Re: locking a file renders repos unusable with older clients

On Wed, Mar 30, 2005 at 04:09:05PM +0200, Tobias Ringstr??m wrote:
> Branko ??ibej wrote:
>
> >Creating those new tables should take close to no time at all. I doubt
> >there's a repository that's so busy that you can't afford to lock it
> >for one second.
>
> No, I was thinking that it would be hard to aquire the write lock, not
> that it taks a long time to upgrade. AFAIK, new readers can still "get
> in" while svnadmin is waiting for the write lock.
>
> /Tobias
>

I believe you are correct that POSIX does not guarantee FCFS.

In fact, "our very own" Shlomi Fish wrote a patch to the linux
kernel to add FCFS semantics to fcntl:

http://www.uwsg.iu.edu/hypermail/linux/kernel/0208.2/1164.html

In practice (and probabilistically), writers eventually get the
lock.

What I usually do to briefly quiesce high traffic systems is add
a firewall or iptables rule to drop SYNs on the service I want to
quiesce. Existing connections run down, no new connections are
allowed. Because the SYNs are dropped rather than rejected, clients
don't even get error messages; their TCP just retries.

--ben

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Wed Mar 30 16:57:38 2005