Ethan Rose created HDDS-11721:
---------------------------------
Summary: Container export may fail for Schema v2 container while
it is being scanned
Key: HDDS-11721
URL: https://issues.apache.org/jira/browse/HDDS-11721
Project: Apache Ozone
Issue Type: Sub-task
Components: Ozone Datanode
Reporter: Ethan Rose
Assignee: Ethan Rose
For container schema versions older than v3, there is a DB handle cache that
uses reference counting to handle cache eviction. The container scanner is
holding a reference to the DB as it is scanning the blocks while the container
is being exported. The scanner operates without a lock which is expected since
it is a slow, read-only background process. However, the export code makes an
assumption that every caller who has allocated a schema v2 DB reference has
done so under a container lock, and it will block until the other thread has
released the lock, meaning that once it tries to free the DB, it knows there
are no other references. In this case the scanner has a reference but no lock,
so export obtains the write lock and attempts to evict the DB from the cache,
but it fails due to the scanner's reference.
This is incorrect because export should be a read-only operation. It is a copy,
not a move, so it should not be evicting from the cache. We should remove the
cache eviction step, and probably convert the write lock it takes to at least a
read lock.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
