Race condition in AbstractCachingProcessingPipeline causes threads to hang.
---------------------------------------------------------------------------
Key: COCOON-2208
URL: https://issues.apache.org/jira/browse/COCOON-2208
Project: Cocoon
Issue Type: Bug
Components: * Cocoon Core
Affects Versions: 2.1.11
Reporter: Sylvain Wallez
Priority: Blocker
There's a race condition in AbstractCachingProcessingPipeline.waitForLock() :
if "lock" is not null, there is a possibility that the thread that created the
lock releases it before the current thread starts waiting on it (see below).
When this happens, the thread waits forever since no thread will ever call
notify() on the lock.
The very bad effect is that this progressively blocks the entire thread pool of
the servlet engine, which ends up rejecting new incoming connections.
And BTW, calling containsKey() just before get() unneedingly doubles the number
of lookups.
{noformat}
protected boolean waitForLock(Object key) {
if(transientStore != null) {
Object lock = null;
synchronized(transientStore) {
String lockKey = PIPELOCK_PREFIX+key;
if(transientStore.containsKey(lockKey)) {
// cache content is currently being generated, wait for
other thread
lock = transientStore.get(lockKey);
}
}
// START OF RACE CONDITION ZONE
if(lock != null) {
try {
// become owner of monitor
synchronized(lock) {
// END OF RACE CONDITION ZONE
lock.wait();
}
} catch (InterruptedException e) {
if(getLogger().isDebugEnabled()) {
getLogger().debug("Got interrupted
waiting for other pipeline to finish processing, retrying...",e);
}
return false;
}
if(getLogger().isDebugEnabled()) {
getLogger().debug("Other pipeline finished
processing, retrying to get cached response.");
}
return false;
}
}
return true;
}
{noformat}
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
