Buildbot infrastructure upgrade
Hannu Nyman
hannu.nyman at iki.fi
Mon Mar 15 20:32:03 GMT 2021
Petr Štetiar kirjoitti 13.3.2021 klo 8.03:
> Petr Štetiar <ynezz at true.cz> [2021-03-12 16:38:46]:
>
> Hi,
>
>> I would like to let you know, that we've prepared upgrade of the Buildbot
>> based infrastructure and there is a plan to make those changes alive.
> we've just finished the migration.
>
I suspect that the scheduling logic in the new buildbot is still wrong.
Looking at the grid and builders views, it seems that once there are new
commits, buildbot restarts building all targets in alphabetical order.
Instead it should launch the "longest waiting" target next, like the old
buildbot did. The current logic pretty much disables the building of the
alphabetically last targets...
https://buildbot.openwrt.org/master/images/#/builders
https://buildbot.openwrt.org/master/images/#/grid
So far in the last three days, apm821xx/nand has been build 16 times,
apm821xx/sata 13 times, bcm47xx/generic only 4 times, and x86/64 just 1 time.
That was from master images buildbot, but the same logical behaviour probably
goes in all buildbot instances. (master, 21.02, 19.07, images & packages)
master packages shows the same:
https://buildbot.openwrt.org/master/packages/#/builders
(Evidence of the faulty logic can also be seen in the download directory, the
the timestamp on the "alphabetically last" targets are mainly from Saturday,
while the a* targets are all from today.)
Hopefully the logic can be corrected to be more like the previous buildbot's
scheduling logic, where targets were built pretty nicely in round-robin fashion.
More information about the openwrt-adm
mailing list