> 2) At http://www.rectangular.com/kinosearch/wiki/ScalingUp you discuss
> possible cluster arrangements, etc. I'm thinking of the best way to apply
> this idea without buying more hardware (for now). In our clustering
> solution, there is no "master" node, can you foresee any problem having
> *all* nodes the "master" (with the search load being spread around and the
> results being aggregated on whichever node the client hit).
> This topology is used to spread the HTTP (and DB) load around the cluster
> allowing for higher loads and easy scaling (and HA). Having a single
> search master breaks this somewhat. We'll migrate searching onto a
> separate cluster later, but for now I need to make things fit.
Hi Marvin,
Dunno if you missed the above post (or you're playing catchup): any
comments on whether all search nodes can be "masters" for
aggregating/sorting results?
Regards
Henry
_______________________________________________
KinoSearch mailing list
KinoSearch@rectangular.com
http://www.rectangular.com/mailman/listinfo/kinosearch
> possible cluster arrangements, etc. I'm thinking of the best way to apply
> this idea without buying more hardware (for now). In our clustering
> solution, there is no "master" node, can you foresee any problem having
> *all* nodes the "master" (with the search load being spread around and the
> results being aggregated on whichever node the client hit).
> This topology is used to spread the HTTP (and DB) load around the cluster
> allowing for higher loads and easy scaling (and HA). Having a single
> search master breaks this somewhat. We'll migrate searching onto a
> separate cluster later, but for now I need to make things fit.
Hi Marvin,
Dunno if you missed the above post (or you're playing catchup): any
comments on whether all search nodes can be "masters" for
aggregating/sorting results?
Regards
Henry
_______________________________________________
KinoSearch mailing list
KinoSearch@rectangular.com
http://www.rectangular.com/mailman/listinfo/kinosearch