[Buildbot-commits] [Buildbot] #2399: Pushing multiple changes to a git repository watched by a GitPoller causes redundant builds to be performed.

[Buildbot]

#2399: Pushing multiple changes to a git repository watched by a GitPoller causes
redundant builds to be performed.
-------------------+-------------------------------------------------
Reporter:  juj     |      Owner:
    Type:  defect  |     Status:  new
Priority:  major   |  Milestone:  undecided
 Version:  0.8.7   |   Keywords:  GitPoller, mergeRequests, scheduler
-------------------+-------------------------------------------------
 I've now observed this multiple times, the pattern is as follows:

 1. All builders are idle
 2. I make multiple commits to a local git repository
 3. I push the commits at one go to upstream github repository.
 4. All the builders will initiate a build to a random non-latest commit.
 While they are building, a message 'x pending' will show in the title bar
 of the waterfall page for each builder.
 5. After those builds finish, all builders will initiate a second build to
 the latest pushed commit.

 I do not have any mergeRequests parameters in my config.
 http://buildbot.net/buildbot/docs/latest/manual/cfg-builders.html#merging-
 build-requests states that mergeRequests should be on by default, so I am
 expecting that each builder would only build the latest commit of the
 multiple commits that were pushed upstream.

 An example can be seen here: http://clb.demon.fi:8114/waterfall

 1. At Thu 08 Nov 2012 14:46:48, all builders are idle.
 2. Very shortly after, I push these commits
 https://github.com/realXtend/naali/compare/1c5df8c67547...ecf5a43ee4dbc9ae6e9fee5a5005b2dcf537f5a0
  This happens by fast-forwarding a branch master to be up-to-date with the
 branch incoming. Buildbot waterfall page doesn't show this in changes list
 (perhaps another bug?), but GitPoller does react on this, and at 14:50:30,
 it triggers builds.

 a. The builder osx-Tundra starts building
 c58d8c79e2cea406ee481b3ffa06ec48914d80fa
 b. The builder ubuntu-Tundra starts building
 3dc37d3f240162e2be6ea059d1a2b8244a81bafd
 c. The builder vs2008-Tundra starts building
 220d63a14a776d1a2355bc372c036f3ea61337eb

 Note that all these commits were not even part of the commits that were
 pushed. They seem completely random.

 3. After each of these finish, each builder immediately starts a new
 build:
 a. The builder osx-Tundra starts building
 ecf5a43ee4dbc9ae6e9fee5a5005b2dcf537f5a0
 b. The builder ubuntu-Tundra starts building
 ecf5a43ee4dbc9ae6e9fee5a5005b2dcf537f5a0
 c. The builder vs2008-Tundra starts building
 ecf5a43ee4dbc9ae6e9fee5a5005b2dcf537f5a0

 which is as expected, since that's the latest change that was pushed.

 More examples can be seen in the same waterfall page in the history. At
 13:52:43, multiple commits were also pushed at one go, at which point
 ubuntu-Tundra-incoming decided to do a random first build before building
 the latest commit.

-- 
Ticket URL: <http://trac.buildbot.net/ticket/2399>
Buildbot <http://buildbot.net/>
Buildbot: build/test automation