ubuntu-phone team mailing list archive

[Didier Roche <didrocks@xxxxxxxxxx>]

Le 28/03/2014 21:19, Sergio Schvezov a écrit :
> On Fri, Mar 28, 2014 at 2:38 PM, Didier Roche <didrocks@xxxxxxxxxx> wrote:
> > Hey,
> snip
>
> > On a finale note on the CI status:
> > Some people would have noticed that the CI Train jenkins got killed multiple
> > times by the OOM. The machine went high on memory as jobs were piling up
> > (the archive had a lot of builds and powerpc was taking time to catch up, as
> > more jobs running there) and 1 GB wasn't enough. As we can't add more RAM
> > dynamically on this machine, after the 4th outage within 2 hours, we decided
> > to add as a band-aid some swap file to it. It was a coincidence that the
> > discussion on finishing the move to prodstack (with more memory) happened
> > some hours before the outage... (and the move should hopefully happen next
> > week).
> Not sure if there is really a technical limitation, but can't you just
> spin up more than one jenkins instance and link back to the
> spreadsheet accordingly?
>
> Even if you move to another stack; this would also help with those
> jenkins outages as we'd only lose some capacity instead of all of it.
>
> Anyways, that's my Friday suggestion with not too much thought into or
> details for a more informed one.

Remember that CI Train is a temporary solutions, part of the shortcuts 
that was done to put it into production in two days instead of multiple 
weeks of development is that the metadata or on the file system. So to 
keep and go on from the same state, we need to share the FHS. When I 
asked for that some months ago, It seemed attaching the same volume to 
multiple machines isn't something prodstack and canonistack is suitted 
for in term of rules and technology for efficient disk integrity.