[IGNITE-9840] Possible deadlock on transactional future on client node in case of network problems or long GC pauses

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.6
• Fix Version/s: 2.8
• Component/s: clients
• Assignee: Ilya Lantukh
• Reporter: Andrei Aleksandrov
• Created: 10/Oct/18 14:41
• Updated: 08/Nov/18 10:06
• Resolved: 08/Nov/18 09:58

Description

Steps to reproduce:

1)Start the server node with next timeouts. DefaultTxTimeout should be greater than other:

 

<property name="peerClassLoadingEnabled" value="true"/>
<property name="failureDetectionTimeout" value="60000"/>
<property name="clientFailureDetectionTimeout" value="60000"/>
<property name="networkTimeout" value="60000"/>
<property name="gridName" value="name"/>
<!--Transaction timeout setting-->
<property name="transactionConfiguration">
    <bean class="org.apache.ignite.configuration.TransactionConfiguration">
        <property name="DefaultTxTimeout" value="600000"/>
    </bean>
</property>


<property name="idleConnectionTimeout" value="5000"/>
<property name="connectTimeout" value="5000"/>
<property name="ackTimeout" value="20000"/>

On the server side you should create a cache with next parameters:

 

 

<bean class="org.apache.ignite.configuration.CacheConfiguration">
    <property name="name" value="CACHE"/>
    <property name="cacheMode" value="PARTITIONED"/>
    <property name="atomicityMode" value="TRANSACTIONAL"/>
    <property name="writeSynchronizationMode" value="FULL_SYNC"/>
    <property name="backups" value="1"/>
    <property name="statisticsEnabled" value="true"/>

2)After that start the client with the next code:

IgniteCache<String, Object> cache = ignite.getOrCreateCache("CACHE");

try (Transaction tx = ignite.transactions().txStart()) {

    cache.put("Key", new Object());

    System.out.println("Stop me");

    //here we will get long GC pause on server side

    Thread.sleep(10000);

    // Commit the transaction.
    tx.commitAsync().get();
}

 

On step "Stop me" you should suspend all the thread on the server side to emulate the networking problem or long GC pause on the server side.

Finally, you will face in client node next:

[2018-10-10 16:46:10,157][ERROR][nio-acceptor-tcp-comm-#28%GRIDC1%][root] Critical system error detected. Will be handled accordingly to configured handler [hnd=StopNodeOrHaltFailureHandler [tryStop=false, timeout=0, super=AbstractFailureHandler [ignoredFailureTypes=UnmodifiableSet [SYSTEM_WORKER_BLOCKED]]], failureCtx=FailureContext [type=SYSTEM_WORKER_BLOCKED, err=class o.a.i.IgniteException: GridWorker [name=grid-timeout-worker, igniteInstanceName=GRIDC1, finished=false, heartbeatTs=1539179057570]]]

Also, the similar issue could be reproduced in 2.4. In both cases looks like we have a deadlock during trying to display the TxEntryValueHolder. Looks like this values are already used by the transaction with long DefaultTxTimeout .

java.lang.Thread.State: WAITING
at sun.misc.Unsafe.park(Unsafe.java:-1)
at java.util.concurrent.locks.LockSupport.park(LockSupport.java:304)
at org.apache.ignite.internal.util.future.GridFutureAdapter.get0(GridFutureAdapter.java:177)
at org.apache.ignite.internal.util.future.GridFutureAdapter.get(GridFutureAdapter.java:140)
at org.apache.ignite.internal.processors.cache.binary.CacheObjectBinaryProcessorImpl.metadata0(CacheObjectBinaryProcessorImpl.java:526)
at org.apache.ignite.internal.processors.cache.binary.CacheObjectBinaryProcessorImpl.metadata(CacheObjectBinaryProcessorImpl.java:510)
at org.apache.ignite.internal.processors.cache.binary.CacheObjectBinaryProcessorImpl$2.metadata(CacheObjectBinaryProcessorImpl.java:193)
at org.apache.ignite.internal.binary.BinaryContext.metadata(BinaryContext.java:1265)
at org.apache.ignite.internal.binary.BinaryUtils.type(BinaryUtils.java:2407)
at org.apache.ignite.internal.binary.BinaryObjectImpl.rawType(BinaryObjectImpl.java:302)
at org.apache.ignite.internal.binary.BinaryObjectExImpl.toString(BinaryObjectExImpl.java:205)
at org.apache.ignite.internal.binary.BinaryObjectExImpl.toString(BinaryObjectExImpl.java:186)
at org.apache.ignite.internal.binary.BinaryObjectImpl.toString(BinaryObjectImpl.java:919)
at java.lang.String.valueOf(String.java:2994)
at java.lang.StringBuilder.append(StringBuilder.java:131)
at org.apache.ignite.internal.processors.cache.transactions.TxEntryValueHolder.toString(TxEntryValueHolder.java:161)
...

On the client side, it could be looked like a hanging transaction because we waiting on:

tx.commitAsync().get();

 

Issue Links

• links to: GitHub Pull Request #5268