[
https://issues.apache.org/jira/browse/SOLR-18144?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jan Høydahl reassigned SOLR-18144:
----------------------------------
Assignee: Eric Pugh
> Regression in Schema Designer and .system collection
> ----------------------------------------------------
>
> Key: SOLR-18144
> URL: https://issues.apache.org/jira/browse/SOLR-18144
> Project: Solr
> Issue Type: Bug
> Components: Schema Designer
> Affects Versions: 9.9, 9.10, 9.10.1
> Reporter: Jan Høydahl
> Assignee: Eric Pugh
> Priority: Major
>
> When using Schema designer in solr 9.9 or 9.10, it complains with an error
> message "Collection not found: .system". We first thought it was related to
> SOLR-15832 but it turns out to be a new bug. Here is an analysis by Claude
> Code which I believe is accurate.
> h2. Root Cause
> The regression was introduced by commit 82c50e4291e (SOLR-16503: Use
> Http2SolrClient in SolrClientCache, SchemaDesigner, cherry-picked to
> branch_9x on Feb 10, 2025).
> h3. What Changed
> SchemaDesignerConfigSetHelper was migrated from using raw HTTP via
> CloudLegacySolrClient.getHttpClient() to using GenericSolrRequest +
> CloudSolrClient.
> Before (CloudLegacySolrClient.getHttpClient().execute(httpPost)):
> - Bypassed CloudSolrClient's ZooKeeper collection validation entirely
> - The raw HTTP POST arrived at the Solr servlet (HttpSolrCall)
> - HttpSolrCall.autoCreateSystemColl() detected: POST + .system not in ZK →
> auto-created the collection
> - getStoredSampleDocs used raw HTTP GET; HTTP 404 was explicitly handled as
> "not found is OK"
> After (request.process(cloudClient, BLOB_STORE_ID)):
> - CloudSolrClient performs a ZooKeeper lookup before sending any HTTP request
> - If .system doesn't exist in ZK → immediately throws
> SolrException("Collection not found: .system")
> - HttpSolrCall.autoCreateSystemColl() is never reached → .system is never
> auto-created
> - getStoredSampleDocs now only catches SolrServerException, not
> SolrException — so "Collection not found" propagates up
> h3. Fix Options
> # Best for branch_9x: Before any blob operation in
> SchemaDesignerConfigSetHelper, check if .system exists and explicitly create
> it (CollectionAdminRequest.createCollection(SYSTEM_COLL, null, 1, 1)). Also
> fix getStoredSampleDocs to catch SolrException for the "not found" case (like
> the old HTTP 404 handling).
> # Also fix getStoredSampleDocs: Add SolrException to the catch block and
> return Collections.emptyList() when the collection doesn't exist — matching
> the old behavior where HTTP 404 was silently treated as "no documents stored.
> # Long-term/main: The main branch already has the proper fix — commit
> 242c1fc4e8a (SOLR-17852) migrated Schema Designer entirely to the FileStore
> API, eliminating .system / BlobStore dependency. This backport would be more
> work but more correct.
> The most targeted fix would be to ensure .system is created lazily in
> SchemaDesignerConfigSetHelper before blob reads/writes, and to also fix
> getStoredSampleDocs to handle the "collection not found" case gracefully.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
