News for hackers. Stuff that matters.
Re:Distributed Locks Are Dead; Long Live Distributed Locks by sriram_malhar on Wednesday June 05, 2019 @12:34PM (#20106136)
More accurately, these are leases, not locks in the traditional sense. The lease expires when the corresponding client session ends, which is detected by the absence of a heartbeat from the client.
Re:Distributed Locks Are Dead; Long Live Distributed Locks by hcnews on Wednesday June 05, 2019 @02:11PM (#20107412)
Locks are always leases in a distributed system, at least in my experience (and for a sane system).
Re:Distributed Locks Are Dead; Long Live Distributed Locks by hinkley on Wednesday June 05, 2019 @02:28PM (#20107623)
Plus ça change, plus c'est le même chose.

I think the first time I encountered the idea of leases it was dabbling in some CORBA code, back when enough people knew what that acronym means that Sun Microsystems thought they should include an orb in the dev kit. And then again in their RPC mechanism for Java.

Since then I've encountered the concept from others only a handful of times. Object leases aren't that important if your distributed state is straightforward. Platonic REST with no state doesn't need them, nor really do the fully stateful servers we had for about 20 years. And then there are concensus protocols like Raft which fill in some of the gaps in between.

Re:Distributed Locks Are Dead; Long Live Distributed Locks by hinkley on Wednesday June 05, 2019 @06:43PM (#20109873)
Having read the article, I'm not sure I agree.

First, leases aren't mentioned at all. Second, committing changes looks a lot like optimistic locking, due to the version number they're assigning to the lock.

Re:Distributed Locks Are Dead; Long Live Distributed Locks by grogers on Wednesday June 05, 2019 @08:40PM (#20110504)
It's not a lease, anyone can force close your session (which unlocks your locks), which takes effect immediately, not after waiting out a lease holding grace period.