Chronicle-Map Lock Contention

[awinstan]

Hi,

We're experimenting with Chronicle-Map and we're experiencing some odd inconsistent behavior on service startup (where Chronicle-Map is instantiated). It appears that most of our "important" threads are experiencing huge contention with the SegmentHeader lock with the following stack trace extract:

"XX-thread-31" #264 daemon prio=5 os_prio=0 cpu=656390.17ms elapsed=1830.23s tid=0x0000ffee18044000 nid=0x1f9 runnable  [0x0000ffedc4842000]
   java.lang.Thread.State: RUNNABLE
	at net.openhft.chronicle.hash.impl.BigSegmentHeader.tryUpdateLockMillis(BigSegmentHeader.java:214)
	- parking to wait for  <0x0000fff526044fa8> (a java.util.concurrent.ForkJoinPool)
	at net.openhft.chronicle.hash.impl.BigSegmentHeader.tryUpdateLock0(BigSegmentHeader.java:178)
	at net.openhft.chronicle.hash.impl.BigSegmentHeader.innerTryUpdateLock(BigSegmentHeader.java:166)
	at net.openhft.chronicle.hash.impl.BigSegmentHeader.updateLock(BigSegmentHeader.java:500)
	at net.openhft.chronicle.map.impl.CompiledMapQueryContext$UpdateLock.lock(CompiledMapQueryContext.java:1122)
	at net.openhft.chronicle.map.MapMethods.put(MapMethods.java:80)
	at net.openhft.chronicle.map.VanillaChronicleMap.put(VanillaChronicleMap.java:910)
        ...

JFR traces confirm virtually all CPU time is burned in BigSegmentHeader.updateLock.

This issue is inconsistent impacting some hosts and not others - see the following CPU graph where post deployment a subset of the hosts have their CPU pegged:

Restarting the process temporarily resolves the issue but it comes back during our next deployment (I'm not sure what triggers this as the deploy process should be virtually identical to my manual process restarts).

Corretto Java 17 on Graviton3 instances. Chronicle-Map Version 3.26ea3

The Chronicle-Map is instantiated as follows:

ChronicleMapBuilder.of(ByteString.class, valueClass)
              .name(name)
              .entries(capacity)
              .averageKeySize(averageKeySizeBytes)
              .averageValueSize(averageValueSizeBytes)
              .keyMarshaller(new ByteStringMarshaller(name))
              .valueMarshaller(valueMarshaller)
              .createPersistedTo(persistenceFile.get())

[JerryShea]

Hello @awinstan - this is not something we have come across before. Investigating this would take a bit of time and we are busy supporting customers and their projects. For us to give this prioritised support can I suggest you review the offerings on https://chronicle.software/support/

[RobertoHuang]

@awinstan @JerryShea Encountering the same problem, may I ask if this has been resolved？