I noticed a recent drop in build times for the Gerrit-Codegen jobs, but I can't explain it. Has anyone noticed this for other jobs?
I looked into the details a bit, but I still can't explain it. There was a big drop between jobs 347, 348 and 349:
#347 took 14 min 13 sec
#348 took 9 min 7 sec
#349 took 5 min 8 sec
Since then all codegen jobs are taking about 6 minutes. I look at the timing results from Maven's "Reactor Summary" at the end of the job's console output, and added them up and the results are as follows:
#347 took 6 min
#348 took 4 min 17 sec
#349 took 3 min 55 sec
So the actual build time is getting shorter. Looking at the individual components one can see only small speedups in each bundle or feature, with a couple of exceptions. One of the EMF models (oeprt.xtumlrt.interactions.model) as well as the codegen-feature and xtext-feature are building a lot faster. The rest are building faster as well, but the time drop is not as dramatic. Yet, these bundles have not changed.
But that still doesn't explain the huge difference with the totals above:
#347 difference of 8 min
#348 difference of almost 5 minutes
#349 difference of more than 1 minute
So the big difference seems to be before the builds actually start, resolving and downloading dependencies, and computing the reactor order. But looking at the trace there is something even stranger. The console output for job #347 doesn't show any "fetching" or "downloading" of artifacts, but it is the one that took longer!. The #348 shows fetching and downloading a lot of artifacts. Job #349, like #347 doesn't show fetching/downloading. More recent jobs show fetching but not downloading.
Can anyone explain? Is it because of the updated TPs? Job #347 was triggered by Gerrit change
https://git.eclipse.org/r/99368 which updated the TPs. Since #348 ran after that gerrit was merged, this would probably explain the job downloading new artifacts, but not the time difference.
If it matters, job #347 ran on June 15th at 7:13pm EST, job #348 ran on June 15th at 10:03am EST, and job #349 ran on June 15th at 10:16am.
Any ideas? I'm not complaining ;) But I'm quite interested in understanding this unexpected and welcome improvement.
--
Ernesto Posse
Zeligsoft
