1282011 - worker for linux64 in order to run talos

Status: RESOLVED FIXED

(Taskcluster :: Workers, defect)

Description

[Joel Maher ( :jmaher ) (UTC -8)]

after many experiments in buildbot and on hardware (i.e. bug 1269784), it looks as though we need to support hardware in the lab without docker involved.  

My understanding is we need a generic worker created which will run on Linux.  I am not sure if this can easily be done using existing workers/tools (I understand :aki might have a simple generic worker, or we could build off the osx worker possibly).

Once this is done, we can work on creating the talos jobs and splitting the hardware pool between buildbot and taskcluster.

Comment 1

[Aki Sasaki (not active)]

I have a python scriptworker which is designed for security-intensive tasks only.  It does not currently support Windows.

The TC team has genericworker and the upcoming taskcluster-worker which are more likely what you want.

Comment 2

[Pete Moore [:pmoore][:pete]]

We should meet to discuss timeframes. The two options Aki mentions are strong contenders (generic-worker and taskcluster-worker).

taskcluster-worker would be a better long term solution, but some core parts of taskcluster-worker still have yet to be written, thus roll out may take longer.

generic-worker support would be quicker to roll out, since generic-worker already runs on linux, but it currently misses some key features:

    a) clean up between tasks
    b) runs tasks as separate users
    c) install itself as a service
    d) roll out AMIs

So if you can wait for taskcluster-worker, that would be better.

Comment 3

[Joel Maher ( :jmaher ) (UTC -8)]

thanks for the information Pete!  This is a good summary of what features are there.  

I am patient and can wait until later this quarter, but if we cannot get started working on taskcluster-worker come late August, we should figure out generic-worker to get us by until we can build the correct solution.

Comment 4

[Dustin J. Mitchell [:dustin] (he/him)]

Jonas, Wander, Joel, and I are meeting tomorrow to talk about this.

Comment 5

[Chris Cooper [:coop] (he/him)]

We punted on this for now in favor of making progress on Mac, correct?

Comment 6

[Dustin J. Mitchell [:dustin] (he/him)]

Yes, and because we'll want to set this up on Ubuntu 16.04, which will only be configured on the new hardware, which has not been delivered yet, and the datacenter for which is not yet known :)

Comment 7

[Pete Moore [:pmoore][:pete]]

Is this running on taskcluster-worker now? Or is this still needed?

Thanks!

Comment 8

[Joel Maher ( :jmaher ) (UTC -8)]

all jobs are on talos now, lets close this down.  I don't know what "worker" we are using on linux, I really wish taskcluster just had one worker and not so many options.

Comment 9

[Pete Moore [:pmoore][:pete]]

(In reply to Joel Maher ( :jmaher ) (UTC-4) from comment #8)
> I don't know what "worker" we are using on linux, I really wish taskcluster just had one worker and not
> so many options.

Looks like it is taskcluster-worker, e.g. finding a recent task:

https://queue.taskcluster.net/v1/task/X8wJVd27T-mC3bOLdBAFWw

We see the workerType is "gecko-t-linux-talos" and provisionerId is "releng-hardware".

Looking in puppet, we see this is using taskcluster-worker:

https://hg.mozilla.org/build/puppet/file/85fdd4706fb3/manifests/moco-nodes.pp#l24