Hi Andy,
On 11/27/13 11:58, Andrew Hanushevsky wrote:
> Hi Matevz.
>
> On Wed, 27 Nov 2013, Matevz Tadel wrote:
>
>> Now, the question, if I'd have in my site redirector's config:
>> all.manager meta xrootd-rrdns.cms.org
>> this most likely will only get it to report to one meta manager, not all of
>> them, right? Which means we have to manually list true names of all meta
>> managers on all sites of the (US) federation.
> Please use the following directive:
>
> all.manager meta xrootd-rrdns.cms.org+:<port>
>
> The plus says to use all the addresses assigned to the DNS entry. Otherwise, you
> are right, you will get an unpredictable setup as each mananger will use one
> *random* address. The client does this automatically so you don't have to worry
> about it there.
>
>> "all" and "any" directives don't seem to really matter for meta (or std)
>> managers, only for peer and proxy (although it's not explicit in the docs /
>> configuration syntax). Would it make sense to introduce "rrdns" flag that would
>> cause (periodic / on error?) expansion of the manager host name into a list of
>> addresses?
> See above, we already have a way of handling this. The "all" does have a use
> when you are trying to load balance across all the managers. This can be of
> significant help when you have a lage working set of files lookups as the cache
> entries are equally distributed across all the managers.
We're setting up RR-DNS for US-CMS now ... and I still don't understand what
any/all do when subscribing a manager to a set of meta-managers.
all.manager meta all cmsxrootd.fnal.gov+ 1213
I thought that namespace cache on (meta) managers is only populated with things
that they actively query. Are managers actually pushing their caches upwards to
meta-managers? But then neither any nor all is any good as the caches will still
end up on one meta-manager only and client connections come in from a random one.
I thought this kicks in for client requests that can not be served locally and
have to be redirected upwards.
Please explain! :)
Best,
Matevz
########################################################################
Use REPLY-ALL to reply to list
To unsubscribe from the XROOTD-DEV list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1
|