1243436 - Update b2g_bumper to commit to mozilla-inbound instead of b2g-inbound

Status: RESOLVED WONTFIX

(Release Engineering :: General, defect)

Description

[[:fabrice] Fabrice Desré]

Since b2g-inboud is not merged back to m-c anymore, we need to change the integration branch the bumper is pushing to.

Comment 1

[Hal Wine [:hwine] use NI!]

Based on chat in #releng, it's not clear that this is the intended response to the change in tiers.

Looking for an 'ack' from Sheriffs

Comment 2

[[:fabrice] Fabrice Desré]

Attachment: rip-b2g-inbound.patch

Comment 3

[[:fabrice] Fabrice Desré]

Doug?

Comment 4

[Lawrence Mandel [:lmandel] (use needinfo)]

AFAIK this will have no impact on Firefox. After speaking with Hal, it is best to ensure that one of the sheriffs is aware of this change before it is made in case there's something we're missing. Unless there is a real impact on Firefox, we should go ahead with this change and say goodbye to b2g-inbound once tomcat gives us an ack.

Comment 5

[Phil Ringnalda (:philor)]

ack, because from our perspective it's nothing more than 20-30 "NPOTB DONTBUILD" pushes per day, mildly annoying when one push-races you but otherwise a no-op.

Since the point of b2g-inbound+bumper was to see how Gaia pushes worked out on top of a known-good Gecko, while other trees showed how Gecko pushes worked on top of a known-good Gaia, it seems like this will be a nearly useless and constantly confusing waste of resources to run, but that's not a sheriffing concern, as long as the people who are looking at B2G test results have decided they don't care.

Comment 6

[Hal Wine [:hwine] use NI!]

Comment on attachment 8712753 [details] [diff] [review]
rip-b2g-inbound.patch

Review of attachment 8712753 [details] [diff] [review]:
-----------------------------------------------------------------

The changes to master.py look reasonable - r+ on that

The changes to beagle.py will break vcs-sync -- those are for a separate application (vcs-sync), r- on that.

Feel free to carry forward the r+ and land & deploy the master.py change (ideally with a fresh patch).

Comment 7

[Phil Ringnalda (:philor)]

Probably the most significant thing you're letting yourself in for with this is the absolute certainty that those multiple times a day nightlies that are still being built off mozilla-central will be built off random broken Gaia commits, and will then sit on those broken commits for days on end, as a sheriff picks a commit off mozilla-inbound to merge without looking at what b2g looked like at that cset, merges it on Friday, and then everyone leaves for the weekend.

Comment 8

[Carsten Book [:Tomcat]]

(In reply to Lawrence Mandel [:lmandel] (use needinfo) from comment #4)
> AFAIK this will have no impact on Firefox. After speaking with Hal, it is
> best to ensure that one of the sheriffs is aware of this change before it is
> made in case there's something we're missing. Unless there is a real impact
> on Firefox, we should go ahead with this change and say goodbye to
> b2g-inbound once tomcat gives us an ack.

ack from me - but philor raises a valid point for the b2g folks. It would be good to have someone from the CD team looking at the results from b2g tests from this bumperbots so we are sure we don't ship broken b2g nightlys since they are not sheriffed anymore.

Comment 9

[Phil Ringnalda (:philor)]

Mmm, no. The whole point of making b2g tier-3 is that we are *not* going to choose an m-i revision to merge to m-c based on b2g's needs. If we do this, if we bump b2g on m-i, then we are going to choose m-i revisions based on the latest one which is good on tier-1 jobs, period. So the options for b2g nightlies with m-i bumping are:

* ship broken nightlies

* do not have nightlies

* redesign the b2g nightly pipeline to upload them to a staging spot, and have a manual QA signoff before they ship

* some grotesque and impossible thing with someone watching m-c 24/7, checking whether anything merged to it can be shipped, and for things which cannot, watch them 24/7 until they are no longer the tip, because at any time anyone with an LDAP account can use self-serve to trigger a new set of nightlies thinking they are only triggering desktop nightlies but actually also triggering b2g nightlies

Comment 10

[[:fabrice] Fabrice Desré]

Attachment: rip-b2g-inbound.patch v2

Changes to beagle.py removed.

Comment 11

[Hal Wine [:hwine] use NI!]

Comment on attachment 8713255 [details] [diff] [review]
rip-b2g-inbound.patch v2

Review of attachment 8713255 [details] [diff] [review]:
-----------------------------------------------------------------

still lgtm :D

Comment 12

[[:fabrice] Fabrice Desré]

pushed to default: https://hg.mozilla.org/build/mozharness/rev/b80c0164707c

Comment 13

[Armen [:armenzg]]

This has caused a bad tree closure on inbound. See bug 1244219.

Can we please back this out until bug 1244219 can handle this better? The solution might be to force b2g bumps to produce desktop builds.

If we don't back out until we fix that bug is that we will have bustages being merged around, closing trees and make lives of the sheriffs harder.

Comment 14

[Hal Wine [:hwine] use NI!]

Per discussion in #releng with agreement of sheriffs, temporarily backed out this change so trees will stay open over the weekend.

(In reply to Armen Zambrano [:armenzg] - Engineering productivity from comment #13)
> This has caused a bad tree closure on inbound. See bug 1244219.

This change did NOT cause the tree closure. It did tickle some undesirable behavior in newish CI code - SETA. 

There is a manual workaround (manually trigger some missing builds), but that takes both time and personnel.

There are about 5 ways to fix bug 1244219, so some time will be spent deciding which is best, then implementing, testing, and deploying.

bug 1244219 now blocks this bug.

Comment 16

[[:fabrice] Fabrice Desré]

Hal, according to https://bugzilla.mozilla.org/show_bug.cgi?id=1244219#c11 we can re-land that patch.

Can you take care of that? Thanks!

Comment 17

[Chris AtLee [:catlee]]

We're shutting down the bumper.