Huang Kuan Hao created HDDS-16326:
-------------------------------------
Summary: Cache the container chunks directory instead of stat-ing
it on every chunk operation
Key: HDDS-16326
URL: https://issues.apache.org/jira/browse/HDDS-16326
Project: Apache Ozone
Issue Type: Improvement
Reporter: Huang Kuan Hao
Assignee: Huang Kuan Hao
ContainerUtils.getChunkDir builds a new File and calls exists() on the
container's chunks directory on every call:
File chunksDir = new File(chunksPath);
if (!chunksDir.exists()) {
throw new StorageContainerException(..., UNABLE_TO_FIND_DATA_DIR);
}
It is reached through ContainerLayoutVersion.getChunkFile (line 115), which
FilePerBlockStrategy calls on every WriteChunk (line 153) and every ReadChunk
(line 240), plus the stream/finish/delete paths. So every chunk operation
resolves and stat()s a directory that is fixed for the container's lifetime.
On the read path the stat is pure added work: the read otherwise goes straight
to open().
Fix: resolve and validate the chunks directory once (e.g. cache the File on
KeyValueContainerData when the container is opened/loaded) and reuse it on the
data plane, instead of new File + exists() per operation.
Note on behavior: the per-op exists() currently front-runs a missing or
unmounted chunks dir with UNABLE_TO_FIND_DATA_DIR. After caching, a chunks dir
that disappears mid-life is still caught by the subsequent open()/writeData,
which triggers volume-failure handling (onFailure) - only the error code
differs in that rare case. The container scanner (KeyValueContainerCheck) is
unaffected: it stats getChunksPath() directly and reports MISSING_CHUNKS_DIR on
its own, without going through getChunkDir.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
