[Gluster-users] Geo-replication updates

Aravinda avishwan at redhat.com
Tue Dec 6 17:00:09 UTC 2016

Geo-replication feature got lot of improvements in 3.7, 3.8 and 3.9 
releases. We will write about these features soon. A few updates from 
last week:

- Kotresh discovered awesome option in rsync(`--ignore-missing-args`) 
which helps reducing rsync retries and improving the sync performance if 
your workload involves lots of unlinks. This option is only available 
with Rsync version 3.1.0.
   Patch sent to include this rsync flag by default.
   http://review.gluster.org/16010 (under review)

- Geo-rep creates entries in slave before triggering rsync for data 
modifications. Rsync will sync data using aux-gfid-mount. Added 
`--existing` flag to rsync command to reduce rsync retries if entry does 
not exists in Slave(Note: Entry failures are handled separately)
   http://review.gluster.org/16010 (under review)

- Improved local node detection
   Geo-rep monitor process spawns one worker process for each local 
brick. To identify the local node it was doing real network check, patch 
sent to identify the local node by comparing with Brick host UUID(From 
volume info) with Host UUID(gluster system:: uuid get)
   http://review.gluster.org/16035 (under review)

- Added additional details to Geo-replication events for Events 
   http://review.gluster.org/15858 (merged)

- Fixed a issue with status output during Hybrid crawl
   http://review.gluster.org/15869 (merged)

- Patch sent to avoid Geo-rep workers restart when 
`--log-rsync-performance` config changes
   http://review.gluster.org/15816 (under review)

- Patch sent to fix a worker crash during cleanup and exit
   http://review.gluster.org/15686 (under review)

Last but not least, we have rewritten upstream Geo-replication 
documentation. Please let us know if it is useful.


More information about the Gluster-users mailing list