[
https://issues.apache.org/jira/browse/IGNITE-12654?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vyacheslav Koptilin updated IGNITE-12654:
-----------------------------------------
Description:
Example of heap dump:
||Class Name||Shallow Heap||Retained Heap||
|top
org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopologyImpl
@ 0x809f03d0|88|1 381 118 968|
|grp org.apache.ignite.internal.processors.cache.CacheGroupContext @
0x809f04c8|96|1 381 121 912|
|locParts java.util.concurrent.atomic.AtomicReferenceArray @ 0x81656c30|16|1
380 925 496|
|array java.lang.Object[1024] @ 0x81656c40|4 112|1 380 925 480|
|org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition
@ 0xb5f2bcd8|24|318 622 384|
|org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition
@ 0xb5f28d90|96|318 618 624|
|org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition
@ 0xb5ed4ac8|24|318 618 576|
|org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition
@ 0xb5f2e7f8|24|318 618 528|
|grp org.apache.ignite.internal.processors.cache.CacheGroupContext @
0x809f04c8|96|1 381 121 912|
|state org.apache.ignite.internal.util.future.GridFutureAdapter$Node @
0xe8ed4cd0|24|318 618 624|
|val
org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopologyImpl$$Lambda$58
@ 0xe8ed4cb8|24|24|
|arg$1
org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopologyImpl
@ 0x809f03d0|88|1 381 118 968|
The number of {{GridFutureAdapter$Node}}'s and
{{GridDhtPartitionTopologyImpl$$Lambda$58}}'s looks weird. It seems that the
following code is the root cause:
{code:java|title=GridDhtPartitionTopologyImpl.java}
/**
* Finds local partitions which don't belong to affinity and runs eviction
process for such partitions.
*
* @param updateSeq Update sequence.
* @param aff Affinity assignments.
* @return {@code True} if there are local partitions need to be evicted.
*/
private boolean checkEvictions(long updateSeq, AffinityAssignment aff) {
...
// After all rents are finished resend partitions.
if (!rentingFutures.isEmpty()) {
final AtomicInteger rentingPartitions = new
AtomicInteger(rentingFutures.size());
for (IgniteInternalFuture<?> rentingFuture : rentingFutures) {
rentingFuture.listen(f -> {
int remaining = rentingPartitions.decrementAndGet();
if (remaining == 0) {
lock.writeLock().lock();
try {
this.updateSeq.incrementAndGet();
if (log.isDebugEnabled())
log.debug("Partitions have been scheduled to
resend [reason=" +
"Evictions are done [grp=" +
grp.cacheOrGroupName() + "]");
ctx.exchange().scheduleResendPartitions();
}
finally {
lock.writeLock().unlock();
}
}
});
}
}
return hasEvictedPartitions;
}
{code}
was:
Example of heap dump:
||Class Name||Shallow Heap||Retained Heap||
|top
org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopologyImpl
@ 0x809f03d0|88|1 381 118 968|
|grp org.apache.ignite.internal.processors.cache.CacheGroupContext @
0x809f04c8|96|1 381 121 912|
|locParts java.util.concurrent.atomic.AtomicReferenceArray @ 0x81656c30|16|1
380 925 496|
|array java.lang.Object[1024] @ 0x81656c40|4 112|1 380 925 480|
|org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition
@ 0xb5f2bcd8|24|318 622 384|
|org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition
@ 0xb5f28d90|96|318 618 624|
|org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition
@ 0xb5ed4ac8|24|318 618 576|
|org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition
@ 0xb5f2e7f8|24|318 618 528|
|grp org.apache.ignite.internal.processors.cache.CacheGroupContext @
0x809f04c8|96|1 381 121 912|
|state org.apache.ignite.internal.util.future.GridFutureAdapter$Node @
0xe8ed4cd0|24|318 618 624|
|val
org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopologyImpl$$Lambda$58
@ 0xe8ed4cb8|24|24|
|arg$1
org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopologyImpl
@ 0x809f03d0|88|1 381 118 968|
The number of {{GridFutureAdapter$Node}}'s and
{{GridDhtPartitionTopologyImpl$$Lambda$58}}'s looks weird. It seems that the
following code is the root cause:
{code:java|GridDhtPartitionTopologyImpl.java}
/**
* Finds local partitions which don't belong to affinity and runs eviction
process for such partitions.
*
* @param updateSeq Update sequence.
* @param aff Affinity assignments.
* @return {@code True} if there are local partitions need to be evicted.
*/
private boolean checkEvictions(long updateSeq, AffinityAssignment aff) {
...
// After all rents are finished resend partitions.
if (!rentingFutures.isEmpty()) {
final AtomicInteger rentingPartitions = new
AtomicInteger(rentingFutures.size());
for (IgniteInternalFuture<?> rentingFuture : rentingFutures) {
rentingFuture.listen(f -> {
int remaining = rentingPartitions.decrementAndGet();
if (remaining == 0) {
lock.writeLock().lock();
try {
this.updateSeq.incrementAndGet();
if (log.isDebugEnabled())
log.debug("Partitions have been scheduled to
resend [reason=" +
"Evictions are done [grp=" +
grp.cacheOrGroupName() + "]");
ctx.exchange().scheduleResendPartitions();
}
finally {
lock.writeLock().unlock();
}
}
});
}
}
return hasEvictedPartitions;
}
{code}
> Some of rentingFutures in GridDhtPartitionTopologyImpl may accumulate a huge
> number of eviction callbacks
> ---------------------------------------------------------------------------------------------------------
>
> Key: IGNITE-12654
> URL: https://issues.apache.org/jira/browse/IGNITE-12654
> Project: Ignite
> Issue Type: Bug
> Affects Versions: 2.8
> Reporter: Vyacheslav Koptilin
> Assignee: Vyacheslav Koptilin
> Priority: Major
> Fix For: 2.9
>
>
> Example of heap dump:
> ||Class Name||Shallow Heap||Retained Heap||
> |top
> org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopologyImpl
> @ 0x809f03d0|88|1 381 118 968|
> |grp org.apache.ignite.internal.processors.cache.CacheGroupContext @
> 0x809f04c8|96|1 381 121 912|
> |locParts java.util.concurrent.atomic.AtomicReferenceArray @ 0x81656c30|16|1
> 380 925 496|
> |array java.lang.Object[1024] @ 0x81656c40|4 112|1 380 925 480|
> |org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition
> @ 0xb5f2bcd8|24|318 622 384|
> |org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition
> @ 0xb5f28d90|96|318 618 624|
> |org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition
> @ 0xb5ed4ac8|24|318 618 576|
> |org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition
> @ 0xb5f2e7f8|24|318 618 528|
> |grp org.apache.ignite.internal.processors.cache.CacheGroupContext @
> 0x809f04c8|96|1 381 121 912|
> |state org.apache.ignite.internal.util.future.GridFutureAdapter$Node @
> 0xe8ed4cd0|24|318 618 624|
> |val
> org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopologyImpl$$Lambda$58
> @ 0xe8ed4cb8|24|24|
> |arg$1
> org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopologyImpl
> @ 0x809f03d0|88|1 381 118 968|
> The number of {{GridFutureAdapter$Node}}'s and
> {{GridDhtPartitionTopologyImpl$$Lambda$58}}'s looks weird. It seems that the
> following code is the root cause:
> {code:java|title=GridDhtPartitionTopologyImpl.java}
> /**
> * Finds local partitions which don't belong to affinity and runs
> eviction process for such partitions.
> *
> * @param updateSeq Update sequence.
> * @param aff Affinity assignments.
> * @return {@code True} if there are local partitions need to be evicted.
> */
> private boolean checkEvictions(long updateSeq, AffinityAssignment aff) {
> ...
> // After all rents are finished resend partitions.
> if (!rentingFutures.isEmpty()) {
> final AtomicInteger rentingPartitions = new
> AtomicInteger(rentingFutures.size());
> for (IgniteInternalFuture<?> rentingFuture : rentingFutures) {
> rentingFuture.listen(f -> {
> int remaining = rentingPartitions.decrementAndGet();
> if (remaining == 0) {
> lock.writeLock().lock();
> try {
> this.updateSeq.incrementAndGet();
> if (log.isDebugEnabled())
> log.debug("Partitions have been scheduled to
> resend [reason=" +
> "Evictions are done [grp=" +
> grp.cacheOrGroupName() + "]");
> ctx.exchange().scheduleResendPartitions();
> }
> finally {
> lock.writeLock().unlock();
> }
> }
> });
> }
> }
> return hasEvictedPartitions;
> }
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
