[
https://issues.apache.org/jira/browse/SOLR-18466?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Mathew updated SOLR-18466:
--------------------------
Description:
The {{search()}} function in streaming expressions currently ignores \_route\_
for shard selection and queries every active slice of the target collection.
Users with known routing keys therefore incur unnecessary request and response
overhead, which can be significant on large clusters. This change adds support
for \_route\_ param in the streaming expression search function, reusing
{{docCol.getRouter().getSearchSlices()}} which is used for shard selection in
normal distributed search.
*Changes:*
Two files in {{{}solr/solrj-streaming{}}}:
*1. the CloudSolrStream class:*
the existing method:
{code:java}
List<Slice> getSlices(
String collectionName,
CloudSolrClient cloudSolrClient,
boolean checkAlias)
{code}
is kept for existing calls from {{{}TopicStream{}}}, {{{}TextLogitStream{}}},
{{{}FeaturesSelectionStream{}}}, and {{StatementImpl}} (all out of scope for
this change); these are unaffected and continue to receive all active slices
A new overloaded helper method is added:
{code:java}
List<Slice> getSlices(
String collectionName,
CloudSolrClient cloudSolrClient,
boolean checkAlias,
SolrParams solrParams)
{code}
When {{solrParams}} contains a nonempty \_route\_, each collection's slices are
resolved via {{docCol.getRouter().getSearchSlices(routeKeys, params, docCol)}}
and when {{_route_}} is absent (or empty), all active slices are returned
exactly as before.
*2. the TupleStream class:*
{{CloudSolrStream.getSlices(collection, cloudSolrClient, true, solrParams)}} is
called after the request params are assembled and passed to the route aware
{{{}getSlices{}}}.
*Example usages:*
{noformat}
expr=search(collection1,
zkHost="localhost:9983",
qt="/export",
q="*:*",
fl="id,a_s,a_i,a_f",
sort="a_f asc, a_i asc", _route_="routeKeyA!")
{noformat}
{noformat}
expr=search(collection1,
zkHost="localhost:9983",
qt="/export",
q="*:*",
fl="id,a_s,a_i,a_f",
sort="a_f asc, a_i asc", _route_="routeKeyA!,routeKeyB!,routeKeyC!")
{noformat}
*One note on backward compatibility:*
Previously, \_route\_ was ignored during shard selection for streaming
expression {{{}search(){}}}. With this change, nonempty route values are passed
to {{{}docCol.getRouter().getSearchSlices(routeKeys, params, docCol){}}}. For
collections using the implicit router, an invalid \_route\_ _value now causes
the request to fail. Previously, the same request searched all active shards
despite the invalid_ \_route\_. This should be documented in the release/
upgrade notes.
Tests cover single and multiple route keys, composite id routing, missing and
empty route values, and invalid implicit router shard names.
was:
The {{search()}} function in streaming expressions currently ignores \_route\_
for shard selection and queries every active slice of the target collection.
Users with known routing keys therefore incur unnecessary request and response
overhead, which can be significant on large clusters. This change adds support
for \_route\_ param in the streaming expression search function, reusing
{{docCol.getRouter().getSearchSlices()}} which is used for shard selection in
normal distributed search.
*Changes:*
Two files in {{{}solr/solrj-streaming{}}}:
*1. the CloudSolrStream class:*
the existing method:
{code:java}
List<Slice> getSlices(
String collectionName,
CloudSolrClient cloudSolrClient,
boolean checkAlias)
{code}
is kept for existing calls from {{{}TopicStream{}}}, {{{}TextLogitStream{}}},
{{{}FeaturesSelectionStream{}}}, and {{StatementImpl}} (all out of scope for
this change); these are unaffected and continue to receive all active slices
A new overloaded helper method is added:
{code:java}
List<Slice> getSlices(
String collectionName,
CloudSolrClient cloudSolrClient,
boolean checkAlias,
SolrParams solrParams)
{code}
When {{solrParams}} contains a nonempty {_}{{{\{_route_}}}}{_}{_}, each
collection's slices are resolved via
{{docCol.getRouter().getSearchSlices(routeKeys, params, docCol)}} and when
{{_route}}{_} is absent (or empty), all active slices are returned exactly as
before.
*2. the TupleStream class:*
{{CloudSolrStream.getSlices(collection, cloudSolrClient, true, solrParams)}} is
called after the request params are assembled and passed to the route aware
{{{}getSlices{}}}.
*Example usages:*
{noformat}
expr=search(collection1,
zkHost="localhost:9983",
qt="/export",
q="*:*",
fl="id,a_s,a_i,a_f",
sort="a_f asc, a_i asc", _route_="routeKeyA")
{noformat}
{noformat}
expr=search(collection1,
zkHost="localhost:9983",
qt="/export",
q="*:*",
fl="id,a_s,a_i,a_f",
sort="a_f asc, a_i asc", _route_="routeKeyA,routeKeyB,routeKeyC")
{noformat}
*One note on backward compatibility:*
Previously, \_route\_ was ignored during shard selection for streaming
expression {{{}search(){}}}. With this change, nonempty route values are passed
to {{{}docCol.getRouter().getSearchSlices(routeKeys, params, docCol){}}}. For
collections using the implicit router, an invalid \_route\_ {_}value now causes
the request to fail. Previously, the same request searched all active shards
despite the invalid \_route\_{_}. This should be documented in the release/
upgrade notes.
Tests cover single and multiple route keys, composite id routing, missing and
empty route values, and invalid implicit router shard names.
> Support _route_ in the streaming expression search function to prune shard
> fan-out
> ----------------------------------------------------------------------------------
>
> Key: SOLR-18466
> URL: https://issues.apache.org/jira/browse/SOLR-18466
> Project: Solr
> Issue Type: Improvement
> Components: SolrJ, streaming expressions
> Reporter: Mathew
> Priority: Minor
> Labels: pull-request-available
> Time Spent: 10m
> Remaining Estimate: 0h
>
> The {{search()}} function in streaming expressions currently ignores
> \_route\_ for shard selection and queries every active slice of the target
> collection. Users with known routing keys therefore incur unnecessary request
> and response overhead, which can be significant on large clusters. This
> change adds support for \_route\_ param in the streaming expression search
> function, reusing {{docCol.getRouter().getSearchSlices()}} which is used for
> shard selection in normal distributed search.
> *Changes:*
> Two files in {{{}solr/solrj-streaming{}}}:
> *1. the CloudSolrStream class:*
> the existing method:
> {code:java}
> List<Slice> getSlices(
> String collectionName,
> CloudSolrClient cloudSolrClient,
> boolean checkAlias)
> {code}
> is kept for existing calls from {{{}TopicStream{}}}, {{{}TextLogitStream{}}},
> {{{}FeaturesSelectionStream{}}}, and {{StatementImpl}} (all out of scope for
> this change); these are unaffected and continue to receive all active slices
> A new overloaded helper method is added:
> {code:java}
> List<Slice> getSlices(
> String collectionName,
> CloudSolrClient cloudSolrClient,
> boolean checkAlias,
> SolrParams solrParams)
> {code}
> When {{solrParams}} contains a nonempty \_route\_, each collection's slices
> are resolved via {{docCol.getRouter().getSearchSlices(routeKeys, params,
> docCol)}} and when {{_route_}} is absent (or empty), all active slices are
> returned exactly as before.
> *2. the TupleStream class:*
> {{CloudSolrStream.getSlices(collection, cloudSolrClient, true, solrParams)}}
> is called after the request params are assembled and passed to the route
> aware {{{}getSlices{}}}.
> *Example usages:*
> {noformat}
> expr=search(collection1,
> zkHost="localhost:9983",
> qt="/export",
> q="*:*",
> fl="id,a_s,a_i,a_f",
> sort="a_f asc, a_i asc", _route_="routeKeyA!")
> {noformat}
> {noformat}
> expr=search(collection1,
> zkHost="localhost:9983",
> qt="/export",
> q="*:*",
> fl="id,a_s,a_i,a_f",
> sort="a_f asc, a_i asc", _route_="routeKeyA!,routeKeyB!,routeKeyC!")
> {noformat}
> *One note on backward compatibility:*
> Previously, \_route\_ was ignored during shard selection for streaming
> expression {{{}search(){}}}. With this change, nonempty route values are
> passed to {{{}docCol.getRouter().getSearchSlices(routeKeys, params,
> docCol){}}}. For collections using the implicit router, an invalid \_route\_
> _value now causes the request to fail. Previously, the same request searched
> all active shards despite the invalid_ \_route\_. This should be documented
> in the release/ upgrade notes.
> Tests cover single and multiple route keys, composite id routing, missing and
> empty route values, and invalid implicit router shard names.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
