We are seeing multiple cases where an index was deleted > 1 month ago but after a node restart the shards of this index becomes unassigned due to DANGLING_INDEX_IMPORTED.
Actually, the corresponding index uuuid folder isn't getting deleted from a node's data directory.
I have the following questions:
How can I ensure that the index folder is deleted from each of the cluster nodes after an index has been deleted.
Is there an ES recommended way for deleting indices?
Can this happen because an index was not closed before deleting it?
One thing I believe can cause this is if your cluster is not correctly configured and can suffer from split-brain scenarios. Please make sure you have followed these guidelines as this otherwise can/will lead to data loss.
This should be automatic in a correctly configured cluster.
In our current setup, all nodes are dataNodes as well as master eligible.
We've set min_master_nodes to n/2+1 and have our own custom algo to ensure split-brain never occurs.
Is there a way to disable import of dangling indexes?
I am not entirely sure there are no other causes, but the fact that you end up with dangling indicxes in a running cluster could indicate that your custom algorithm is not working properly under all conditions and failure scenarios...
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.