Page MenuHomePhabricator

Migrate SCA cluster to SCB (Jessie and Node 4.2)
Closed, ResolvedPublic

Description

The SCA cluster is currently on trusty, while RESTBase is on Jessie. This situation doubles the efforts needed by our small team to ensure smooth service deployment at each occurrence. Since the grand plan is to (eventually) switch all of WMF's infrastructure to Jessie, the Services team would like to migrate the SCA cluster to Jessie as soon as possible.

Related Objects

Event Timeline

mobrovac raised the priority of this task from to Needs Triage.
mobrovac updated the task description. (Show Details)
mobrovac added projects: Services, acl*sre-team.
mobrovac added subscribers: mobrovac, akosiaris, Joe and 3 others.

I 'd be happy if SCA moves on to Jessie relatively soon. So +1 from me. That being said, we first need to make sure SCA services actually work on Jessie. So in alphabetical order we need to make sure these services work fine:

  • apertium
  • citoid
  • cxserver
  • mathoid
  • zotero

One thing that we'll need is to make all those services to use base::service_unit to manage their services, and write systemd unit files as well as the current upstart jobs.

Alternatively, we could set up two Jessie boxes as scb and then use that for new services / migrate over existing services one by one.

One thing that we'll need is to make all those services to use base::service_unit to manage their services, and write systemd unit files as well as the current upstart jobs.

We can also use a sysvinit script as we do for RESTBase to ensure a smooth transition.

In T96017#1206446, @Joe wrote:

One thing that we'll need is to make all those services to use base::service_unit to manage their services, and write systemd unit files as well as the current upstart jobs.

That should be an easy change in service::node, especially given that we already have all possible init scripts readily standing by.

Gage set Security to None.
mark lowered the priority of this task from High to Medium.Jul 22 2015, 10:45 AM
mark added a subscriber: mark.

We've recently discussed about how to move forward with this, and it will likely involve a new temporary service cluster indeed.

Since it was mentioned in the Ops meeting, let's add it to the task:
xulrunner is only present in Wheezy, starting with 31, Firefox/Iceweasel no longer use/build a separate library of XUL (it's now built in to make faster changes by not caring about reverse deps). So, we'll need build src:xulrunner from Wheezy for jessie-wikimedia.

xulrunner is only present in Wheezy, starting with 31, Firefox/Iceweasel no longer use/build a separate library of XUL (it's now built in to make faster changes by not caring about reverse deps). So, we'll need build src:xulrunner from Wheezy for jessie-wikimedia.

For Zotero, we need the version 29 of XULRunners, cf. https://github.com/zotero/translation-server/issues/10 . Moreover, a special package has been built for our needs - https://github.com/wikimedia/mediawiki-services-zotero-translation-server . @akosiaris packaged these for Trusty.

xulrunner is only present in Wheezy, starting with 31, Firefox/Iceweasel no longer use/build a separate library of XUL (it's now built in to make faster changes by not caring about reverse deps). So, we'll need build src:xulrunner from Wheezy for jessie-wikimedia.

For Zotero, we need the version 29 of XULRunners, cf. https://github.com/zotero/translation-server/issues/10 . Moreover, a special package has been built for our needs - https://github.com/wikimedia/mediawiki-services-zotero-translation-server . @akosiaris packaged these for Trusty.

In production we use version 24esr1 (and not 29, nor 31). See http://apt.wikimedia.org/wikimedia/pool/main/x/xulrunner/. As for the special package built for WMF, it's almost the same as upstream, albeit removing various ilconceived ideas like copying shared objects for xulrunner and I don't remember what else into cwd. That package is now fully javascript (and a .chrome manifest file) making it very easy to upgrade and also decoupled from the version of xulrunner hosted on the "building" machine. Of course all this is just information and mostly moot since xulrunner is nowadays unsupported. We should be targeting at deprecating and removing zotero from our infrastructure ASAP (yesterday would be preferable, but there is so much we can do)

Removed the Blocked-on-Operations tag as all prerequisites are done and we can start migrating services to it at leisure

mobrovac renamed this task from Migrate SCA cluster to Jessie to Migrate SCA cluster to SCB (Jessie and Node 4.2).Jan 15 2016, 11:26 PM
mobrovac claimed this task.
mobrovac edited projects, added Services (done); removed Services.

This has been completed a looong time ago. Closing.