Implement a rack consul lock. about choregraphie

[kamaradclimber]

Suggested algorithm looks ok to me.

I'm a bit concerned by how this will be used: if I use the "rack" mode, I would still need to specif the concurrency level for nodes (but default case would be to allow the full rack whatever its size).

I would suggest to subclass ConsulLock to:

• use the rack name as the consul lock id (we usually set it to the node name)
• by default concurrency is 1 which corresponds to 1 rack allowed at a time
• we add a new field by holder (by rack) containing the list of nodes in that rack using the lock

{"version":1,"concurrency":1,"holders":{"rack1": "today"}, rack1: {"node1": "today"}}

your algorithm still holds: we maintain the list of nodes in the rack using the lock and release the lock only when this list is empty.

It supports correctly the possiblity to have multiple racks under the lock (default to 1) and allow unlimited number of nodes in the same rack to be in the lock.

What do you think?

[ashangit]

Fine for me I will just use a json like that one to be able to easily get all holder by racks:
{"version":1,"concurrency":1,"holders":{"rack1": "today"}, "racks": {"rack1": {"node1": "today"}}}

[ashangit]

As seen with @Annih we are thinking of using this structure instead to still have nodes has holders:
{"version":1,"holders":{"node1": {"ts": "today", "rack": "rack1", "concurrency":1}}

[kamaradclimber]

fine by me. Note that this is an upgrade of the format so bump the version