefire heap.
The test's memory consumption is non-deterministic — it depends on CPU/memory
speed rather than a fixed workload.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
