FYI: eu.dl.sourceforge.net distfile corruption

[Matthias Andree]

On Sun, 07 Mar 2004, Oliver Eikemeier wrote:

> >Yes. The sites should be randomized so that the load (and possibly
> >breakage) is spread.
> 
> We have RANDOMIZE_MASTER_SITES for this. This should probably enabled
> by default. Breakage shouldn't be spread, especially it is important
> to be able to exculde a defect or hacked server.

If a re-fetch (after detecting MD5 mismatch) makes sure that a different
server is tried, and reports the original server that the altered file
was downloaded from, this should be isolated quickly.

> >The way it is now, the first site listed bears the major part of
> >ports-induced downloads. This would make the Irish Heanet site suffer
> >most of the SourceForge-related downloads for port installs. I consider
> >this unfair. If the ports system had tried to re-fetch (after broken
> >checksum) from a different site, the download would have succeeded BTW.
> 
> Maybe. On the other hand it may be strange to try twenty mirrors to
> find that the checksum mismatches on all of them.

I recall having read "checksum mismatch, re-fetching one more time".
That's one, not twenty, or is the message misleading?

> >And no, I have nothing to do with Heanet, I'm just worried because they
> >are on top of your list. Besides, there are more European download sites
> >for SF, for instance, Switch and Cesnet.
> 
> Switch and Cesnet don't carry all projects distfiles, I added
> MASTER_SITE_SOURCEFORGE_EXTENDED for them.

Ah, didn't know about that. Thanks for the clarification.

> Generally, the problem you mention are valid, but consider the way the port
> system works and are not specific to sourceforge mirrors. A general solution
> that includes i.e. MASTER_SITE_APACHE and MASTER_SITE_GNU is appreciated.

Of course a general solution is preferred over any special casing - I've
just argued about the currently "interesting" example because it's often
easier to follow then.

-- 
Matthias Andree

Encrypt your mail: my GnuPG key ID is 0x052E7D95