Hi Shamik, Your assumptions on that are correct.
As far as explicit '/8' at query time is concerned, that's the only way the router would get to know that it's a 3-level id and not a 2-level one i.e. e.g. shard.keys='myapp!' Hash range to be fetched: <first-16 bits of murmur hash of myapp>0000 to <first-16 bits of murmur hash of myapp>ffff Intended Hash Range: <8 bits from the hash of myapp>000000 to <8 bits from hash of myapp>ffffff (larger range). If you don't specify a '/8', it will operate as a regular composite id and try to use 16 bits from the 1st component instead of the intended 8 (in case of tri-level). Does that make sense? On Mon, May 19, 2014 at 5:06 PM, Shamik Bandopadhyay <sham...@gmail.com> wrote: > Hi, > > Need some clarification on multilevel composite-id routing in SolrCloud > .I'm currently composite id routing using the following pattern *topic!url* . > This is aimed at run-time de-duplication based on topic field. As I'm > adding support for language search, I felt the need to include language > parameter for better multi-tenancy. Here's the new key structure I'm > thinking of --> *language!topic!url*. > An example would be : english!12345!www.testurl.com > > Now, during query time, I'll always have language parameter at my disposal. > I was thinking,of leveraging the shard.key parameter to specify > *shard.keys=language! > *, which will route the request to the right shard and bring back english > content. Is this a valid assumption ? > > Also, as per my understanding, the three fields will default to 8,8 and 16 > bits of the routing hash. What'll be a valid scenario for providing > a custom allocation of bits for these fields. I was referring to the > following article http://searchhub.org/2014/01/06/10590/ , but was not > entirely sure on this section. > > > *At query time:* > > *To query all records for myapp: shard.keys=myapp/8!* > > *Note the explicit mention of 8 bits in case of querying by component 1 > only i.e. app level. This is required because the usage of the router as 2 > or 3 level isn’t implicit. Specifying ’8′ bits for the component highlights > the use of ’3′ level router*. > Any feedback will be much appreciated. > > Thanks, > Shamik -- Anshum Gupta http://www.anshumgupta.net
