Running validation compaction sequentially
Set the snapshot_override option to run validation compaction sequentially rather than in parallel. Switching repair snapshot behavior might be necessary for latency sensitivity reasons.
Set the snapshot_override
configuration option to true to run
validation compaction sequentially for both subrange and incremental repairs. The
repair takes a snapshot of all data on all the nodes and then each node validates
the snapshot one at a time.
Switching the repair snapshot behavior might be necessary for latency sensitivity
reasons. The snapshot override option determines the manner in which Merkle trees
are built during validation compaction: all at once in parallel (default); or one
node at a time to reduce impact on server load.
Attention: Be aware that
enabling the snapshot override results in considerably slower repair completion
times. Roughly estimate the slowdown as equivalent to the replication factor.
For instance, with an RF of 3, the repairs builds the Merkle tree three times,
one at a time, so sequential compaction could take at least three times as long
as parallel.
opscenterd.conf
The location of the opscenterd.conf file depends on the type of installation:- Package installations: /etc/opscenter/opscenterd.conf
- Tarball installations: install_location/conf/opscenterd.conf
Procedure
- Open opscenterd.conf for editing.
-
Set the
snapshot_override
option toTrue
:[repair_service] snapshot_override=True
- Restart opscenterd.
- Monitor the subrange and incremental repair progress on the Status tab. Be sure to observe the slower progress of validation compactions in the Validation Compactions pane. After differences are determined by Merkle tree comparisons, the streaming progress continues at its regular pace as shown in the Streams pane.