I don't understand what you are trying to tell me but to sum it up, a rebalancing of primary nodes with ES is not possible. -1 for ES.
So I have to stick with SOLR which can rebalance leaders " /admin/collections?action=REBALANCELEADERS" Regards Bernd Am Montag, 3. November 2014 13:43:11 UTC+1 schrieb Jörg Prante: > > I think the write load issue has not a relation to primary/replica shard > setting. All writes should go to all shards of an index. > > If you have a shard count that is not dividable by the count of nodes, > write load could be skewed, so one or more nodes have more work than > others. > > This can be fixed at index creation time by setting a shard number so that > each node carries the same number of shards for an index. > > Best, > > Jörg > > On Mon, Nov 3, 2014 at 1:32 PM, Bernd Fehling <bernd....@gmail.com > <javascript:>> wrote: > >> Exactly what I see, that one server which has now 2 primaries has higher >> load because of two primaries writing while bulk loading. >> And this is the cause why I want to have a rebalance and the primary back >> to its origin to have a well distributed load. >> So no chance to realize this with ES? >> >> But if the index is generated I get a well distributed system. >> Or is this just luck and generating the index several times I get always >> a different result? >> >> Regards, >> Bernd >> >> Am Montag, 3. November 2014 13:09:53 UTC+1 schrieb Jörg Prante: >>> >>> There is nothing wrong, this is expected behavior. >>> >>> It does not matter where the primaries are. Primaries and replica have >>> exactly the same information about cluster state and do exactly the same >>> amount of work. The only exception is that primary shards do the write >>> operation first before sending them to replicas. >>> >>> So there is no need to "rebalance". >>> >>> Best, >>> >>> Jörg >>> >>> On Mon, Nov 3, 2014 at 1:05 PM, Bernd Fehling <bernd....@gmail.com> >>> wrote: >>> >>>> I have an index with 4 shards and 1 replica so that each server has 1 >>>> primary and 1 replica. >>>> If the system is set up everything is well distributed. >>>> If 1 server goes down the primary is moved to one of the replicas but >>>> if the server comes up >>>> again the system is not rebalanced. The primary says where it is and >>>> does not move >>>> back to its origin server. >>>> Any idea whats wrong, is there a setting for this? >>>> >>>> Regards >>>> Bernd >>>> >>>> -- >>>> You received this message because you are subscribed to the Google >>>> Groups "elasticsearch" group. >>>> To unsubscribe from this group and stop receiving emails from it, send >>>> an email to elasticsearc...@googlegroups.com. >>>> To view this discussion on the web visit https://groups.google.com/d/ >>>> msgid/elasticsearch/e2bb0dad-8540-4f34-87a1-253f984c2d22% >>>> 40googlegroups.com >>>> <https://groups.google.com/d/msgid/elasticsearch/e2bb0dad-8540-4f34-87a1-253f984c2d22%40googlegroups.com?utm_medium=email&utm_source=footer> >>>> . >>>> For more options, visit https://groups.google.com/d/optout. >>>> >>> >>> -- >> You received this message because you are subscribed to the Google Groups >> "elasticsearch" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to elasticsearc...@googlegroups.com <javascript:>. >> To view this discussion on the web visit >> https://groups.google.com/d/msgid/elasticsearch/36b7cba2-7d2c-49c0-8323-944961289f10%40googlegroups.com >> >> <https://groups.google.com/d/msgid/elasticsearch/36b7cba2-7d2c-49c0-8323-944961289f10%40googlegroups.com?utm_medium=email&utm_source=footer> >> . >> >> For more options, visit https://groups.google.com/d/optout. >> > > -- You received this message because you are subscribed to the Google Groups "elasticsearch" group. To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/elasticsearch/4a5ba3bd-7a65-4bb4-a054-e573d587b955%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.