[Build flow plugin] Cannot load balance multiple workflows

[Ty Satrang]

http://i.imgur.com/VEZ1Erc.png

For some reason my flow jobs like to all run on a single node instead of choosing an idle node.

This is a problem for me because I am using the Node and Label parameter plugin to run the jobs on the same VM as the flow job so I can copy the reports and logs to the parent workspace for publishing.

I tried using the Least Load plugin, but I get the same behavior.

I also tried the Throttle Concurrent Builds Plug-in to restrict the jobs per node to 1, to no effect.

I am just realizing this now but, maybe it would be easier to get the files to the parent workspace another way and let the flow jobs all run on whatever single node they want. (maybe? https://github.com/dnozay/build-flow-toolbox-plugin )

[Ginga, Dick]

Its strategy is to try to use the same node it last built on. If you have available executors, it will use one. if you have a label assigned to many slaves, each with one executor, it will do what you want.

 

One way to do what you want is to have each “archive” whatever files you want which puts them onto the Jenkins server and then use “Copy artifacts from another project”

--
You received this message because you are subscribed to the Google Groups "Jenkins Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to jenkinsci-use...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/jenkinsci-users/3d1d0150-3125-4b54-a846-b0a6687022cf%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ty Satrang]

"if you have a label assigned to many slaves, each with one executor, it will do what you want"

My slaves do all share the same label and I tried to "Restrict where this project can be run" and I get the same behavior.

"One way to do what you want is to have each “archive” whatever files you want which puts them onto the Jenkins server and then use “Copy artifacts from another project”"

I suppose this could work, but does it copy the files to the workspace or does it just archive them? I am using a plugin to publish reports from the workspace.

[Ginga, Dick]

Maybe I misunderstand your problem. Let’s keep it simple.

 

The reason the multiple runs of your build flow job run on the same node is because, when selecting a node, it tries to run it one the same node it previously ran on (#2 in the Strategy below). If your nodes had just 1 executor defined, then any consecutive runs would run on different nodes. But with 8 executors, it just runs it there.

 

Scheduling strategy

Some slaves are faster, while others are slow. Some slaves are closer (network wise) to a master, others are far away. So doing a good build distribution is a challenge. Currently, Jenkins employs the following strategy:

1. If a project is configured to stick to one computer, that's always honored.
2. Jenkins tries to build a project on the same computer that it was previously built.
3. Jenkins tries to move long builds to slaves, because the amount of network interaction between a master and a slave tends to be logarithmic to the duration of a build (IOW, even if project A takes twice as long to build as project B, it won't require double network transfer.) So this strategy reduces the network overhead.

If you have interesting ideas (or better yet, implementations), please let me know.

To view this discussion on the web visit https://groups.google.com/d/msgid/jenkinsci-users/0a9cb8a5-d3ef-4f02-8506-0e4b0a986493%40googlegroups.com.

[Ty Satrang]

Thank you, I did see this copy pasta in another thread, and I did try changing the scheduling strategy with a plug-in. It seems that this particular problem is somewhat exclusive to the Build Flow plugin. It doesn't just queue jobs up, as you can see in the image link, it runs the flow jobs all at once on a single node.

[Ginga, Dick]

Can you provide a simple example of your main buildflow job?

To view this discussion on the web visit https://groups.google.com/d/msgid/jenkinsci-users/e787a0c5-7c45-4100-af52-05269ce704d6%40googlegroups.com.

[Ty Satrang]

I am not sure how you want to receive that, but here is a gist of the config that will reproduce the issue. 

Flow: https://gist.github.com/tsatrangtw/73a391aa75b45cb38235
Downsteam: https://gist.github.com/tsatrangtw/4c20c9a85ce63081c160

[Ginga, Dick]

Ok, so your upper job Sample Flow does

 

Build(Sample downstream job, node: name-name)

 

And Sample downstream job does a find grep command

 

And your image represents starting Sample flow job 9 times. Since there are 9 available executors on dvm-qa5, all 9 Sample flow jobs start. I expect that 8 of them pend waiting on Sample downstream job to finish (unless you checked “Run concurrent”)

 

And I think you wanted those 9 jobs to be distributed among other nodes in your group like dvm-qa2 and 1?

 

I think that I don’t know how to get it to do what you want but I would not call it a “problem”. This strategy “Jenkins tries to build a project on the same computer that it was previously built.” Is called “affinity” and it helps a great deal when jobs are pulling source code and/or building. It can greatly speed up the process.

To view this discussion on the web visit https://groups.google.com/d/msgid/jenkinsci-users/aea6321b-8ab0-4311-b405-3dda4f533458%40googlegroups.com.

[Ty Satrang]

I am using Jenkins to run tests for several teams, and I would like to be able to run tests concurrently rather than serially. I have a master job that takes parameters (team, tests, code to pull) that I pass using a build flow. I use Jenkins so that I can have this happen on demand or on a schedule. It appears I may be going about this the wrong way. Do you have any suggestions?

[Stefan Thomasson]

Hi,

If i understood your description, i am wondering if this Isn't the expected behaviour for buildflow as it is a "fly weight" job and therefore do not consume any executors?

To view this discussion on the web visit https://groups.google.com/d/msgid/jenkinsci-users/aea6321b-8ab0-4311-b405-3dda4f533458%40googlegroups.com.

[Ty Satrang]

I think you must be right, my solution was to stop caring where the flow runs, and focus on an alternative method of retrieving the report/log data. I used the Build Flow Toolbox which lets you copy files with the flow:

job = build('downstream job')
toolbox.copyFiles(job.workspace, build.workspace)
//Unrelated but also useful:
out.println job.build.log