Apache
/
lucene
mirror of https://github.com/apache/lucene.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

SOLR-11297: Message 'Lock held by this virtual machine' during startup. Solr is trying to start some cores twice

This commit is contained in:
Erick Erickson 2017-09-25 12:12:31 -07:00
parent feec5c6f09
commit 6391a75a50
2 changed files with 40 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
solr/CHANGES.txt
View File

@ -145,6 +145,9 @@ Bug Fixes
* SOLR-11363: JSON Facet API: repeated values in a numeric points field with docValues enabled
were double counted. (Hossman, yonik)
* SOLR-11297: Message "Lock held by this virtual machine" during startup. Solr is trying to start some cores twice.
(Luiz Armesto, Shawn Heisey, Erick Erickson)
Optimizations
----------------------

40
solr/core/src/java/org/apache/solr/core/CoreContainer.java
View File

@ -637,9 +637,10 @@ public class CoreContainer {
if (zkSys.getZkController() != null) {
zkSys.getZkController().throwErrorIfReplicaReplaced(cd);
}
solrCores.waitAddPendingCoreOps(cd.getName());
core = createFromDescriptor(cd, false, false);
} finally {
solrCores.removeFromPendingOps(cd.getName());
if (asyncSolrCoreLoad) {
solrCores.markCoreAsNotLoading(cd);
}
@ -935,7 +936,13 @@ public class CoreContainer {
// first and clean it up if there's an error.
coresLocator.create(this, cd);
SolrCore core = createFromDescriptor(cd, true, newCollection);
SolrCore core = null;
try {
solrCores.waitAddPendingCoreOps(cd.getName());
core = createFromDescriptor(cd, true, newCollection);
} finally {
solrCores.removeFromPendingOps(cd.getName());
}
return core;
} catch (Exception ex) {
@ -970,7 +977,6 @@ public class CoreContainer {
throw new SolrException(SolrException.ErrorCode.BAD_REQUEST,
"Error CREATEing SolrCore '" + coreName + "': " + ex.getMessage() + rootMsg, ex);
}
}
/**
@ -979,6 +985,26 @@ public class CoreContainer {
* @param dcore a core descriptor
* @param publishState publish core state to the cluster if true
*
* WARNING: Any call to this method should be surrounded by a try/finally block
* that calls solrCores.waitAddPendingCoreOps(...) and solrCores.removeFromPendingOps(...)
*
* <pre>
* <code>
* try {
* solrCores.waitAddPendingCoreOps(dcore.getName());
* createFromDescriptor(...);
* } finally {
* solrCores.removeFromPendingOps(dcore.getName());
* }
* </code>
* </pre>
*
* Trying to put the waitAddPending... in this method results in Bad Things Happening due to race conditions.
* getCore() depends on getting the core returned _if_ it's in the pending list due to some other thread opening it.
* If the core is not in the pending list and not loaded, then getCore() calls this method. Anything that called
* to check if the core was loaded _or_ in pending ops and, based on the return called createFromDescriptor would
* introduce a race condition, see getCore() for the place it would be a problem
*
* @return the newly created core
*/
private SolrCore createFromDescriptor(CoreDescriptor dcore, boolean publishState, boolean newCollection) {
@ -1258,7 +1284,12 @@ public class CoreContainer {
} else {
CoreLoadFailure clf = coreInitFailures.get(name);
if (clf != null) {
try {
solrCores.waitAddPendingCoreOps(clf.cd.getName());
createFromDescriptor(clf.cd, true, false);
} finally {
solrCores.removeFromPendingOps(clf.cd.getName());
}
} else {
throw new SolrException(SolrException.ErrorCode.BAD_REQUEST, "No such core: " + name );
}
@ -1431,7 +1462,8 @@ public class CoreContainer {
// TestLazyCores
if (desc == null || zkSys.getZkController() != null) return null;
// This will put an entry in pending core ops if the core isn't loaded
// This will put an entry in pending core ops if the core isn't loaded. Here's where moving the
// waitAddPendingCoreOps to createFromDescriptor would introduce a race condition.
core = solrCores.waitAddPendingCoreOps(name);
if (isShutDown) return null; // We're quitting, so stop. This needs to be after the wait above since we may come off