: Maybe this verbiage implies exactly one replica? Though the debug output
: seems to imply otherwise?
:
: > When a Solr node receives a search request, the request is routed behind
: the scenes to a replica of a shard that is part of the collection being
: searched.
:
: https://solr.apache.org/guide/8_4/distributed-requests.html
Only one replica (per shard) gets any given "stage" request.
what you're seeing lower on that page is examples of the older (pre-cloud)
syntax for requesting a request be routed to multiple shards -- using a
COMMA (',') seperated list of "shards" and the ability
ask solr to pick a random replica of each shard by futher specifying a
PIPE ('|') delimiated list of URLs.
basically the idea was in a mutlti shard, multi-replica request, you could
either send the request to a specific list of shards (',') where you had
already picked which replica URL it shoudl send to for each shard, or you
could have solr pick the replicas for you by sending a list ('|') of
replica urls.
...but that's with the 'shards' param indented for users...
The structure of those "lists of lists" was/is used in a various debugging
formats and made it's way into the 'shard.url' param that you are asking
about regarding requests to individual replicas -- which is relic of a
bygone era that existed to support the ShardAugmenterFactory (aka:
fl=[shard])
Full historical analysis in the jira where we removed it in 9.3...
https://issues.apache.org/jira/browse/SOLR-9378
:
: On Tue, Oct 24, 2023 at 4:44 PM Doug Turnbull <douglas.turnb...@reddit.com>
: wrote:
:
: > This may be a silly question, but I can't seem to find an answer. Perhaps
: > just my google-fu is weak.
: >
: > If I query a SolrCloud cluster, with debug=true, In the tracking output, I
: > will see during GET_TOP_IDS a list of N replicas per shard.
: >
: > shards.url=
: >
http://solr-node-1:8983/solr/my_collectiion_shard1_replica_1234|http://solr-node-1:8983/solr/my_collectiion_shard1_replica_5678
: >
: > Does this imply ALL of these replicas are queried, and the first response
: > is aggregated back into the response? Or is EXACTLY ONE replica queried?
: >
: > AND
: >
: > To reduce tail latency - ie waiting for the slowest core - is there a way
: > to control how many replicas per shard are requested?
: >
: > Thanks
: > -Doug
: >
:
-Hoss
http://www.lucidworks.com/
