[Libreoffice] "random" build failures in sfx2/sdi/sfxitems.sdi

[Christian Lohmaier]

Hi Lionel, *,

On Sun, Dec 11, 2011 at 6:06 PM, lionel at mamane.lu <lionel at mamane.lu> wrote:
> On Sun, Dec 11, 2011 at 05:48:54PM +0100, Christian Lohmaier wrote:
>> On Sun, Dec 11, 2011 at 10:58 AM, lionel at mamane.lu <lionel at mamane.lu> wrote:
>
>>> BTW, it took about 20 minutes for the full log to arrive at
>>> http://tinderbox.libreoffice.org/MASTER/status.html; any reason for
>>> such a significant delay?
>
>> The cronjob that regenerates the page
>
> Ah, it is a cronjob, I see.

Yes, it needs to be a cronjob since processing the mails creates
intermediate files that are larger than the ulimit for mail-processing
(and at least on the original server, there were not enough
user-rights to configure it differently.
Thus the mails are batch-processed via cron to bypass this limit.

Another reason why it is a cronjob is that it was also gathering
checkin-data from the OOo-childworkspaces and did gather information
from EIS - and obviously you don't want to spam those systems (and
querying hundreds of cws takes time....).
But as the OOo-part of the tinderbox setup is history, and there is no
git integration, all that the cronjob needs to do is to process the
mailed buildlogs.

Thus two possibilities:
* shorten cronjob interval
* look whether the ulimit is still an issue or can be disabled, and if
so set $REGENERATE_AFTER_MAIL=1 to have the pages regenerated as soon
after a mail is parsed.
(but doesn't make sense with batch-processing, then it makes more
sense to process all piled-up mail first, and then only regenerate the
html pages once)

ciao
Christian