On Wed, Sep 24, 2008 at 4:42 PM, Grant Ingersoll <[EMAIL PROTECTED]> wrote: > 1. Can someone describe the stages (I think I know them, but there isn't > much docs on it)?
The original SOLR-303 has the best description I think. 1) get ids of top docs, and merge them 2) get the stored fields from each shard, do highlighting > What if you feel you need a new stage? What would be the > method for adding it? (not saying I need one, but I wouldn't mind writing it > up on the Wiki page above). I think, I would just need to have my mythical > component return a value (in distributedProcess) that is in between the > values of whatever stage I wish to interject in, right? Right, that's why there are big numeric gaps between the existing stages. > The stages seem to > be very query-centric (which makes some sense) It's not a generic distributed computation platform :-) > 2. How does the modifyRequest piece work? I think the idea is that a > component later in the chain can piggyback onto an existing request, right? Both to piggyback, or modify. For example, the highlight component knows enough to turn off highlighting when appropriate. > So, perhaps in looking at my latest patch for S-651, I really should just > piggyback onto the retrieve docs request, right, that is likely coming from > the QueryComponent anyway? Sounds very reasonable. > 3. Anyone have suggestions on testing distributed SearchComponents? Right > now, S-651 is just calling the distributeProcess call directly, but I was > wondering whether we want to think about abstracting a test harness from the > TestDistributedSearch class that starts up multiple Jetty instances. > Thoughts? > > 4. Seems like we should put some effort into making the other components > support distributed operations, right? Is there an existing issue for > this? I see https://issues.apache.org/jira/browse/SOLR-785. Should we > handle it on an ad-hoc basis and just open issues for each one? separate issue for each one. They can be committed independently... no need to couple them. -Yonik
