On 10/18/2016 06:40 PM, Daniel Shahaf wrote:
> [ I asked this on IRC but haven't got a reply, so trying again... ]
> I understand there's an issue of jenkins slowness.
> What I know of that issue is that it's suspected to be caused by
> b1b64d98e5971437abb3fe793f6fb58eab1ef4fd [to jenkins.debian.net.git] and
> that its symptom is
> I assume the issue is unique to that one jenkins job.
I thought I remembered Holger mention reproducible_scheduler.py is also
much slower since the switch to SQLAlchemy (the commit you are
As for reproducible_scheduler -- Mattia guessed it was because we now
query each source package (to see if it exists) instead of relying on
sqlite's "REPLACE INTO" functionality -- but turns out this is not the
case. From the diff you can see we always checked to see if the source
package exists, so no additional sql queries were added.
> Is there any further concrete data available, please? All successful
> builds linked on that page are "slow" builds (25min+), do we have a full
> console log of a "fast" build for comparison? Or a log of build
> duration as a function of time, that goes far enough into the past to
> have some "fast" builds in it? (So I can check whether the build
> durations correlate with when b1b64d98 was deployed.)
> Basically, anything you can tell me about the "fast" builds beyond "They
> existed" would help :-)
I'm super curious now too! I'll revert the change to SQLAlchemy and see
if the there is a noticeable speed up for either of these scripts tomorrow.
> Reproducible-builds mailing list
Reproducible-builds mailing list