1377065 - something has gone wrong while rewriting or rebasing your commits. The commits being pushed no longer match what was requested.

Status: RESOLVED FIXED

(Conduit Graveyard :: Transplant, enhancement)

Description

[Carsten Book [:Tomcat]]

from https://bugzilla.mozilla.org/show_bug.cgi?id=1357021 - seems it landed on autoland but also the autoland bot commented with 

We're sorry - something has gone wrong while rewriting or rebasing your commits. The commits being pushed no longer match what was requested. Please file a bug.

is there any action Fischer (the dev) need to do ?

Comment 1

[Fischer [:Fischer]]

(In reply to Carsten Book [:Tomcat] from comment #0)
> from https://bugzilla.mozilla.org/show_bug.cgi?id=1357021 - seems it landed
> on autoland but also the autoland bot commented with 
> 
> We're sorry - something has gone wrong while rewriting or rebasing your
> commits. The commits being pushed no longer match what was requested. Please
> file a bug.
> 
> is there any action Fischer (the dev) need to do ?
Thanks for filing this bug.
Just for information, heard dholbert said he has seen that in the past when someone accidentally triggers autoland twice.
No sure if my case related to that.
And I check out the autoland, the bug 1357021 patches are checked-in.

Comment 2

[Gregory Szorc [:gps]]

I don't see anything obvious in the autoland logs. It certainly looks like this could be the result of triggering autoland multiple times. But I don't see evidence for or against it.

Comment 3

[:glob ✱]

it looks like what's happening here is the "duplicate request" logic only checks for a duplicate in-flight request.

i suspect what's happening is the first request is being autolanded, reviewboard is being signalled, however it doesn't update the UI of an existing page.  the user then requests a second landing without having refreshed the page, resulting in the duplicate.

Comment 4

[:glob ✱]

Attachment: autoland: improve duplicate request detection (bug 1377065)

Refuse to land commits that have already landed, in addition to those waiting
to be landed.

Review commit: https://reviewboard.mozilla.org/r/158840/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/158840/

Comment 5

[Gregory Szorc [:gps]]

Comment on attachment 8887967 [details]
autoland: improve duplicate request detection (bug 1377065)

https://reviewboard.mozilla.org/r/158840/#review164378

Looks good.

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/hgcustom/version-control-tools/rev/941cdd5b6e13
autoland: improve duplicate request detection r=gps

Comment 7

[:glob ✱]

looking at this before deploying to production, i think i have the wrong logic:

> WHERE (landed IS NULL or NOT landed)

we want to block revisions that are in-flight (landed IS NULL) and those that have already landed (landed == true).
in other words we should only allow new requests, or that failed to land previously.
pretty sure this clause should be

> WHERE (landed IS NULL or landed)


as i'm second guessing myself, i'd like a sanity check from you gps.

Comment 8

[Gregory Szorc [:gps]]

Yeah, the change to `landed IS NULL or landed` seems reasonable.

Attempting to model tri-state in optional columns is wonky. I really wish there were an enum tracking explicit state.

Comment 9

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/hgcustom/version-control-tools/rev/dcd16732ac79
autoland: fix duplicate landing detection logic r=gps