[NFR] - Add documentation on how to upgrade with minimal downtime to end users

[niden]

Tim:

Firstly let me say I am really excited about Phalcon, I cant wait to try it out.

I'm just a little worried about downtime for users when I need to upgrade the Phalcon core or make changes to my application (and I'm sure a lot of users from other php frameworks are too).

It would be great if you outlined strategies for upgrading with minimal downtime. I am guessing the use of load balancers upstream and then manually moving clients between each server to perform the upgrade must be involved... but I'm just guessing. Maybe there are other easier ways.

These are all questions and it would be great to have them clarified before we jump ship.

Thanks again for what looks like an amazing framework.

[niden]

Phalcon (Lead Developer, Phalcon PHP Framework)

Hi Tim, thanks, we don’t have a specific strategy to upgrade phalcon on many servers, in our unique experience with that, we’ve used git to push the changes to remote repositories, but we are not sure if that is a good way, we’ll let open this issue to get help from other members of the community. Thanks

[niden]

Tim

I'm not treating updates to Phalcon any differently than I treat updates to PHP (core) or other extensions. Presumably, if a few moments of down time is going to be an issue you would have at least two servers behind something like HAProxy.

If that's the case, just do it in a cascading action. My plan is to fork Phalcon for the purposes of updating, and pull from the main branch when stuff goes in that I need. At that point, it's a matter of running a script that updates the code on the server, installs the updated module and restarts the web server - one at a time. Puppet or similar would be best to orchestrate that.

A much cleaner way would be to package up your phalcon module in a rpm/deb package and host a repository so you can just use yum/apt to do this without needing to build it on every server using it. I'm not going to do that for a while, because this Phalcon is obviously going to remain a rapidly moving target for quite some time.

Either way, you really need some kind of reverse proxy in place if short interruptions are going to be a deal breaker.

[niden]

David

I don't know if this should be a priority until closer to version 1.0. This is still the glory time where you can continually break the code.