[okfn-help] eu3 load issues

[Rufus Pollock]

cc'ing okfn-help (rather just sysadmin alias) -- seems more
appropriate (I think we should either deprecate sysadmin alias or turn
it into a dedicated mailing list ...)

On 19 August 2010 23:58, William Waites <william.waites at okfn.org> wrote:
>  As Friedrich pointed out a little while ago, many
> log messages on eu3 like this:
>
> Aug 19 22:42:08 eu3 kernel: 4gb seg fixup, process ldconfig (pid 15268),
> cs:ip 73:08060475
>
> Seems to be an issue with 32 bit instances running
> in a 64 bit host under Xen, known issue for some
> time (ca 2007).
>
> This person seems to have had an experience which
> sounds remarkably similar to ours:
>
> http://oilq.org/fr/node/5891
>
> I graphed the frequency of these messages versus
> system load (tried to get them to line up nicely)
> here:
>
> http://styx.org/~ww/eu3/

Amazing detective work will and friedrich

> It seems inconclusinve whether there's a correlation.
>
> Assuming this is the problem, I can see three
> possible things we can do:
>
>  * upgrade the kernel on eu3 to something more
>    recent (means rebuilding it) and hope the problem
>    goes away.
>  * upgrade eu3 to a 64-bit host (means rebuilding
>    it and expensive)
>  * start divesting ourselves of AWS (means rebuilding
>    it but may be cheaper and in the long run better)

I an immediate initial think would be to separate DB and Frontend
(webapp). This would also give us a clearer picture of what is
happening.

> Friedrich has opened a ticket with AWS to see what
> they have to say, but we have no support contract
> with them so we'll see what and when we hear back
> from them.

Yes, if you aren't paying for support you basically get no answer ...
Could try the forums too?

> In the meantime, continuing to look at other
> explanations...

It would be great if we could reproduce the problem -- we could spin
up a temporary AWS box, put on a ckan instance and fire lots of
requests at it ...

rufus