[Dirvish] Expire removes most recent backup?

[Steve Ramage]

Paul Slootman wrote:
>
> If I do "rm -rf / tmp/bla" as root, that would be a disaster too.
> However, I wouldn't want to see the rm command modified to stop me from
> doing so.
>   
This isn't the correct analogy, as f will basically force it to go 
through and not prompt. You are overriding the default behavior in rm. 
Not sure what distro you run, but Ubuntu/Debian will make you type in 
something less "YES I KNOW WHAT I'M DOING" when trying to do something 
like remove the kernel package, or other harmful activity.
> Doing backups is an important task. You *need* to know what you're
> doing. Using appropriate logging and check scripts as can be found in
> the wiki or faq are useful for detecting failed backups, so that you
> won't end up in the situation that the most recent (successful!) backup
> was weeks or months ago. If, on the other hand, your most recent backup
> was yesterday, and dirvish-expire wants to remove that, then you've made
> en error in the expire rules, there's no other way this could happen
> (barring the system time being set forward by days / weeks / months).
>
>   
But in this case there are two failures. One you deleted the file, and 
two dirvish deleted the image it contains.
In the case of the most recent image, the harddrive crashed, and you 
have to wait a week to get a new one. Or in my case, I decommisioned 
some machines and didn't worry about it because I had dirvish running. 
It turns out I lost about a month, of stuff.
> If you don't get fancy with the expiry rules and just go for e.g. a
> plain +1 month, then by design the most recent backup simply can't be
> the first to be removed. And if the backup breaks without you noticing
> it, the most recent one will in fact be the only one that dirvish-expire
> will already refuse to remove.
>
> Perhaps you need to show the list your configuration and the relevant
> lines of the image summary files (where the expiry time is stored).
>
>
> Paul Slootman
> _______________________________________________
> Dirvish mailing list
> Dirvish at dirvish.org
> http://www.dirvish.org/mailman/listinfo/dirvish
>   

I don't think that dirvish should allow by default for the newest image 
to be expired ever, though a new directive could be used. I don't see 
how it could cause problems. dirvish-runall && dirvish-expire probably 
won't work because I believe dirvish-expire will only run if ALL of your 
machines completed successfully. If one of your servers is down, and you 
have some massive logs that run for about a week and are rotated, backed 
up daily. Those logs will start stock pilling VERY fast, on disk because 
your old images are not getting cycled out.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.dirvish.org/pipermail/dirvish/attachments/20080105/8d71e081/attachment.html