Hanisha Koneru created HDDS-6526:
------------------------------------
Summary: Container deletion should not depend on UsedBytes counter
Key: HDDS-6526
URL: https://issues.apache.org/jira/browse/HDDS-6526
Project: Apache Ozone
Issue Type: Bug
Components: SCM
Reporter: Hanisha Koneru
HDDS-5359 fixes the issues with how blockCount and usedBytes are updated. But
even after the fix in HDDS-5359, usedBytes cannot be trusted to be an accurate
representation of the actual number of bytes in the container. This is because
usedBytes is updated in memory first when a chunk is written and then updated
in DB during the putBlock call.
Since blockCount is reliable after HDDS-5359. So SCM should delete a container
based on the blockCount and not on usedBytes.
Also, when a DN receives a delete container command from SCM, it should double
check that there are no valid blocks in the container before deleting it. This
is an extra check on the DN side to avoid deleting a non-empty container.
For older containers for which blockCount could be wrong, the extra check on
the DN side will ensure that non-empty containers are not deleted. Note that in
this scenario, SCM should not mark the container as DELETED before DN can
confirm that the container is actually empty.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
