ids参数表在solr查询中如何工作？ 并能帮助我调试分片问题吗？

Question

We have a large number of solr shards and are trying to setup multi-levels of aggregators. My understanding is that one aggregator should have no more than 200 cores associated with it. Our current plan has our first level of aggregators working on 100 cores each. We will then have another level of aggregators collecting these together. So far this is not working for us and when we query our second level we are getting 500 Internal Server Errors. Digging into this we find our level one aggregators are give NPEs. We've also found there's some translation of the query going on. For instance, we give our level 2 aggregator something like this:

http://l2agghostname:8080/solr/core-00/select?q=*

It sends the following to the level 1:

http://l1agghostname:8080/solr/core-00/select?ids=a6_370573660942_76697809790_0,a7_370573660942_76697809790_4&wt=xml&q=*

That suggests it's receiving the IDs to return, but I'm not sure exactly what that "ids" parameter is supposed to do. If I plug that same query directly in to the level 1 aggregator I get the same error, however if I give it only one doc id, as so:

http://l1agghostname:8080/solr/core-00/select?ids=a6_370573660942_76697809790_0&wt=xml&q=*

Then it will return the information!

This seems very weird but I'm also not sure I should be spending time trying to understand how this ids parameter is working. Am I following a red herring?

PS: http://l1agghostname:8080/solr/core-00/select?q=* does return results as expected.

Answer 1

What we eventually found was that Solr was not able to handle searches like this. Solr would send the initial query through the stack. The aggregators would then collect the IDs and send those back. Solr is then supposed to retrieve the documents using IDs. Once the IDs were returned it would not be able to recall the path to retrieve them all. It thus sent the queries to the wrong machines and errors occurred.