[
https://issues.apache.org/jira/browse/IGNITE-10511?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vyacheslav Koptilin updated IGNITE-10511:
-----------------------------------------
Description:
See attached thread dump:
disco-event-worker hangs on removeExplicitNodeLocks() on GridCacheMapEntry
which is held by GridDistributedTxRemoteAdapter acquired in
GridCacheMapEntry.innerSet().
CacheObjectBinaryProcessorImpl is waiting on metadata message on discovery,
which can be processed due to disco-event-worker is stuck.
Possible fix:
{code:java}
public void onNodeLeft(final ClusterNode node) {
if (isDone() || !enterBusy())
return;
cctx.mvcc().removeExplicitNodeLocks(node.id(), initialVersion());
try {
onDiscoveryEvent(new IgniteRunnable() {
@Override public void run() {
if (isDone() || !enterBusy())
return;
...
}
});
}
finally {
...
}
}
{code}
As we can see most of the processing is done async in IgniteRunnable() in
exchange-worker.
We can move
{code:java}
cctx.mvcc().removeExplicitNodeLocks(node.id(), initialVersion());
{code}
inside this Runnable's body.
was:
See attached thread dump:
disco-event-worker hangs on removeExplicitNodeLocks() on GridCacheMapEntry
which is held by GridDistributedTxRemoteAdapter acquired in
GridCacheMapEntry.innerSet().
CacheObjectBinaryProcessorImpl is waiting on metadata message on discovery,
which can be processed due to disco-event-worker is stuck.
Possible fix:
{quote}public void onNodeLeft(final ClusterNode node) {
if (isDone() || !enterBusy())
return;
cctx.mvcc().removeExplicitNodeLocks(node.id(), initialVersion());
try {
onDiscoveryEvent(new IgniteRunnable() {
@Override public void run() {
if (isDone() || !enterBusy())
return;
...
{quote}
As we can see most of the processing is done async in IgniteRunnable() in
exchange-worker.
We can move
{quote}cctx.mvcc().removeExplicitNodeLocks(node.id(), initialVersion());
{quote}
inside this Runnable's body.
> disco-event-worker can be deadlocked by BinaryContext.metadata running is sys
> striped pool waiting for cache entry lock
> -----------------------------------------------------------------------------------------------------------------------
>
> Key: IGNITE-10511
> URL: https://issues.apache.org/jira/browse/IGNITE-10511
> Project: Ignite
> Issue Type: Bug
> Reporter: Pavel Voronkin
> Priority: Major
> Attachments: race.txt
>
>
> See attached thread dump:
> disco-event-worker hangs on removeExplicitNodeLocks() on GridCacheMapEntry
> which is held by GridDistributedTxRemoteAdapter acquired in
> GridCacheMapEntry.innerSet().
> CacheObjectBinaryProcessorImpl is waiting on metadata message on discovery,
> which can be processed due to disco-event-worker is stuck.
> Possible fix:
> {code:java}
> public void onNodeLeft(final ClusterNode node) {
> if (isDone() || !enterBusy())
> return;
> cctx.mvcc().removeExplicitNodeLocks(node.id(), initialVersion());
> try {
> onDiscoveryEvent(new IgniteRunnable() {
> @Override public void run() {
> if (isDone() || !enterBusy())
> return;
>
> ...
> }
> });
> }
> finally {
> ...
> }
> }
> {code}
> As we can see most of the processing is done async in IgniteRunnable() in
> exchange-worker.
>
> We can move
> {code:java}
> cctx.mvcc().removeExplicitNodeLocks(node.id(), initialVersion());
> {code}
> inside this Runnable's body.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
