You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@ignite.apache.org by Zhenya Stanilovsky <ar...@mail.ru> on 2021/10/13 09:38:00 UTC
Re[4]: apache ignite 2.10.0 heap starvation
Node is going down due to full gc triggering, you can avoid it by [1] but you obtain ALL not only life objects.
Additionally you can try to attach with async prof [2] or probably visual vm
[1] https://stackoverflow.com/questions/23393480/can-heap-dump-be-created-for-analyzing-memory-leak-without-garbage-collection
[2] https://github.com/jvm-profiling-tools/async-profiler
>heap dump generation does not seems to be working.
>whenever I tried to generate the heap dump, node is going down, bit strange,
>what else we could analyze
>On Tue, Oct 12, 2021 at 7:35 PM Zhenya Stanilovsky < arzamas123@mail.ru > wrote:
>>hi, highly likely the problem in your code - cpu usage grow synchronously with heap increasing between 00.00 and 12.00.
>>You need to analyze heap dump, no additional settings will help here.
>>
>>>On the same subject, we have made the changes as suggested
>>>
>>>nodes are running on 8 CORE and 128 GB MEM VMs, i've added the following jvm parameters
>>>
>>>-XX:ParallelGCThreads=4
>>>-XX:ConcGCThreads=2
>>>-XX:MaxGCPauseMillis=200
>>>-XX:InitiatingHeapOccupancyPercent=40
>>>
>>>Not used any of these below, using the default values for all these, which is 8 (as the number of cores)
>>>
>>> <property name="systemThreadPoolSize" value="8"/>
>>> <property name="publicThreadPoolSize" value="8"/>
>>> <property name="queryThreadPoolSize" value="8"/>
>>> <property name="serviceThreadPoolSize" value="8"/>
>>>
>>>I could still see our heap is increasing, but atleast I could see a pattern now (not like earlier which is almost exponential)
>>>
>>>Attaching the screenshots of heap, CPU, GC and start script with all the jvm arguments used.
>>>what do you think I should be changing to run to use heap effectively
>>>
>>>
>>>On Wed, Sep 29, 2021 at 2:35 PM Ibrahim Altun < ibrahim.altun@segmentify.com > wrote:
>>>>after many configuration changes and optimizations, i think i've solved the heap problem.
>>>>
>>>>here are the changes that i applied to the system;
>>>>JVM changes -> https://medium.com/@hoan.nguyen.it/how-did-g1gc-tuning-flags-affect-our-back-end-web-app-c121d38dfe56 helped a lot
>>>>
>>>>nodes are running on 12CORE and 64GB MEM servers, i've added the following jvm parameters
>>>>
>>>>-XX:ParallelGCThreads=6
>>>>-XX:ConcGCThreads=2
>>>>-XX:MaxGCPauseMillis=200
>>>>-XX:InitiatingHeapOccupancyPercent=40
>>>>
>>>>on ignite configuration i've changed all thread pool sizes, which were much more than these;
>>>> <property name="systemThreadPoolSize" value="12"/>
>>>> <property name="publicThreadPoolSize" value="12"/>
>>>> <property name="queryThreadPoolSize" value="12"/>
>>>> <property name="serviceThreadPoolSize" value="12"/>
>>>> <property name="stripedPoolSize" value="12"/>
>>>> <property name="dataStreamerThreadPoolSize" value="12"/>
>>>> <property name="rebalanceThreadPoolSize" value="12"/>
>>>>
>>>>Here is the 16 hours of GC report;
>>>>https://gceasy.io/diamondgc-report.jsp?p=c2hhcmVkLzIwMjEvMDkvMjkvLS1nYy5sb2cuMC5jdXJyZW50LS04LTU4LTMx&channel=WEB
>>>>
>>>>
>>>>
>>>>On 2021/09/27 17:11:21, Ilya Korol < llivezking@gmail.com > wrote:
>>>>> Actually Query interface doesn't define close() method, but QueryCursor
>>>>> does.
>>>>> In your snippets you're using try-with-resource construction for SELECT
>>>>> queries which is good, but when you run MERGE INTO query you would also
>>>>> get an QueryCursor as a result of
>>>>>
>>>>> igniteCacheService.getCache(ID, IgniteCacheType.LABEL).query(insertQuery);
>>>>>
>>>>> so maybe this QueryCursor objects still hold some resources/memory.
>>>>> Javadoc for QueryCursor states that you should always close cursors.
>>>>>
>>>>> To simplify cursor closing there is a cursor.getAll() method that will
>>>>> do this for you under the hood.
>>>>>
>>>>>
>>>>> On 2021/09/13 06:17:21, Ibrahim Altun < i...@segmentify.com > wrote:
>>>>> > Hi Ilya,>
>>>>> >
>>>>> > since this is production environment i could not risk to take heap
>>>>> dump for now, but i will try to convince my superiors to get one and
>>>>> analyze it.>
>>>>> >
>>>>> > Queries are heavily used in our system but aren't they autoclosable
>>>>> objects? do we have to close them anyway?>
>>>>> >
>>>>> > here are some usage examples on our system;>
>>>>> > --insert query is like this; MERGE INTO "ProductLabel" ("productId",
>>>>> "label", "language") VALUES (?, ?, ?)>
>>>>> > igniteCacheService.getCache(ID,
>>>>> IgniteCacheType.LABEL).query(insertQuery);>
>>>>> >
>>>>> > another usage example;>
>>>>> > --sqlFieldsQuery is like this; >
>>>>> > String sql = "SELECT _val FROM \"UserRecord\" WHERE \"email\" IN (?)";>
>>>>> > SqlFieldsQuery sqlFieldsQuery = new SqlFieldsQuery(sql);>
>>>>> > sqlFieldsQuery.setLazy(true);>
>>>>> > sqlFieldsQuery.setArgs(emails.toArray());>
>>>>> >
>>>>> > try (QueryCursor<List<?>> ignored = igniteCacheService.getCache(ID,
>>>>> IgniteCacheType.USER).query(sqlFieldsQuery)) {...}>
>>>>> >
>>>>> >
>>>>> >
>>>>> > On 2021/09/12 20:28:09, Shishkov Ilya < sh...@gmail.com > wrote: >
>>>>> > > Hi, Ibrahim!>
>>>>> > > Have you analyzed the heap dump of the server node JVMs?>
>>>>> > > In case your application executes queries are their cursors closed?>
>>>>> > > >
>>>>> > > пт, 10 сент. 2021 г. в 11:54, Ibrahim Altun < ib...@segmentify.com >:>
>>>>> > > >
>>>>> > > > Igniters any comment on this issue, we are facing huge GC
>>>>> problems on>
>>>>> > > > production environment, please advise.>
>>>>> > > >>
>>>>> > > > On 2021/09/07 14:11:09, Ibrahim Altun < ib...@segmentify.com >>
>>>>> > > > wrote:>
>>>>> > > > > Hi,>
>>>>> > > > >>
>>>>> > > > > totally 400 - 600K reads/writes/updates>
>>>>> > > > > 12core>
>>>>> > > > > 64GB RAM>
>>>>> > > > > no iowait>
>>>>> > > > > 10 nodes>
>>>>> > > > >>
>>>>> > > > > On 2021/09/07 12:51:28, Piotr Jagielski < pj...@touk.pl > wrote:>
>>>>> > > > > > Hi,>
>>>>> > > > > > Can you provide some information on how you use the cluster?
>>>>> How many>
>>>>> > > > reads/writes/updates per second? Also CPU / RAM spec of cluster
>>>>> nodes?>
>>>>> > > > > >>
>>>>> > > > > > We observed full GC / CPU load / OOM killer when loading big
>>>>> amount of>
>>>>> > > > data (15 mln records, data streamer + allowOverwrite=true). We've
>>>>> seen>
>>>>> > > > 200-400k updates per sec on JMX metrics, but load up to 10 on
>>>>> nodes, iowait>
>>>>> > > > to 30%. Our cluster is 3 x 4CPU, 16GB RAM (already upgradingto
>>>>> 8CPU, 32GB>
>>>>> > > > RAM). Ignite 2.10>
>>>>> > > > > >>
>>>>> > > > > > Regards,>
>>>>> > > > > > Piotr>
>>>>> > > > > >>
>>>>> > > > > > On 2021/09/02 08:36:07, Ibrahim Altun < ib...@segmentify.com >>
>>>>> > > > wrote:>
>>>>> > > > > > > After upgrading from 2.7.1 version to 2.10.0 version ignite
>>>>> nodes>
>>>>> > > > facing>
>>>>> > > > > > > huge full GC operations after 24-36 hours after node start.>
>>>>> > > > > > >>
>>>>> > > > > > > We try to increase heap size but no luck, here is the start>
>>>>> > > > configuration>
>>>>> > > > > > > for nodes;>
>>>>> > > > > > >>
>>>>> > > > > > > JVM_OPTS="$JVM_OPTS -Xms12g -Xmx12g -server>
>>>>> > > > > > >>
>>>>> > > >
>>>>> -javaagent:/etc/prometheus/jmx_prometheus_javaagent-0.14.0.jar=8090:/etc/prometheus/jmx.yml>
>>>>>
>>>>> > > > > > > -Dcom.sun.management.jmxremote>
>>>>> > > > > > > -Dcom.sun.management.jmxremote.authenticate=false>
>>>>> > > > > > > -Dcom.sun.management.jmxremote.port=49165>
>>>>> > > > > > > -Dcom.sun.management.jmxremote.host=localhost>
>>>>> > > > > > > -XX:MaxMetaspaceSize=256m -XX:MaxDirectMemorySize=1g>
>>>>> > > > > > > -DIGNITE_SKIP_CONFIGURATION_CONSISTENCY_CHECK=true>
>>>>> > > > > > > -DIGNITE_WAL_MMAP=true
>>>>> -DIGNITE_BPLUS_TREE_LOCK_RETRIES=100000>
>>>>> > > > > > > -Djava.net.preferIPv4Stack=true">
>>>>> > > > > > >>
>>>>> > > > > > > JVM_OPTS="$JVM_OPTS -XX:+AlwaysPreTouch -XX:+UseG1GC>
>>>>> > > > > > > -XX:+ScavengeBeforeFullGC -XX:+DisableExplicitGC>
>>>>> > > > > > > -XX:+UseStringDeduplication
>>>>> -Xloggc:/var/log/apache-ignite/gc.log>
>>>>> > > > > > > -XX:+PrintGCDetails -XX:+PrintGCDateStamps>
>>>>> > > > > > > -XX:+PrintTenuringDistribution -XX:+PrintGCCause>
>>>>> > > > > > > -XX:+UseGCLogFileRotation -XX:NumberOfGCLogFiles=10>
>>>>> > > > > > > -XX:GCLogFileSize=100M">
>>>>> > > > > > >>
>>>>> > > > > > > here is the 80 hours of GC analyize report:>
>>>>> > > > > > >>
>>>>> > > >
>>>>> https://gceasy.io/my-gc-report.jsp?p=c2hhcmVkLzIwMjEvMDgvMzEvLS1nYy5sb2cuMC5jdXJyZW50LnppcC0tNS01MS0yOQ==&channel=WEB >
>>>>>
>>>>> > > > > > >>
>>>>> > > > > > > do we need more heap size or is there a BUG that we need to
>>>>> be aware?>
>>>>> > > > > > >>
>>>>> > > > > > > here is the node configuration:>
>>>>> > > > > > >>
>>>>> > > > > > > <?xml version="1.0" encoding="UTF-8"?>>
>>>>> > > > > > > <beans xmlns=" http://www.springframework.org/schema/beans ">
>>>>> > > > > > > xmlns:xsi=" http://www.w3.org/2001/XMLSchema-instance ">
>>>>> > > > > > > xsi:schemaLocation=">
>>>>> > > > > > > http://www.springframework.org/schema/beans >
>>>>> > > > > > >
>>>>> http://www.springframework.org/schema/beans/spring-beans.xsd ">>
>>>>> > > > > > > <bean id="ignite.cfg">
>>>>> > > > > > > class="org.apache.ignite.configuration.IgniteConfiguration">>
>>>>> > > > > > > <property name="gridLogger">>
>>>>> > > > > > > <bean class="org.apache.ignite.logger.log4j2.Log4J2Logger">>
>>>>> > > > > > > <constructor-arg type="java.lang.String">
>>>>> > > > > > > value="/etc/apache-ignite/ignite-log4j2.xml"/>>
>>>>> > > > > > > </bean>>
>>>>> > > > > > > </property>>
>>>>> > > > > > > <property name="communicationSpi">>
>>>>> > > > > > > <bean>
>>>>> > > >
>>>>> class="org.apache.ignite.spi.communication.tcp.TcpCommunicationSpi">>
>>>>> > > > > > > <property name="usePairedConnections" value="true"/>>
>>>>> > > > > > > </bean>>
>>>>> > > > > > > </property>>
>>>>> > > > > > > <property name="failureDetectionTimeout" value="60000"/>>
>>>>> > > > > > > <property name="systemThreadPoolSize" value="128"/>>
>>>>> > > > > > > <property name="publicThreadPoolSize" value="128"/>>
>>>>> > > > > > > <property name="queryThreadPoolSize" value="128"/>>
>>>>> > > > > > > <property name="serviceThreadPoolSize" value="128"/>>
>>>>> > > > > > > <property name="stripedPoolSize" value="128"/>>
>>>>> > > > > > > <property name="dataStreamerThreadPoolSize" value="4"/>>
>>>>> > > > > > > <property name="rebalanceThreadPoolSize" value="16"/>>
>>>>> > > > > > >>
>>>>> > > > > > > <!-- Explicitly enable peer class loading. -->>
>>>>> > > > > > > <property name="peerClassLoadingEnabled" value="true"/>>
>>>>> > > > > > >>
>>>>> > > > > > > <!-- Enable deploymentSpi,>
>>>>> > > > > > > /usr/share/apache-ignite/libs/segmentify directory will be
>>>>> checked>
>>>>> > > > > > > every 5 seconds for changed files-->>
>>>>> > > > > > > <property name="deploymentSpi">>
>>>>> > > > > > > <bean>
>>>>> > > > class="org.apache.ignite.spi.deployment.uri.UriDeploymentSpi">>
>>>>> > > > > > > <property name="temporaryDirectoryPath">
>>>>> > > > > > > value="/tmp/temp_ignite_libs"/>>
>>>>> > > > > > > <property name="uriList">>
>>>>> > > > > > > <list>>
>>>>> > > > > > >>
>>>>> > > > > > > <va...@localhost>
>>>>> > > > /usr/share/apache-ignite/libs/segmentify/</value>>
>>>>> > > > > > > </list>>
>>>>> > > > > > > </property>>
>>>>> > > > > > > </bean>>
>>>>> > > > > > > </property>>
>>>>> > > > > > >>
>>>>> > > > > > > <property name="cacheConfiguration">>
>>>>> > > > > > > <list>>
>>>>> > > > > > > <!-- Partitioned cache example configuration (Atomic>
>>>>> > > > mode). -->>
>>>>> > > > > > > <bean>
>>>>> > > > class="org.apache.ignite.configuration.CacheConfiguration">>
>>>>> > > > > > > <property name="name" value="default"/>>
>>>>> > > > > > > <property name="atomicityMode" value="ATOMIC"/>>
>>>>> > > > > > > <property name="backups" value="1"/>>
>>>>> > > > > > > </bean>>
>>>>> > > > > > > </list>>
>>>>> > > > > > > </property>>
>>>>> > > > > > >>
>>>>> > > > > > > <!-- Explicitly configure TCP discovery SPI to provide list
>>>>> of>
>>>>> > > > > > > initial nodes. -->>
>>>>> > > > > > > <property name="discoverySpi">>
>>>>> > > > > > > <bean>
>>>>> > > > class="org.apache.ignite.spi.discovery.tcp.TcpDiscoverySpi">>
>>>>> > > > > > > <property name="networkTimeout" value="60000"/>>
>>>>> > > > > > > <property name="ipFinder">>
>>>>> > > > > > > <bean>
>>>>> > > > > > >>
>>>>> > > >
>>>>> class="org.apache.ignite.spi.discovery.tcp.ipfinder.vm.TcpDiscoveryVmIpFinder">>
>>>>>
>>>>> > > > > > > <property name="addresses">>
>>>>> > > > > > > <list>>
>>>>> > > > > > > <!-- THERE ARE 10 NODES -->>
>>>>> > > > > > > </list>>
>>>>> > > > > > > </property>>
>>>>> > > > > > > </bean>>
>>>>> > > > > > > </property>>
>>>>> > > > > > > </bean>>
>>>>> > > > > > > </property>>
>>>>> > > > > > >>
>>>>> > > > > > > <!-- Enabling Apache Ignite native persistence. -->>
>>>>> > > > > > > <property name="dataStorageConfiguration">>
>>>>> > > > > > > <bean>
>>>>> > > > class="org.apache.ignite.configuration.DataStorageConfiguration">>
>>>>> > > > > > > <property name="defaultDataRegionConfiguration">>
>>>>> > > > > > > <bean>
>>>>> > > > > > >
>>>>> class="org.apache.ignite.configuration.DataRegionConfiguration">>
>>>>> > > > > > > <property name="persistenceEnabled">
>>>>> > > > value="true"/>>
>>>>> > > > > > > <property name="checkpointPageBufferSize">
>>>>> > > > > > > value="#{ 2L * 1024 * 1024 * 1024}"/>>
>>>>> > > > > > > <property name="maxSize" value="#{ 40L * 1024 *>
>>>>> > > > > > > 1024 * 1024 }"/>>
>>>>> > > > > > > </bean>>
>>>>> > > > > > > </property>>
>>>>> > > > > > > <property name="storagePath">
>>>>> > > > value="/srv/ignite/persist"/>>
>>>>> > > > > > > <property name="walPath" value="/srv/ignite/wal"/>>
>>>>> > > > > > > <property name="walArchivePath" value="/srv/ignite/wal"/>>
>>>>> > > > > > > <property name="walMode" value="LOG_ONLY"/>>
>>>>> > > > > > > <property name="walSegmentSize" value="#{ 256L * 1024 *>
>>>>> > > > 1024 }"/>>
>>>>> > > > > > > <property name="walFlushFrequency" value="5000"/>>
>>>>> > > > > > > <property name="maxWalArchiveSize" value="#{ 512L * 1024>
>>>>> > > > * 1024 }"/>>
>>>>> > > > > > > <property name="writeThrottlingEnabled" value="true"/>>
>>>>> > > > > > > <property name="checkpointFrequency" value="300000"/>>
>>>>> > > > > > > <property name="checkpointWriteOrder" value="SEQUENTIAL">
>>>>> > > > />>
>>>>> > > > > > > </bean>>
>>>>> > > > > > > </property>>
>>>>> > > > > > > </bean>>
>>>>> > > > > > >>
>>>>> > > > > > >>
>>>>> > > > > > > -->
>>>>> > > > > > > < https://www.segmentify.com/ >İbrahim Halil AltunSenior
>>>>> Software>
>>>>> > > > Engineer+90>
>>>>> > > > > > > 536 3327510 • segmentify.com →
>>>>> < https://www.segmentify.com/ >UK •>
>>>>> > > > Germany •>
>>>>> > > > > > > Turkey < https://www.segmentify.com/ecommerce-growth-show >>
>>>>> > > > > > > < https://www.g2.com/products/segmentify/reviews >>
>>>>> > > > > > >>
>>>>> > > > > >>
>>>>> > > > >>
>>>>> > > >>
>>>>> > > >
>>>>> >
>>>>>
>>>
>>> --
>>>Thanks & Regards,
>>>Naveen Bandaru
>>
>>
>>
>>
>
> --
>Thanks & Regards,
>Naveen Bandaru