Staggered Deploys

[tzmanics]

Add the ability to stagger deploys. Deploy 1/2 the servos, wait some period of time, then deploy the other half. This is intended to provide a quick solution for zero downtime deploys.

I discussed briefly with Jack and the medusa implementation should be pretty straight forward. When it queue's up the deploys (in Kue), it can queue half of them with a delayed start.

Possible UX could be:

$ modulus deploy --staggered=30

30 is the number of seconds to wait. If no time is specified default to 5 seconds.

[theworkflow]

If the servos are expendable, would it be better to just deploy the code into a new servo then destroy the old one once the deploy is completed?

[jackboberg]

I think most of this discussion should be on internal tools. The only bit that needs to be addressed on this repo is adding support for the flags if/when the feature lands.

[cryptoquick]

If we're deployed on Modulus for production, we shouldn't bring down our entire API for minutes every time we re-deploy. This is a critical feature.

[tzmanics]

Hi @cryptoquick we definitely agree and have zero-downtime deployment on our roadmap to be tackled soon. Thanks for reaching out!

[cryptoquick]

That's appreciated, thanks... We just went into production this morning, so... I'm really not sure if I'm comfortable having API downtime on each deployment, especially when working with CI.

[troxler]

@tzmanics Any news on this or maybe an ETA?
I don't like it that my app's users see an error message every time I deploy.

[tzmanics]

Hi @troxler, we have zero-downtime deployment on schedule for release this quarter. We'll keep you updated to when the release happens.

[troxler]

@tzmanics Great! Thanks for the update.