Mozilla.org Testing and Deployment Proposal

[Paul McLanahan]

Hello Webdevs,

Our deployment process for mozilla.org (bedrock) is currently
suboptimal. It is as follows:

1. PR is reviewd and merged to 'master' when ready.
2. Stage env is automatically updated within 10 minutes or so.
3. Changes are tested on stage (www.allizom.org).
4. Approved changes are pushed to prod.

Problems with this situation:

1. We have a dev instance, but it's broken due to neglect, but does
have chief. We'd like it to be the automated one and update from
'master' branch on a schedule.
2. Stage should be the one using chief so that we may also test the
deployment of the changes.
3. We currently just deploy 'master' to prod, but should be tagging releases.

Proposed process:

1. PR is reviewd and merged to 'master' when ready.
2. Dev env is automatically updated within 10 minutes or so.
3. Changes are tested on dev (www-dev.allizom.org).
4. Approved changes are pushed to stage for testing via chief.
5. Once approved on stage, 'master' is tagged (tagging scheme TBD.
Most likely what SUMO does).
6. Tag is pushed to github.
7. Tag is deployed to prod using chief.

Any feedback will be much appreciated. We want to know if you're doing
other things and like it, or definitely if you're doing this and hate
it. If we hear no wailing or gnashing of teeth by tomorrow we'll take
it as tacit approval and begin implementing this.

Thanks!

pmac and your friendly webprod devs.

Paul McLanahan
Mozilla Webdev
--pmac

[Luke Crouch]

MDN process:

1. mdndev reviews and merges PR to master
2. cron updates developer-dev every ~5-10m
3. WebQA pings #mdndev and we push the commit hash to stage via chief
4. WebQA verifies stage
5. We push the same commit hash to prod
6. WebQA verifies prod

So ...

1. We don't actually test on dev :/
2. We don't use tags - just commit hashes

It works pretty well for us though.

-L

> _______________________________________________
> dev-webdev mailing list
> dev-w...@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-webdev

[Schalk Neethling]

So Dev env is automatically updated within 10 minutes or so from master
right? This sounds good +1 from me.

--
Kind Regards,
Schalk Neethling
Front-End Engineer
Mozilla

[Paul McLanahan]

Point taken on tagging. We had a discussion about this in #webprod
before I posted this. I wasn't aware of how useful or easy-to-access
the chief logs were. If that's sufficient for people then that's great
for me and less to do.

Rob:

I think we get a similar file. Some IT magic does it I believe. Good point.

Luke:

That sounds pretty good, and pretty close to what we want minus
tagging. The testing I'd like to see on dev isn't really formal. It'd
just be nice to have a place that's a current reflection of 'master'
so that you can be sure you've not broken things in an unexpected way
before using chief on stage. For us, Dev is also the place where
translations are tested because our new "redirect to en-US for a page
unless the translation is marked active" system is off when DEV=True.
So this env is probably more important for our l10n folks than we
devs.

Will:

Oops! We saw lots of tags and assumed :) Your issue is clear however.

Thanks for the feedback everyone! This is great.

Paul

Paul McLanahan
Mozilla Webdev
--pmac

Paul McLanahan
Mozilla Webdev
--pmac

On Thu, Nov 29, 2012 at 4:18 PM, Luke Crouch <lcr...@mozilla.com> wrote:
> MDN process:
>
> 1. mdndev reviews and merges PR to master
> 2. cron updates developer-dev every ~5-10m
> 3. WebQA pings #mdndev and we push the commit hash to stage via chief
> 4. WebQA verifies stage
> 5. We push the same commit hash to prod
> 6. WebQA verifies prod
>
> So ...
>
> 1. We don't actually test on dev :/
> 2. We don't use tags - just commit hashes
>
> It works pretty well for us though.
>
> -L
>
>

[Paul McLanahan]

On Thu, Nov 29, 2012 at 7:23 PM, Andrei Hajdukewycz <san...@mozilla.com> wrote:
> One question: Are we going to have branches that track the current commit
> hash of staging and/or production, or just do everything purely with commit
> hashes and only have the master branch?

At the moment my thought is that the latter is the way. If you think
using branches for tracking could be useful I'm open to the idea.

Paul