Large-scale production installations, and another question

mike mike503 at gmail.com
Sun Jun 29 09:09:23 UTC 2008


On 6/29/08, dormando <dormando at rydia.net> wrote:

First off - thanks for the quick reply. On a Saturday night, too!

> It's weighted by a few things:
>
> - Nodes that are down are removed.
> - If IO monitoring is working, the list of paths are sorted, in order of
> least busy nodes.

perfect. this is as intelligent as it can get.

> - If no_verify=0 is set in the get_paths call, paths returned are only
> ones which have a valid HEAD response on the storage node. For
> performance reasons it's generally a good idea to set no_verify=1.
> Unless you cannot return paths that could potentially be
> missing/wrong/broken/etc.

so if no_verify=0, then it issues a HEAD to see if the file exists, if
so, it replies back "yes, i have the file" to the tracker, which
combines those replies from the various storage nodes marked as the
ones that have it, and then sends that to the client?

couldn't there be a process which runs constantly and knocks out nodes
with invalid HEAD replies as being out of date/needing fixed (i'm sure
those tools exist and i just don't have enough experience with
mogilefs yet...)


More information about the mogilefs mailing list