[Labs-l] *COMPLETE* Planned maintenance

[Petr Bena]

990 GIGABYTES in single file?

"Holy Jesus on a stick" --Debra Morgan

On Fri, Oct 11, 2013 at 7:21 AM, Marc-André Pelletier
<mpelletier at wikimedia.org> wrote:
> Hello all,
>
> The switch to the new NFS server hardware is now complete, with only 12
> hours of delay.  :-)
>
> Seriously though, the filesystem copy took so long because - despite
> having prepared with a copy a day early so that only a quick rsync would
> suffice - there ended up being a bit over 4T worth of files that were
> touched in that 24h period.
>
> Log files, by their nature, are constantly updated.  Normally, that
> shouldn't have been a major issue since the rsync would just copy them
> again...  except that many of those log files were *hundreds of gigs* in
> size.  This caused the filesystem resync to take a bit over 11 hours.
>
> While Cyberbot was the overall winner, with very nearly 2G of logs to
> his name, our winner for most impressive single log is yifeibot with
> 990G in a single log file!
>
> I would really much rather not have to turn quota on on our file system:
> it is very useful to be occasionally able to handle huge datasets.
> However, if users abuse the freedom I will have no choice but to do so
> in order to protect reliability and QoS.
>
> So, the delay having been that long, some things may have broken that I
> was no longer able to notice (I've been at this for 14h straight, now,
> and need sleep).  I'll be on hand tomorrow to help work out any kinks
> that may have slipped in.
>
> On the positive side, however, we are now on new hardware for the NFS
> server and it seems to be working quite fine.  Yeay!
>
> -- Marc
>
>
> _______________________________________________
> Labs-l mailing list
> Labs-l at lists.wikimedia.org
> https://lists.wikimedia.org/mailman/listinfo/labs-l