Marking this as tested since I was able to follow it successfully once I
figured out *where* to grab the autogenerated password.

Fix some typos in obsolete proposals

See merge request !32

We do not backup repositories in the job anymore, and clarify what is where

We decided against repartitionning the disk at all. To quote the above
ticket:

> the rationale, for the SAN, is especially that we have a lot of
> devices already for the LUNs. a single LUN creates six devices in
> the kernel (3 paths for each controller, so six), which makes
> commands like lsblk and multipath -ll really unreadable quick. by
> having three LUNs per VM, we explode this to eighteen (18!) devices
> per VM, which is really a mess. i even wonder if we might hit the
> number of maximum devices allowed internally...
>
> point is: one LUN per partition per VM is messy. and anyways, it's
> not how "the cloud" works, typically; when you ask for a 100GB VM
> from Amazon, they don't give you the VM partitionned in whatever
> they think is the right chunk, they just give you a f*kn 100GB disk,
> period. in our case, ganeti-instance-debootstrap hacks at this to
> prepend a MBR (and with GPT, we should probably have an EFI
> partition), but whatever: point is it's just one fkn disk.

turns out age() doesn't help at all: it compares the timestamp to the
current day at midnight. just subtracting the timestamps works much
better: it's actually accurate and just works.

also add refer to the wait stuff in the show running queries.

 1. stop the database *before* making changes on the backup server

 2. stop puppet just in case it restarts the services

 3. make backups *before* purging the old package

By doing it this way, we *should* avoid the gaps we were having
before. See also team#40776.

TPA-RFC-34: end of office hours

See merge request !31

spelling, style tweaks

See merge request !30

That way people will immediately see they have to act and when.

Mostly correcting stuff caught by my spell checker.