rebalance commands in GD2

[atinmu]

APIs to be defined:

rebalance start
rebalance fix-layout start
rebalance start [force]
rebalance stop
rebalance status

[atinmu]

@aravindavk can you please help in defining the ReST APIs for the above?

[nbalacha]

Information on the current rebalance behaviour:

Rebalance [fix-layout] start:

1. The rebalance process runs on every node of the volume
2. The rebalance process handles a single volume.
3. The rebalance process uses the -rebalance.vol volfile which is the same as the Fuse client volfile but only contains xlators from dht and down (No perf xlators, etc). It also seems to have io-stats loaded on top of DHT.
4. Rebalance requires that all child subvols be up as the selfheal which sets the layout of the directories requires all subvols to be available.. The process will terminate if any subvolume goes down.
5. A commit hash will need to be generated and passed to the process

Rebalance start force:
Same as above but skips some space checks. This is used when the rebalance is started by the remove-brick operation.

Rebalance stop:
Stops the ongoing rebalance operation. The process will finish migrating any files it might be in the middle of before terminating so it might take some time to actually stop.

[nbalacha]

Rebalance command:

Started running /usr/local/sbin/glusterfs version 4.0dev1 (args: /usr/local/sbin/glusterfs -s localhost --volfile-id rebalance/vol1 --xlator-option *dht.use-readdirp=yes --xlator-option *dht.lookup-unhashed=yes --xlator-option *dht.assert-no-child-down=yes --xlator-option replicate.data-self-heal=off --xlator-option replicate.metadata-self-heal=off --xlator-option replicate.entry-self-heal=off --xlator-option *dht.readdir-optimize=on --process-name rebalance --xlator-option *dht.rebalance-cmd=1 --xlator-option *dht.node-uuid=96f35379-4a82-4207-812f-d91462348cc6 --xlator-option *dht.commit-hash=3502166428 --socket-file /var/run/gluster/gluster-rebalance-55c320ec-8693-4d04-ab26-052f0032ab96.sock --pid-file /var/lib/glusterd/vols/vol1/rebalance/96f35379-4a82-4207-812f-d91462348cc6.pid -l /var/log/glusterfs/vol1-rebalance.log)

[atinmu]

Checklist

• ReST APIs availibility in wiki
• CLI commands
• Sufficient logging
• Availibility of metrics data if applicable
• Xlator options table upto date?
• Test cases & test coverage of the functionality

[kshlm]

@nbalacha Thanks for the intial inputs. The basic requirements for rebalance [start [force] | stop] should be easily doable. rebalance status should be doable as well.

Rebalance requires that all child subvols be up as the selfheal which sets the layout of the directories requires all subvols to be available.. The process will terminate if any subvolume goes down.

GD2 can only try to ensure that all subvols are up before trying to launch rebalance. We could try to kill the rebalance process when subvols are down. But based on the command line for the rebalance process, it looks like the rebalance process itself ensures that already.

A commit hash will need to be generated and passed to the process

What is this hash? Any random number or something specific?

[prashanthpai]

PR #487 has started implementing this.