Re: [Fab-user] Revert deployments

[Christian Vest Hansen]

On Mon, Jan 19, 2009 at 7:03 PM, Miguel A. Lamy <address@hidden> wrote:
> Hi all,
>
> I am evaluating Fabric and a I'm bit lost sorry... I think I'm putting dumb 
> question.
>
> I want to deploy Java Web application into several servers. In the past I 
> used some shell scripts that stoped
> myservers, rsynced my app to the servers, backuped the databases (Mysql, 
> SqlServer and Oracle) and started
> the application server (tipical Jboss and Oracle). I use ssh in all servers 
> with public keys.
>
> I was doing the deployment  server by server and sometimes the process didn't 
> run as expected and I have
> crashs during the startup of the application because of some program errors 
> or inconsistency in the databases.

A tool such as Fabric works best in homogeneous environments, ie. when
all database servers are alike and all application servers are alike.
In this mindset, any inconsistencies in the environment are bugs.

When each server is different, then each server must be handled
differently. This encourages fiddling, and fiddling leads to mistakes,
and mistakes leads to... suffering.... in the form of down time, ahem.
:)

That said, Fabric is a pretty generic remote automation tool. :)

>
> So I thought may be I can use some tool out there that simplify that work and 
> revert my process with no
> pain...
>
> I read some documentation from Fabric site and from the github.com but still 
> don't know how can I run
> myscripts in a way that can "detect" errors in deployment and run some 
> rollback code that reverts errors
> during the deployment.

Fabric cannot undo arbitrary shell commands.

Fabric runs with a fail-fast mentality by default, but this can be
overwriten on most remote operations by setting a `fail` kwarg to
either "ignore" or "warn". When you invoke an operation, the call can
be wrapped with a `try ... except SystemExit` of sorts, and that will
give you a chance to do something when bad stuff happens.

But what I usually do is to just implement a "rollback" command in my
fabfile, and when I see that my deployment have failed, I can then
decide whether to run it or not.

>
> I understand that I can use variables to store command result status and 
> define rollback commands that can be
> executed but my question is... can Fabric "detect" errors and automatically 
> execute rollback functions in
> case of some trouble ?

An "error", in the sense of invoking some remote shell command, is
when that command returns with a code other than 0.

Fabric detects this and acts upon it in accord to the "fail" kwarg for
that particular operation, which is "abort" by default.

Aborting works by throwing SystemExit, and this is something that you
can catch. However, Fabric was not designed with this use case in
mind, to there is no additional information about what failed encoded
in that exception.

>
> I want to deploy several servers and not be worried if one server didn't 
> deployed correctly at 2.00AM and the
> application server didn't start because of that failure. That server should 
> automatically recover from that
> bad deploy, revert to the previous status so in the morning I can analyse the 
> errors and correct that errors
> not worried that users can't run my application because of that failure.

This is possible to do by calling your own "rollback" command in the
except clause mentioned above.

>
> I'd appreciate any help given :-)
>
> (Sorry for my poor english)
>
> Miguel Lamy
>
>
>
> _______________________________________________
> Fab-user mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/fab-user
>



-- 
Venlig hilsen / Kind regards,
Christian Vest Hansen.