Andy

I am not sure I see the connection between assigning query
to a different czar and your comment "if we explode queries
into 1000's of pieces and then try to handle reassembly in
a single place"

I do like the idea of assigning query to a different char.
It'd help not only with balancing load between czars, but
also with things like gracefully taking down a czar.

Let's have this discussion on the qserv mailing list, there
are others that will surely be interested in this topic.

I'm happy to turn that into story/stories/epics. Let's wait
few more days to see where this email discussion (and qserv
hangout discussion?) will take us, then we can readjust the
plan wrt to czar failover.

Thanks
Jacek


On 09/04/2015 12:36 PM, Andrew Hanushevsky wrote:
> Hi Jacek,
>
> Perhaps we need a new story. My last comment on failover pretty much
> said that I don't think it's worth all of the effort from a practical
> stand-point. However, there is a related item that (sort of) involves
> failover. I'm thinking how we could load balance czar's after the fact.
> The idea is that it might be interesting if we explore being able to
> reassign queries to differt czars after the query is started. This would
> give us flexibility on how we distribue the reassembly load. As it
> stands now, the system probably won't scale very well if we explode
> queries into 1000's of pieces and then try to handle reassembly in a
> single place.
>
> Let me know if this thought process is something we should explore.
>
> Andy

########################################################################
Use REPLY-ALL to reply to list

To unsubscribe from the QSERV-L list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=QSERV-L&A=1