RE: [Opendlm-devel] ODLM/OGFS Recovery -- CR transaction locks and memexp
"Cahill, Ben M" <[email protected]> Fri, 23 Apr 2004 14:46:42 -0700
| Newsgroups | gmane.comp.file-systems.opengfs.devel |
|---|---|
| Message-ID | <[email protected]> |
> >=20 > > The other tricky part is, if we withhold *all* locks from=20 > > OpenGFS (not just the ones that were blocked by the dead=20 > > node), will surviving nodes be able to finish their write=20 > > transactions (and release their CR transaction locks)? They=20 >=20 > A write while holding a read lock?? Actually yes, but not on the resource (which must have an EX lock). The transaction lock (resource in OpenDLM parlance) is global across the cluster. It is used EX only when doing a filesystem journal recovery. Each (normal) write transaction is surrounded by a CR transaction lock. This makes the node doing the recovery playback wait for everyone else to finish writing, then it blocks those other nodes from writing while replaying the journal. >=20 > > need to do this, or else the node that attempts the journal=20 > > recovery will never get the (EX) transaction lock .... > >=20 > > .... Again (false positive/negative), how do we conclusively=20 > > tell which locks were blocked by dead node vs. not? I'm happy to say that I just did an audit of OpenGFS, and there is never a case in which the transaction lock surrounds the acquisition of another lock. Therefore this deadlock scenario is not a factor! (good). This would have been a fundamental design flaw, anyway, and would have caused deadlocks even with memexp. > I was just curious how opengfs retrieved the memexp storage=20 > area of a dead node? =20 > This seems counter-intuitive. Unless every node had a copy=20 > of each others storage area.=20 The memexp storage is all in one place. There is one and only one lock storage server for memexp. That's the SPOF that we're trying to avoid by using OpenDLM. If the memexp server was running on the dead node, too bad, everyone is hosed. Otherwise, the data is available to anyone who asks. -- Ben -- Opinions are mine, not Intel's =20 >=20 > Best Regards, > Don >=20 ------------------------------------------------------- This SF.net email is sponsored by: The Robotic Monkeys at ThinkGeek For a limited time only, get FREE Ground shipping on all orders of $35 or more. Hurry up and shop folks, this offer expires April 30th! http://www.thinkgeek.com/freeshipping/?cpg=12297