CLI support for riak_client:repair_node()

[hmmr]

Expose riak_client:repair_node/0 as riak admin node repair.

Related PRs: OpenRiak/riak#36. Test in OpenRiak/riak_test#29.

[martinsumner]

I'm not sure about repairing multiple nodes, what happens when partitions are covered in more than one nodes - the behaviour might be unpredictable. Having nodes of all as well, I'm not sure it is obvious what will happen in this case.

I've only done non-functional testing for repairing a single node, so I think it should be restricted to repairing a single node at a time. There are are advanced configurable options like double-pair that might not behave as expected with multiple node repairs being triggered.

[hmmr]

Would it, then, be appropriate to extend the action of riak admin node repair with status subcommand which would report if any repairs are in progress? And if so, prevent a new repair from starting?

[martinsumner]

Repairs can be tracked like other handoffs/transfer operations - so there's not a need for a status, in that there is another way of following this .

However, there is a repair_status for each vnode (tracked by the vnode_manager). So just as node_repair is just a wrapper to call repair on all vnodes on the node, there could be a call to repair_status that collates the status across vnodes - but the status is just in_progress/not_found.

There are better ways to track progress - so I'm not sure it is necessary, but it could give you a definitive in_progress/complete statement.

There may be more complications here. I can't remember the history behind this comment:

https://github.com/OpenRiak/riak_test/blob/a9605720be2efa567dde65489b0caf415f7e4d92/tests/node_repair_big.erl#L187

Perhaps I hadn't thought of using repair_status, or perhaps there was an issue with it.

[hmmr]

I have added subcommands repair status and repair stop, and extended start with a check to only allow repair to start when no other repairs are running.