Coda File System

Re: crash in rvmlib_free (not necessarily) during repair

From: <>
Date: Wed, 10 Jul 2013 10:13:33 +0000
On Tue, Jul 09, 2013 at 10:13:24PM +0200, Piotr Isajew wrote:
> > How large log do you have? I was using 16384 for several years but since
> > I once also got "No space left in volume log" I am now now using 65536.
> > (Apparently the data amounts have grown as the time goes :)
> I run the default setting for now. I can increase it of course,
> but I like your idea to check wired connection first.

The default setting is practically too low so I suggest you increase it.
RVM is nowadays a way bigger than it was in the last century.

> True. I don't want to go against the current. Having replicated
> volumes would be great, but the main reason of my interest in coda
> is client-side cache, so if anything other will fail, I'll just
> go with one-server system, and use a machine with RAID for it.

This is not bad. Replicated servers add redundancy but good backups
give almost the same availability even in case of failures, as long as
the client caches are well populated.

> > I have successfully run Coda with geographically spread servers - but
> > the network between them was reliable and such a setup is not something
> > inherently supported by Coda design.
> I suspect, that this kind of setup is extremely sensitive to
> disconnections during writes, but can easily handle scenario,
> where there are disconnections during reads, so depending on
> usage pattern it may work flawlessly or lead to problems I have
> now.

Quite right. Read-only operations do not depend on server-server
connectivity. Our data here consists mostly of "write once, use many
times by many clients" which makes possible problems less noticeable.

A dirty trick for populating large volumes can be using an old
client in fully connected mode. It writes to all servers at once
and they do not have to resolve the copies afterwards.
This is of course not something to rely on in production, but
can be useful as the last resort when nothing else helps ;)
as long as the compatibility persists.

> > Would it be feasible for you Piotr to make a test with the servers on
> > a wired connection?
> I can do this. It's just that putting both servers in the same
> Ethernet eliminates some redundancy in the setup. 

As soon as you put servers say in different rooms it makes the setup much
more resistant e.g. in a case of a local fire. A wired net can be quite
large too - the idea is to make the connection resemble an ideal one (no
packet loss, jitter, duplication), not necessarily having the servers
on the same physical wire or switch.

Anyway, a single-server Coda realm is almost as good, too.

Looking forward to hear whether your servers behave better
on a better network.

Received on 2013-07-10 06:22:45