We are using Jenkins,GerritTrigger setup for CI and it will start build for each commit though all commits came from single push. Since all changes are dependent on each other its enough to make a single build with all changes, but I don't see that option in GerritTrigger plugin.
I believe many companies use Jenkins and Gerrit combination and I am curious to know how they are handling these cases.
Example:
If a developer pushes below 4 commits at once to gerrit it will create 4 changes accordingly in gerrit say 1,2,3,4 and it starts 4 builds in jenkins for all commit
git log --oneline
e3dfdwd CommitD
5fgfdgh CommitC
df34dsf CommitB
a23sdr3 CommitA
Here 4 commits as a whole will pass all tests in jenkins but individually they will fail. Now jenkins builds will fail for A, B, C and will succeed for D as it will checkout A,B,C as they are its dependencies.
In this case though Commit-D is successful it can't be merged as its dependencies are not passed in Jenkins.
It seems reasonable from development to expect jenkins verification for each push instead of each commit. But GerritTrigger can run for each commit only.
Question:
Is there a way to inform jenkins to start build only for commit-D as it will have all dependencies C,B,A ??
Or can we start a build for each git push from development instead of commits?
Sorry if I had any info
I found a way to start build only for commit-D.
I have introduced a gerrittrigger job which runs immediately after every commit, this job will not do any clone/build/verification.
It will just do some set of verifications like, check if given change has needed-by change, dependency exists and dependencies on same branch etc
This job will trigger another main job which does real clone, checkout change, build, verification etc only for changes which pass all validations.
So this will start job always for top commit and approve/reject all dependent changes based on job result.
Though it has few limitations, we found this method is suitable for our workflow
Most companies don't use git and gerrit :) Most companies don't even use git, unfortunately. And most of the ones that do, don't use gerrit. I've consulted for dozens of companies: two use git, and neither of them have even heard of gerrit.
I don't think it's possible to get gerrit to think of pushes as if they were commits. Since each commit in a push can be separately reviewed and rejected, each commit has to be considered and built separately. If you don't want it to work that way, gerrit might not be for you.
Instead, you should squash your related commits locally before pushing them to gerrit. This will achieve the desired results.
Related
I have a Webhook for my repository in github which is triggered when there is a push to master branch and then webhook starts the build in Jenkins
At the end of build process, I change the version in the pom and push it to master again which would trigger the webhook again and make a loop of builds in Jenkins.
Is there something I can do about this? What is the better practice for this scenario?
Thank you
When you deliver a new change you can commit with a specific message like ('pom modified from jenkins build'). Then add a rule in Jenkins Git plugin to ignore commits with that message ("Polling ignores commit with certain messages" using exact or a pattern). This way you avoid an infinite loop!
Try and break the loop by pushing the updated pom to a dedicated "release" branch from which the final release is built (and the webhook you mention is not active)
From there, any new development should first merge the latest release (and get the updated pom), before making new commits for the next feature to merge to master.
I recently made the transition from Subversion to Git for all my repos at work. However, with svn we had commit hooks in place so that our Jenkins job would run for whichever branch was checked into. Now, I'm trying to set this up using Gitlab and there appears to only be one place to add a web hook. It looks like any time something is checked into ANY branch, the web hook will run. Meaning if I have a branch_A associated with jenkins_job_A, something could be checked into branch_B and the commit hook for jenkins_job_A will still run. Is there a branch by branch way to configure these web hooks? Or is there some kind of script I can check into each branch that will act as a commit hook? Or (my fear) is this feature not supported in Gitlab?
I guess you set up GitLab to do a post commit request to http://yourserver/jenkins/git/notifyCommit?url=<URL of the Git repository>? In theory this should trigger the polling on all jobs that are configured with that URL, and in the polling step the jobs should decide whether they should build or not. In practice this will unfortunately cause all jobs to fire.
We worked around this issue by moving the Job configuration into a Jenkinsfile and then use a Multibranch Pipeline.
As an alternative you could also install the GitLab plugin for Jenkins and use the Jenkins integration in GitLab. This will allow you to trigger the correct jobs when commits are pushed on a branch. The downside is that it requires a per-job configuration.
I'm trying to add some very basic functionality that exists in every other modern ci product, but which unfortunately seems to be a completely foreign concept in Jenkins land.
I have the github plugin hooked up, and the git plugin set to build the "inverse" of "origin/master", so that pushing any branch except master triggers a build.
The problem is, if there's a flaky test and the build fails there's no way to restart it in jenkins. I added the Naginator plugin but it rebuilds the last branch that ran, not the branch of the build that you clicked "retry" on. Using the Naginator plugin, it seems that I need the git branch or sha to be a real parameter of the build. But, I can't find a way to set the git branch as a parameter of the build when a build gets triggered.
The only thing I can think of is to split it into two builds that link to the same git repo, and have the second one be a parameterized build that the first one triggers with the GIT_COMMIT value as the parameter. Then, retrying the second one with Naginator should retry it on the same SHA. This isn't a good solution though, it sucks to have to configure 2 builds for every one of my builds.
Does anyone know of a good way to accomplish this? I'm hoping I'm just missing something simple.
Unfortunately i'm unfamiliar with this exact setup, however the Git plugin documentation, section Push notification from repository, mentions that in the trigger url, the <commit ID is optional. If set, it will trigger a build immediately, without polling for changes.
If there is a built-in "button" in some plugin to issue this manually from inside jenkins UI i don't know, if not that could be a nice feature request.
So, if there really is no easy option aviable yet, as a workaround you could write yourself some script which builds and calls the url for a given branch + commit ID.
Trigger url format, as found in Git Plugin docs:
curl http://yourserver/git/notifyCommit?url=<URL of the Git repository>&branches=branch1[,branch2]*][&sha1=<commit ID>]
To give some context the question is about GitLab and Jenkins setup.
I know how to setup a web hook, I know how to setup a job to be triggered by the hook. The problem is that I need to have multiple jobs and only a single entry-point (parent job) trigger for them.
The downstream jobs at the same time need to be git repo aware so I have to set repo url for them. This causes them to be triggered independently by the hook and I don't want that as this means that they are triggered twice.
On the other hand if I don't configure repo url on a downstream job and the parent job triggers it, it fails as it is not able to do a checkout.
I may try to hack around with some 'execute shell' build step, I believe it's not a valid way to go. Has anybody a good tip how to solve that?
For the reference here is the GitLab Jenkins plugin documentation according to which:
Plugin will parse the GitLab payload and extract the branch for which
the commit is being pushed and changes made. It will then scan all Git
projects in Jenkins and start the build for those that:
match url of the GitLab repo
match the configured refspec pattern if any
and match committed GitLab branch
I tried playing around with different settings, without a great result though.
For the project you want to get only local triggers, just enable Don't trigger a built on commit notification in the Additional behaviours of git plugin.
(https://github.com/elvanja/jenkins-gitlab-hook-plugin/issues/11#issuecomment-35385032, as you actually have discovered).
But a better solution could be to make your downstream jobs reference the repository locally cloned by main job (not sure if actually possible), so the plugin will never consider them for schedule a build, as the git url don't match.
On .NET projects, I've used TeamCity and Subversion to run pre-checked builds.
For those unfamiliar with pre-checked builds, the idea is that when you commit, a build is run with your changeset against trunk.
If that build passes, then your changeset is applied to trunk. If it doesn't pass, you're notified and can try again.
That way, builds from trunk should always be green, and no-one else on the team is interrupted by build breakage.
My question is: has anyone achieved a similar workflow on a Ruby on Rails project, using Git and Jenkins (a.k.a. Hudson)? If so, could you please share some hints / tips / documentation?
You might want to have a look at a Jenkins/Gerrit combination. Gerrit is a code review system that you can setup so it will not push your commit to master unless your commit is approved. You can set it up so that Jenkins first has to approve it.
We are starting to use it for a C++ project where Jenkins first tries to build the patch, if it succeeds it is pushed to the main repo. This also doesn't stall git on the client side.
http://jenkins-ci.org/
https://wiki.jenkins-ci.org/display/JENKINS/Gerrit+Trigger
To truely do this, you want to use a Git pre-receive hook. An excellent one (which runs PHP tests when code has been pushed) is detailed in: a blog post on codeutopia.net
However, the last sentence of the entry states:
it will cause the git push to be delayed until its completion.
Which is no fun if your test suite takes minutes to run, and seriously annoying when it takes 10 or 20 minutes (because git on the dev's machine will sit there and wait and not let go until the pre-receive hook exits).
Now, you may be able to write some sort of Hudson script to revert the failing commit when a build breaks. A former team experimented with such an idea, but never implemented it.