Thomas Mueller created OAK-11878:
------------------------------------
Summary: ClusterNodeStoreLock LOCK_TIMEOUT is too large
Key: OAK-11878
URL: https://issues.apache.org/jira/browse/OAK-11878
Project: Jackrabbit Oak
Issue Type: Improvement
Reporter: Thomas Mueller
If index-import fails (eg. the process is killed) then async indexing is
blocked for 100 days, unless someone breaks the lock manually.
I think 6 hours is better because:
* Usually a "diff" of 6 hours works fine, but beyond that it gets harder and
harder for the document store.
* The longest index import phase I have seen is 2.5 hours, for a ~2 billion
nodes repository, indexing many indexes in one go.
oak-core
package org.apache.jackrabbit.oak.plugins.index.importer
{noformat}
/**
* Lock implementation for clustered scenario. The locking is done
* by setting the lease time for the lane to distant future which
* prevent AsyncIndexUpdate from running.
*/
public class ClusterNodeStoreLock implements
AsyncIndexerLock<ClusteredLockToken>
/**
* Use a looong lease time to ensure that async indexer does not start
* in between the import process which can take some time
*/
private static final long LOCK_TIMEOUT = TimeUnit.DAYS.toMillis(100);
{noformat}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
