I'll do a little follow up here.
We've managed to start a pgx server and manipulate hbase graph! :D
PGX "Hello World"
We wrote a small code to insert vertices, edgex, instantiate pgx and run a simple example, this is it:
cfg = GraphConfigBuilder.forPropertyGraphHbase().setName('sinapse').setZkQuorum('bda1node05').build()
opg = OraclePropertyGraph.getInstance(cfg)
a = opg.addVertex()
a.setProperty('nome', 'Felipe')
b = opg.addVertex()
b.setProperty('nome', 'Rhenan')
c = opg.addVertex()
c.setProperty('nome', 'Hugo')
opg.addEdge(a, b, 'Pai de')
opg.addEdge(b, c, 'Pai de')
opg.addEdge(a, c, 'Avo de')
opg.commit()
session = Pgx.createSession('sinapsepgx')
analyst = session.createAnalyst()
pgxGraph = session.readGraphWithProperties(opg.getConfig(), true)
analyst.countTriangles(pgxGraph, true)
And that worked just fine!
Client - Server architecture
The next step, we moved to a client/server mode, starting the start-server script.
We managed to do that just fine too!
This is our config files:
server.conf
{
"port": 7007,
"enable_tls": false,
"enable_client_authentication": false
}
pgx.conf
{
"allow_idle_timeout_overwrite": true,
"allow_local_filesystem": false,
"allow_task_timeout_overwrite": true,
"enable_gm_compiler": true,
"enterprise_scheduler_config": {
"analysis_task_config": {
"priority": "MEDIUM",
"weight": 12,
"max_threads": 12
},
"fast_analysis_task_config": {
"priority": "HIGH",
"weight": 1,
"max_threads": 12
},
"num_io_threads_per_task": 12
},
"preload_graphs": [
{"path": "graphs/sinapse_conf.json",
"name": "sinapse"}
],
"max_active_sessions": 1024,
"max_queue_size_per_session": -1,
"max_snapshot_count": 0,
"memory_cleanup_interval": 600,
"path_to_gm_compiler": null,
"release_memory_threshold": 0.85,
"session_idle_timeout_secs": 0,
"session_task_timeout_secs": 0,
"strict_mode": true,
"tmp_dir": "/tmp"
}
sinapse_conf.json
{
"edge_props": [
{
"name": "relacao",
"type": "string"
}
],
"db_engine": "HBASE",
"vertex_props": [
{
"name": "nome",
"type": "string"
},
{
"name": "cpf",
"type": "string"
}
],
"format": "pg",
"name": "sinapse",
"error_handling": {},
"vertex_id_type": "long",
"attributes": {},
"loading": {},
"zk_quorum": "bda1node05,bda1node06,bda1node07"
}
start-script ran just fine with that, preloaded our hbase graph, works like a charm.
Connected to the server using the pgx client:
./bin/pgx -b http://localhost:7007
And managed to do the same we did in the groovy shell.
That's awesome.
PGX on Yarn
Well, now we are back in our challenge: run and manage PGX on Yarn.
We've copied our pgx.conf file to the hdfs, like this:
hdfs://user/pgx/pgx.conf
{
"allow_idle_timeout_overwrite": true,
"allow_local_filesystem": false,
"allow_task_timeout_overwrite": true,
"enable_gm_compiler": true,
"enterprise_scheduler_config": {
"analysis_task_config": {
"priority": "MEDIUM",
"weight": 12,
"max_threads": 12
},
"fast_analysis_task_config": {
"priority": "HIGH",
"weight": 1,
"max_threads": 12
},
"num_io_threads_per_task": 12
},
"preload_graphs": [
{"path": "graphs/sinapse_conf.json",
"name": "sinapse"}
],
"max_active_sessions": 1024,
"max_queue_size_per_session": -1,
"max_snapshot_count": 0,
"memory_cleanup_interval": 600,
"path_to_gm_compiler": null,
"release_memory_threshold": 0.85,
"session_idle_timeout_secs": 0,
"session_task_timeout_secs": 0,
"strict_mode": true,
"tmp_dir": "/tmp"
}
/opt/oracle/oracle-spatial-graph/property_graph/pgx/yarn/conf/yarn.conf
{
"pgx_yarn_jar_hdfs_path": "hdfs://mpmapas-ns/user/pgx/pgx-yarn-2.7.1.jar",
"pgx_war_hdfs_path": "hdfs://mpmapas-ns/user/pgx/pgx-webapp-2.7.1.war",
"pgx_conf_hdfs_path": "hdfs://mpmapas-ns/user/pgx/pgx.conf",
"pgx_log4j_conf_hdfs_path": "hdfs://mpmapas-ns/user/pgx/log4j2.xml",
"pgx_dist_log4j_conf_hdfs_path": "hdfs://mpmapas-ns/user/pgx/dist_log4j.xml",
"pgx_cluster_host_hdfs_path": "hdfs://mpmapas-ns/user/pgx/cluster-host.tgz",
"zookeeper_connect_string": "bda1node05.pgj.rj.gov.br,bda1node06.pgj.rj.gov.br,bda1node07.pgj.rj.gov.br",
"standard_library_path": "/usr/lib64/gcc/4.8.2",
"min_heap_size": "512m",
"max_heap_size": "12g",
"container_cores": 9,
"container_memory": 0,
"container_priority": 0,
"num_machines": 1
}
Also, @albert recomended us to remove the log4j2.xml from the server/shared-mem/pgx-webapp-2.7.1.war file so we may handle log4j logging using only the file placed on our hdfs folder.
So we've unpacked, removed, repacked the war file, edited the log4j2.xml file on hdfs like this:
hdfs://user/pgx/log4j2.xml
<?xml version="1.0" encoding="UTF-8"?>
<Configuration status="WARN">
<Appenders>
<Console name="Console" target="SYSTEM_OUT">
<PatternLayout pattern="%d{HH:mm:ss,SSS} %p %C{1} - %m%n"/>
</Console>
<File name="LogFile" fileName="file:/tmp/pg_trace.log">
<PatternLayout pattern="%d{HH:mm:ss.SSS} [%t] %-5level %logger{36} - %msg%n"/>
</File>
</Appenders>
<Loggers>
<Root level="debug">
<AppenderRef ref="LogFile"/>
</Root>
<Logger name="oracle.pgx.engine.admin.Ctrl" level="debug">
<AppenderRef ref="LogFile"/>
</Logger>
<Logger name="pgx.dist.cluster_host" level="debug">
<AppenderRef ref="LogFile"/>
</Logger>
</Loggers>
</Configuration>
And finally ran the yarn start server command, just like this:
yarn jar yarn/pgx-yarn-2.7.1.jar yarn/conf/yarn.conf
And we get the bottom of the logfile that seems realy nice!:
18/12/11 16:25:03 INFO zookeeper.ZooKeeper: Client environment:java.io.tmpdir=/tmp
18/12/11 16:25:03 INFO zookeeper.ZooKeeper: Client environment:java.compiler=<NA>
18/12/11 16:25:03 INFO zookeeper.ZooKeeper: Client environment:os.name=Linux
18/12/11 16:25:03 INFO zookeeper.ZooKeeper: Client environment:os.arch=amd64
18/12/11 16:25:03 INFO zookeeper.ZooKeeper: Client environment:os.version=4.1.12-124.14.1.el7uek.x86_64
18/12/11 16:25:03 INFO zookeeper.ZooKeeper: Client environment:user.name=root
18/12/11 16:25:03 INFO zookeeper.ZooKeeper: Client environment:user.home=/root
18/12/11 16:25:03 INFO zookeeper.ZooKeeper: Client environment:user.dir=/opt/oracle/oracle-spatial-graph/property_graph/pgx
18/12/11 16:25:03 INFO zookeeper.ZooKeeper: Initiating client connection, connectString=bda1node05.pgj.rj.gov.br,bda1node06.pgj.rj.gov.br,bda1node07.pgj.rj.gov.br sessionTimeout=10000 watcher=oracle.pgx.yarn.ClientZkClient@32da97fd
18/12/11 16:25:03 INFO zookeeper.ClientCnxn: Opening socket connection to server bda1node07.pgj.rj.gov.br/192.168.8.7:2181. Will not attempt to authenticate using SASL (unknown error)
18/12/11 16:25:03 INFO zookeeper.ClientCnxn: Socket connection established, initiating session, client: /192.168.8.5:33299, server: bda1node07.pgj.rj.gov.br/192.168.8.7:2181
18/12/11 16:25:03 INFO zookeeper.ClientCnxn: Session establishment complete on server bda1node07.pgj.rj.gov.br/192.168.8.7:2181, sessionid = 0x3668759ae4553df, negotiated timeout = 10000
18/12/11 16:25:05 INFO yarn.StartService: waiting for PGX service (yarn appId == 'application_1539869144089_2555') to come up ...
18/12/11 16:25:10 INFO yarn.StartService: retrieved PGX host: http://bda1node07.pgj.rj.gov.br:7007
18/12/11 16:25:10 INFO yarn.StartService: to connect a remote shell to this host, run '$PGX_HOME/bin/pgx --base_url http://bda1node07.pgj.rj.gov.br:7007'
18/12/11 16:25:10 INFO yarn.StartService: to shut the PGX service down, run 'yarn application -kill application_1539869144089_2555'
18/12/11 16:25:10 INFO zookeeper.ZooKeeper: Session: 0x3668759ae4553df closed
18/12/11 16:25:10 INFO zookeeper.ClientCnxn: EventThread shut down
But connecting to it still returns 404 ;(
The last intel I may give you is the yarn stderr log, wich also informs that we are not using log4j correctly:
SLF4J: Class path contains multiple SLF4J bindings.
SLF4J: Found binding in [jar:file:/opt/cloudera/parcels/CDH-5.14.2-1.cdh5.14.2.p0.3/jars/slf4j-log4j12-1.7.5.jar!/org/slf4j/impl/StaticLoggerBinder.class]
SLF4J: Found binding in [jar:file:/u09/hadoop/yarn/nm/filecache/890/pgx-yarn-2.7.1.jar!/org/slf4j/impl/StaticLoggerBinder.class]
SLF4J: See http://www.slf4j.org/codes.html#multiple_bindings for an explanation.
SLF4J: Actual binding is of type [org.slf4j.impl.Log4jLoggerFactory]
ERROR StatusLogger No log4j2 configuration file found. Using default configuration: logging only errors to the console. Set system property 'log4j2.debug' to show Log4j2 internal initialization logging.
18/12/11 16:25:06 INFO yarn.AppMaster: register app
18/12/11 16:25:06 INFO yarn.AppMaster: RM response = [queue=root.users.root,maxCap=<memory:65536, vCores:9>]
18/12/11 16:25:06 INFO yarn.AppMaster: max capability of cluster: <memory:65536, vCores:9>
18/12/11 16:25:06 INFO yarn.AppMaster: attempting to allocate 1 containers
18/12/11 16:25:06 INFO yarn.AppMaster: attempt 1: got 0 containers. Available: <memory:194560, vCores:180>
18/12/11 16:25:06 INFO yarn.AppMaster: attempt 2: got 0 containers. Available: <memory:194560, vCores:180>
18/12/11 16:25:06 INFO yarn.AppMaster: attempt 3: got 1 containers. Available: <memory:129024, vCores:171>
18/12/11 16:25:06 INFO yarn.AppMaster: copy hdfs://mpmapas-ns/user/pgx/pgx-yarn-2.7.1.jar into pgx-yarn.jar
18/12/11 16:25:06 INFO yarn.AppMaster: copy hdfs://mpmapas-ns/user/pgx/pgx-webapp-2.7.1.war into pgx-server.war
18/12/11 16:25:06 INFO yarn.AppMaster: copy hdfs://mpmapas-ns/user/pgx/pgx.conf into conf/pgx.conf
18/12/11 16:25:06 INFO yarn.AppMaster: copy hdfs://mpmapas-ns/user/pgx/log4j2.xml into conf/log4j2.xml
18/12/11 16:25:07 INFO yarn.AppMaster: server env = {CLASSPATH=conf:pgx-server/WEB-INF/lib/*:pgx-yarn.jar:$HADOOP_CONF_DIR}
18/12/11 16:25:07 INFO yarn.AppMaster: server command = $JAVA_HOME/bin/java -Xms512m -Xmx12g oracle.pgx.yarn.PgxService bda1node07.pgj.rj.gov.br $PWD/pgx-server.war 7007 bda1node05.pgj.rj.gov.br,bda1node06.pgj.rj.gov.br,bda1node07.pgj.rj.gov.br /pgx-37a121ce-e028-432c-8761-104027126c3b 1><LOG_DIR>/stdout 2><LOG_DIR>/stderr;
18/12/11 16:25:07 INFO yarn.AppMaster: check for completion
18/12/11 16:25:08 INFO yarn.AppMaster: check for completion
18/12/11 16:25:08 INFO yarn.AppMaster: check for completion
18/12/11 16:25:09 INFO yarn.AppMaster: check for completion
18/12/11 16:25:09 INFO yarn.AppMaster: check for completion
18/12/11 16:25:10 INFO yarn.AppMaster: check for completion
18/12/11 16:25:10 INFO yarn.AppMaster: check for completion
18/12/11 16:25:11 INFO yarn.AppMaster: check for completion
18/12/11 16:25:11 INFO yarn.AppMaster: check for completion
18/12/11 16:25:12 INFO yarn.AppMaster: check for completion
.
.
.
This is the farthest we've managed to go.
We can start our work now! That's realy exciting.
Now I know how to properly start a service, preload, insert, manage data, and we will import our existing graph database to it and do some experimentation.
Would be lovely to have this running on Yarn at the production level.
Thank you all for the extreme dedication and attention.
