Boyi Dai created SOLR-18557:
-------------------------------
Summary: REINDEXCOLLECTION abort/failure can leave the source
collection read-only
Key: SOLR-18557
URL: https://issues.apache.org/jira/browse/SOLR-18557
Project: Solr
Issue Type: Bug
Components: SolrCloud
Reporter: Boyi Dai
When REINDEXCOLLECTION is aborted or fails, ReindexCollectionCmd.cleanup() runs
these steps in order:
# delete the target collection
# delete the checkpoint collection
# turn readOnly off on the source collection
# remove the reindexing state
If step 1 or 2 throws, steps 3 and 4 never run. The source collection then
stays read-only, and its reindexing state is left behind.
One way step 1 throws: the target collection is referenced by an alias.
DeleteCollectionCmd then refuses with "Collection : X is part of aliases:
[...], remove or modify the aliases before removing this collection." Two ways
this happens:
* A user creates an alias on the target while the reindex runs.
REINDEXCOLLECTION doesn't take a collection lock, so CREATEALIAS can run
concurrently.
* With the default in-place mode (no explicit target), the command creates the
alias <source> -> <target> itself. A failure after that point hits the same
error during cleanup.
*Steps to reproduce:*
# Start an async REINDEXCOLLECTION of collection A with target B.
# While it is running, CREATEALIAS C -> B.
# Abort the reindex (cmd=abort).
# Result: B is kept (expected, since it is aliased), but A stays readOnly=true
and /collections/A/.reindexing is not removed.
*Expected:* the source collection is always restored, even when the temporary
collections can't be removed.
*Proposed fix:* run the readOnly reset and the reindexing-state removal in a
finally block, and keep rethrowing the delete error so the failure is still
reported. I have a patch with a test and will open a PR.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
