Chris M. Hostetter created SOLR-18436:
-----------------------------------------
Summary: CloudSolrClient has no fallback option for randomly
selected node when sending V2 requests
Key: SOLR-18436
URL: https://issues.apache.org/jira/browse/SOLR-18436
Project: Solr
Issue Type: Sub-task
Reporter: Chris M. Hostetter
(See parent Jira for larger context: SOLR-18433)
When {{CloudSolrClient.sendRequest}} sees that it has a {{V2}} request it picks
a *SINGLE* _random_ node from {{liveNodes}} and uses that as the _solr_
endpoint when constructing a {{LBSolrClient.Request}}
AFAICT this means two things:
# If the node is partitioned from the client, or otherwise unreachable (maybe
it stops being live concurrently with this request) the {{LBSolrClient}} has no
backup option – the request fails w/o ever being sent
# there is no consistency in where V2 requests are sent if/when someone is
using theV2 API to send a collection admin request (meaning that {{V2Request}}
objects are worse then regular {{CollectionAdminRequests}} for users running
w/o an overseer, who want to follow the recommendations of sending all admin
requests to a consistent node)
----
it seems like the V2 request logic in {{CloudSolrClient.sendRequest()}} should
use _ALL_ liveNodes, in the the same (accidentally consistently ordered: see
SOLR-18435) way as the non-V2 non-requiresCollection code path?
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
