search

ARTbio / GalaxyKickStart

Ansible playbooks for Galaxy Server deployment
GNU General Public License v3.0
24 stars 22 forks source link

[WP] drop usage of miniconda which is not required anymore for galaxy install #331

Closed drosofff closed 5 years ago

drosofff commented 5 years ago

If this works fine let's do it, but I am not sure how robust Galaxy's mechanism for installing Conda is when running more than 1 galaxy handler process.

Ok I will take a closer eye on the issue. Beyond, as I told you some time ago, I am still considering switching to job handler mules as it was done in the dagoba training in June 2018. Do you approve the move ?

mvdbeek commented 5 years ago

I think it would be better to see the final result. At this point I think it would break cloud deployments and deployments with multiple handlers. In principle splitting up things into smaller roles makes sense, sure.

drosofff commented 5 years ago

@mvdbeek I think this PR is almost ready to merge. There is still an adjustment to make because currently the condition when: nginx_galaxy_location != "" only capture a change from no prefix to prefix and misses a change from prefix to no prefix or even from prefix1 to prefix2. But it should not be too difficult to adapt.

There are 2 points I would like your input on:

On a last note: I have done so many deployment using this branch... I did not see problems from letting galaxy install conda in place of the miniconda role.

mvdbeek commented 5 years ago
  • all these adjustments are tricky for the docker part. I have difficulties to figure out exactly what is the general problem: apparently you need to skip services restart during build or deployment and consequently you need a exec /usr/bin/supervisord -c /etc/supervisor/supervisord.conf --nodaemon in the CMD statement. Is that right ?

Yes to all of this (systemd/services don't work in docker).

What's the problem in fact ?

Can you point me to the difficulties you're having ?

And strangely enough the variable galaxy: proxy: is never mentioned. (but cookie-path yes). This is confusing.

Sorry, which variable is that ?

mvdbeek commented 5 years ago

On a last note: I have done so many deployment using this branch... I did not see problems from letting galaxy install conda in place of the miniconda role.

If you did this with the standard

galaxy_web_processes: 2
galaxy_handler_processes: 4

group vars and a normal, new cloud instance I guess it isn't a problem anymore.

drosofff commented 5 years ago

@mvdbeek

Can you point me to the difficulties you're having ?

Hard to describe I did not really mastered the issue. But in general I would say that there is a pb with supervisor at some point, which translates in a problem for nginx not finding the .pid file in the /var/run directory. still a bit mysterious for me. Overall, the difficulty is to code the playbook with keeping in mind that a restart task in the playbook may conflict with the docker build. Saying it another way, ansible playbook is more tricky when you want to preserve docker image building using the playbook. but you know that, don't you ?

Thanks, I will merge. The next step will be to take a decision on mulling or not mulling with uwsgi. I am still lacking a clear view for this !