[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: http.debian.net hash sum mismatch




On 02/04/2014 01:53 AM, Rick Thomas wrote:
> 
> On Feb 3, 2014, at 8:37 PM, Scott Ferguson <scott.ferguson.debian.user@gmail.com> wrote:
> 
>> Mirrors were updating a couple of days ago.... and if you tried to use
>> one during the updating period you would get errors. Could be the problem.
> 
> What would it take to make a mirror update atomically?  For example, download all the updates, get everything staged and ready to go but not yet visible to http clients, then at the flip of a switch, have all the updates become visible at once, perhaps with some kind of a "callback" to the currently active clients to tell them that things have changed and they should re-get everything.  Maybe LVM snapshots would be helpful here?

It wouldn't take anything, if the mirror is following the directions on
http://www.debian.org/mirror/ftpmirror

Specifically "MUST perform a 2-stage sync" which is to avoid this very
problem. "Rationale: if archive mirroring is done in a single stage,
there will be periods of time during which the index files will
reference files not yet mirrored."

> 
> It would require some re-thinking of the protocol used by apt-get/aptitude -- to be sure the stuff you just downloaded is still current and hasn't been changed by an update while you were downloading...  and minimize wasted effort by recognizing an update as early as possible.
> 

I politely disagree on this point, this is something well outside of a
package manager's jurisdiction.  It's up to the mirror to say what's
available.

Another thing to look at is if there are any proxy/caching servers
involved that may be serving old versions of the indexes.

> Just a thought...
> 
> Rick
> 

Its a good thought, that's why the maintainers ask mirrors do this this
way. :)

- PaulNM


Reply to: