Re: [Cloud] What's my URL?

2020-01-12 Thread Alex Monk
https://tools.wmflabs.org/spi-tools On Sun, 12 Jan 2020, 05:36 Roy Smith, wrote: > I've got a django-based tool that I was previously running in test mode > directly on the bastion hosts using runserver. Now I'm trying to move that > over the kubernetes for production. I've got things to the p

Re: [Cloud] What's my URL?

2020-01-12 Thread Alex Monk
You mean, that you just get the path determined by the name of your tool on the server? The tool creation form on toolsadmin does say "The tool name is used as part of the URL for the tool's webservice." On Sun, 12 Jan 2020, 05:47 Roy Smith, wrote: > Never mind, I found it by trial-and-error.

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Brooke Storm
Hi Nux, I took a look, and I see you have DNA running on Grid Engine. Has it ever run ok on either Kubernetes backend (the old “default” or the new “toolforge”)? Brooke Storm Senior SRE Wikimedia Cloud Services bst...@wikimedia.org IRC: bstorm_ > On Jan 11, 2020, a

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Russell Blau
Between steps 1 and 2, did you insert “webservice stop”? If not, try that! :-) Sent from my iPhone > On Jan 11, 2020, at 5:08 PM, Maciej Jaros wrote: > >  > Hi > > I tried the migration path described here: > https://wikitech.wikimedia.org/wiki/News/2020_Kubernetes_cluster_migration#Manuall

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Count Count
Hi! I don't have much luck with a webservice based on the python3.7 image. It is running fine on the legacy K8s cluster. On the new cluster I got a segfault. After stopping the webservice and trying again to get an empty log the pod is now stuck in ContainerCreating. A few minutes ago: tools.fla

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Alex Monk
Hi Count Count, I'm afraid you seem to have no pods on the new cluster to look at: # kubectl get -n tool-flaggedrevspromotioncheck pod No resources found. Alex On Sun, 12 Jan 2020 at 21:07, Count Count wrote: > Hi! > > I don't have much luck with a webservice based on the python3.7 image. It

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Count Count
Yes, I switched back to the old cluster. This is a new tool that was used in production even if only rarely. I can't leave it offline for hours. I have created a test tool as a copy with which I can reproduce the issue: tools.countcounttest@tools-sgebastion-07:~$ kubectl get pods NAME

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Maciej Jaros
Brooke Storm (2020-01-12 20:53): Hi Nux, I took a look, and I see you have DNA running on Grid Engine.  Has it ever run ok on either Kubernetes backend (the old “default” or the new “toolforge”)? Yes, IIRC Bryan did run it on kubernetes for me (on my request from IRC). That was just before

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Maciej Jaros
Russell Blau (2020-01-12 21:40): Between steps 1 and 2, did you insert “webservice stop”?  If not, try that!  :-) Yes, webserivce was off. And I also did try to turn it off and on again ;-). Few times. I also tried "php7.2" and that didn't work either ¯\_(ツ)_/¯ Sent from my iPhone On Ja

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Yongmin Hong via Cloud
Hi. While I know what 'kubernetes' is, I don't have any idea if any of the tools I maintain depends on this k8s migration, and if yes, why. I simply use `jsub` to submit jobs to be submitted, and sit back and expect it to work (and it does). I have no memory of ever touching anything kubectl-re

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Alex Monk
Thanks Count Count. I have identified a new issue with the new k8s cluster and am looking into it. On Sun, 12 Jan 2020 at 21:43, Count Count wrote: > Yes, I switched back to the old cluster. This is a new tool that was used > in production even if only rarely. I can't leave it offline for hours.

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Alex Monk
Hi Count Count, I believe I may have sorted out an issue that prevented some pods (depending partially on luck) from creating containers. Your pod started and container and it crashed, I see a uwsgi.log file with a python module problem and a uwsgi segfault. On Sun, 12 Jan 2020 at 22:12, Alex Monk

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Count Count
> > Your pod started and container and it crashed, I see a uwsgi.log file with > a python module problem and a uwsgi segfault. > Yes. It was working fine with the legacy cluster. The service ist started via webservice --backend=kubernetes python3.7 start Apparently it cannot load the uwsgi share

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Alex Monk
I think I've seen that particular error that you see in stdout/stderr (via kubectl logs) before - on pods that in fact were working. Meanwhile, uwsgi.log says: Python version: 3.7.3 (default, Apr 3 2019, 05:39:12) [GCC 8.3.0] Set PythonHome to /data/project/countcounttest/www/python/venv Fatal

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Chico Venancio
Maybe a venv created in a different python version? Chico Venancio Em dom, 12 de jan de 2020 20:14, Alex Monk escreveu: > I think I've seen that particular error that you see in stdout/stderr (via > kubectl logs) before - on pods that in fact were working. > > Meanwhile, uwsgi.log says: > > Pyt

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Count Count
> > Maybe a venv created in a different python version? > Hmm, I am using a venv with Python 3.7.6. I can try with 3.7.3 tomorrow, which is used in the image. BTW: No version of Python 3.7 is installed on the dev/bastion hosts afaics. Might be a good idea to sync the version to the one used in th

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Alex Monk
Interesting, uwsgi had Python 3.7.3 but `./www/python/venv/bin/python --version` says 3.7.6. Is that a big enough difference to cause problems? On Sun, 12 Jan 2020 at 23:19, Chico Venancio wrote: > Maybe a venv created in a different python version? > > Chico Venancio > > Em dom, 12 de jan de 20

Re: [Cloud] What's my URL?

2020-01-12 Thread Roy Smith
OK, I found where it's documented: https://wikitech.wikimedia.org/wiki/Portal:Toolforge/Tool_Accounts#What_is_a_Tool_Account? "The ability to run a Web service which is visible at https://tools.wmflabs

Re: [Cloud] [Cloud-announce] [Toolforge] New Kubernetes cluster open for beta testers

2020-01-12 Thread Brooke Storm
You have to create the venv in a container using 'webservice shell of the right runtime'. We support Python versions from Debian Jessie, Stretch and Buster by building in containers, so we cannot sync more than one of those to the bastion. We have moved a lot of Python tools back and forth without