[Gluster-users] network.ping-timeout details
Pranith Kumar Karampuri
pkarampu at redhat.com
Wed Aug 6 15:15:44 UTC 2014
Do you think you have any workload you can generate to see the kind of
latency numbers your setup has?
You can probably use 'gluster volume profile <volname> start/info/stop'
to figure out the maximum latencies and configure a number closer to the
numbers you get in this profile?
The following contains steps to perform profiling.
On 08/05/2014 06:53 PM, Scott Merrill wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA512
> I've read the description for the network.ping-timeout option, but the
> admonishment at the end isn't particularly helpful:
> "This reconnect is a very expensive operation and should be avoided."
> Can that be quantified at all? Does this boil down to a "thundering
> herd" problem, such that the problem is exacerbated by lots of
> clients? If the client count is relatively low, is the problem
> reduced in any way?
> The bot in #gluster provides a little more info, but not enough:
> "The reason for the long (42 second) ping-timeout is because
> re-establishing fd's and locks can be a very expensive operation.
> Allowing a longer time to reestablish connections is logical, unless
> you have servers that frequently die."
> If I'm using a distributed replicated volume to ensure availability of
> my data, it seems sub-optimal to block all Gluster I/O for 42 seconds.
> Is there guidance on how to balance the timeout for different scenarios?
> My searching hasn't provided much useful insight on the matter. If
> there are documents available online for me to review, I'll happily
> read them.
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG/MacGPG2 v2.0.22 (Darwin)
> Comment: GPGTools - https://gpgtools.org
> -----END PGP SIGNATURE-----
> Gluster-users mailing list
> Gluster-users at gluster.org
More information about the Gluster-users