Shilun Fan created HDDS-15371:
---------------------------------
Summary: DiskBalancer should use monotonic time for delayed
replica deletion
Key: HDDS-15371
URL: https://issues.apache.org/jira/browse/HDDS-15371
Project: Apache Ozone
Issue Type: Improvement
Components: Ozone Datanode
Reporter: Shilun Fan
Assignee: Shilun Fan
DiskBalancer uses System.currentTimeMillis() to schedule and check delayed
deletion of old replicas after a successful container move.
The delayed deletion is based on a relative delay configured by
replicaDeletionDelay. Since System.currentTimeMillis() is wall-clock time, it
can be affected by system clock changes such as NTP adjustment, manual time
changes, or VM time drift.
If the system clock moves forward, old replicas may become eligible for
deletion earlier than expected. If the system clock moves backward, old
replicas may remain in the pending deletion queue longer than expected.
DiskBalancer already uses Time.monotonicNow() for other relative-time logic,
such as bandwidth throttling with nextAvailableTime. The pending deletion delay
should follow the same pattern and use monotonic time for both deadline
calculation and expiration checks.
This makes delayed replica deletion independent of wall-clock changes and
consistent with other DiskBalancer delay/elapsed-time logic.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
