Failover scripts

[HOF Alerter <ho...@gmail.com>]

Hello World,

I'm preparing to build a two-machine master-slave set-up for Apache SVN
(linux, svnsync mirroring to slave, no cluster). Now, I remember seeing in
the past scripts tailored to this kind of architecture, alas I cannot find
them anymore and not for a lack of googling.

Those included:
- scripts that automagically promote slave to master in case master is down,
- scripts that would edit (replace) post-commit hooks in the above case and
the Apache config.

In case of failover some cleanup is assumed on part of the slave:
unfinished syncs and change of its hooks.Redirecting the users would be
done via loadbalancer that checks for the availability of master the same
way slave does.

Two points on which I am really vague:

- What if the master thinks it is still active and has means to access the
slave. Can we prevent this somehow with hook scripts by blocking the sync
user?
- Switch back after a failover without downtime - possible? Or leave the
slave as master (which makes the excahnge of config and hooks a tad
complicated)?

I could really use your insight and if anyone has experience with this kind
of set-up please shed some light on the subject.

Re: Failover scripts

[Nico Kadel-Garcia <nk...@gmail.com>]

It might be overkill, but what is your time worth? It's fun to build
computers from scratch, and educational. But for a source control
system in a commercial project, it can be much cheaper to simply buy a
well supported package, with automatic updates and good high
availability setups built right in.

I keep going back to the old XKCD cartoon about automation to point
out when it's more efficient to buy a solution:

       https://xkcd.com/1319/

On Fri, Feb 21, 2014 at 6:04 AM, N Maskull <ho...@gmail.com> wrote:
> Wandisco seems like an overkill for this kind of setup.
>
>
> On Fri, Feb 21, 2014 at 11:28 AM, Nico Kadel-Garcia <nk...@gmail.com>
> wrote:
>>
>> On Fri, Feb 21, 2014 at 2:56 AM, HOF Alerter <ho...@gmail.com> wrote:
>> > Hello World,
>> >
>> > I'm preparing to build a two-machine master-slave set-up for Apache SVN
>> > (linux, svnsync mirroring to slave, no cluster). Now, I remember seeing
>> > in
>> > the past scripts tailored to this kind of architecture, alas I cannot
>> > find
>> > them anymore and not for a lack of googling.
>>
>> Save yourself some time. Half a talk with Wandisco about their
>> commercial versions, which have multiple repositories and quorum
>> selection and state reporting built in.
>
>

Re: Failover scripts

[N Maskull <ho...@gmail.com>]

Wandisco seems like an overkill for this kind of setup.


On Fri, Feb 21, 2014 at 11:28 AM, Nico Kadel-Garcia <nk...@gmail.com>wrote:

> On Fri, Feb 21, 2014 at 2:56 AM, HOF Alerter <ho...@gmail.com> wrote:
> > Hello World,
> >
> > I'm preparing to build a two-machine master-slave set-up for Apache SVN
> > (linux, svnsync mirroring to slave, no cluster). Now, I remember seeing
> in
> > the past scripts tailored to this kind of architecture, alas I cannot
> find
> > them anymore and not for a lack of googling.
>
> Save yourself some time. Half a talk with Wandisco about their
> commercial versions, which have multiple repositories and quorum
> selection and state reporting built in.
>

Re: Failover scripts

[Nico Kadel-Garcia <nk...@gmail.com>]

On Fri, Feb 21, 2014 at 2:56 AM, HOF Alerter <ho...@gmail.com> wrote:
> Hello World,
>
> I'm preparing to build a two-machine master-slave set-up for Apache SVN
> (linux, svnsync mirroring to slave, no cluster). Now, I remember seeing in
> the past scripts tailored to this kind of architecture, alas I cannot find
> them anymore and not for a lack of googling.

Save yourself some time. Half a talk with Wandisco about their
commercial versions, which have multiple repositories and quorum
selection and state reporting built in.