External checkpoints to S3 on EMR - amazon-s3

I am trying to deploy a production cluster for my Flink program. I am using a standard hadoop-core EMR cluster with Flink 1.3.2 installed, using YARN to run it.
I am trying to configure my RocksDB to write my checkpoints to an S3 bucket. I am trying to go through these docs: https://ci.apache.org/projects/flink/flink-docs-release-1.3/setup/aws.html#set-s3-filesystem. The problem seems to be getting the dependencies working correctly. I receive this error when trying run the program:
java.lang.NoSuchMethodError: org.apache.hadoop.conf.Configuration.addResource(Lorg/apache/hadoop/conf/Configuration;)V
at com.amazon.ws.emr.hadoop.fs.EmrFileSystem.initialize(EmrFileSystem.java:93)
at org.apache.flink.runtime.fs.hdfs.HadoopFileSystem.initialize(HadoopFileSystem.java:328)
at org.apache.flink.core.fs.FileSystem.getUnguardedFileSystem(FileSystem.java:350)
at org.apache.flink.core.fs.FileSystem.get(FileSystem.java:389)
at org.apache.flink.core.fs.Path.getFileSystem(Path.java:293)
at org.apache.flink.runtime.state.filesystem.FsCheckpointStreamFactory.<init>(FsCheckpointStreamFactory.java:99)
at org.apache.flink.runtime.state.filesystem.FsStateBackend.createStreamFactory(FsStateBackend.java:282)
at org.apache.flink.contrib.streaming.state.RocksDBStateBackend.createStreamFactory(RocksDBStateBackend.java:273
I have tried both leaving and adjusting the core-site.xml and leaving it as is. I have tried setting the HADOOP_CLASSPATH to the /usr/lib/hadoop/share that contains(what I assume are) most of the JARs described in the above guide. I tried downloading the hadoop 2.7.2 binaries, and copying over them into the flink/libs directory. All resulting in the same error.
Has anyone successfully gotten Flink being able to write to S3 on EMR?
EDIT: My cluster setup
AWS Portal:
1) EMR -> Create Cluster
2) Advanced Options
3) Release = emr-5.8.0
4) Only select Hadoop 2.7.3
5) Next -> Next -> Next -> Create Cluster ( I do fill out names/keys/etc)
Once the cluster is up I ssh into the Master and do the following:
1 wget http://apache.claz.org/flink/flink-1.3.2/flink-1.3.2-bin-hadoop27-scala_2.11.tgz
2 tar -xzf flink-1.3.2-bin-hadoop27-scala_2.11.tgz
3 cd flink-1.3.2
4 ./bin/yarn-session.sh -n 2 -tm 5120 -s 4 -d
5 Change conf/flink-conf.yaml
6 ./bin/flink run -m yarn-cluster -yn 1 ~/flink-consumer.jar
My conf/flink-conf.yaml I add the following fields:
state.backend: rocksdb
state.backend.fs.checkpointdir: s3:/bucket/location
state.checkpoints.dir: s3:/bucket/location
My program's checkpointing setup:
env.enableCheckpointing(getCheckpointRate,CheckpointingMode.EXACTLY_ONCE)
env.getCheckpointConfig.enableExternalizedCheckpoints(ExternalizedCheckpointCleanup.RETAIN_ON_CANCELLATION)
env.getCheckpointConfig.setMinPauseBetweenCheckpoints(getCheckpointMinPause)
env.getCheckpointConfig.setCheckpointTimeout(getCheckpointTimeout)
env.getCheckpointConfig.setMaxConcurrentCheckpoints(1)
env.setStateBackend(new RocksDBStateBackend("s3://bucket/location", true))
If there are any steps you think I am missing, please let me know

I assume that you installed Flink 1.3.2 on your own on the EMR Yarn cluster, because Amazon does not yet offer Flink 1.3.2, right?
Given that, it seems as if you have a dependency conflict. The method org.apache.hadoop.conf.Configuration.addResource(Lorg/apache/hadoop/conf/Configuration) was only introduced with Hadoop 2.4.0. Therefore I assume that you have deployed a Flink 1.3.2 version which was built with Hadoop 2.3.0. Please deploy a Flink version which was built with the Hadoop version running on EMR. This will most likely solve all dependency conflicts.
Putting the Hadoop dependencies into the lib folder seems to not reliably work because the flink-shaded-hadoop2-uber.jar appears to have precedence in the classpath.

Related

s3distcp fail with "mapreduce_shuffle does not exist"

When I running command below,
s3-dist-cp --src s3://test/9.19 --dest hdfs:///user/hadoop/test
I got a error about auxService.
20/02/03 07:52:13 INFO mapreduce.Job: Task Id : attempt_1580716305878_0001_m_000000_2, Status : FAILED
Container launch failed for container_1580716305878_0001_01_000004 : org.apache.hadoop.yarn.exceptions.InvalidAuxServiceException: The auxService:mapreduce_shuffle does not exist
In many QnA, I found a solution like this
link.
But there is no process for nodemanager.
[hadoop#ip-172-31-37-115 ~]$ initctl list | grep yarn
hadoop-yarn-timelineserver start/running, process 8149
hadoop-yarn-resourcemanager start/running, process 17331
hadoop-yarn-proxyserver start/running, process 8147
My EMR was created by quick menu with emr-5.28.0.
Is there anyone knows about this problem?
Thanks!
I'm sure there's some way to update the configs, but what I did was create a cluster using the 'advanced' setup and chose these software packages:
Ganglia
Hive
Hue
Mahout
Pig
Tez
Spark
Hadoop
(8 in total)
Most of those, except spark, are installed with the default settings (the first radio button for software packages in quick setup). One of these software packages or something related to it is what causes s3-dist-cp to be installed, and I was able to use it with no problems with that setup.

Flink on EMR cannot access S3 bucket from "flink run" command

I'm prototyping the use of AWS EMR for a Flink-based system that we're planning to deploy. My cluster has the following versions:
Release label: emr-5.10.0
Hadoop distribution: Amazon 2.7.3
Applications: Flink 1.3.2
In the documentation provided by Amazon here: Amazon flink documentation
and the documentation from Flink: Apache flink documentation
both mention directly using S3 resources as an integrated file system with the s3://<bucket>/<file> pattern. I have verified that all the correct permissions are set, I can use the AWS CLI to copy S3 resources to the Master node with no problem, but attempting to start a Flink job using a Jar from S3 does not work.
I am executing the following step:
JAR location : command-runner.jar
Main class : None
Arguments : flink run -m yarn-cluster -yid application_1513333002475_0001 s3://mybucket/myapp.jar
Action on failure: Continue
The step always fails with
JAR file does not exist: s3://mybucket/myapp.jar
I have spoken to AWS support, and they suggested having a previous step copy the S3 file to the local Master node and then referencing it with a local path. While this would obviously work, I would rather get the native S3 integration working.
I have also tried using the s3a filesystem and get the same result.
You need to download your jar from s3 to be available in the classpath.
aws s3 cp s3://mybucket/myapp.jar myapp.jar
and then run the run -m yarn-cluster myapp.jar

Cannot connect Impala-Kudu to Apache Kudu (without Cloudera Manager): Get TTransportException Error

I have successfully installed kudu on Ubuntu (Trusty) as per the official kudu documentations (see http://kudu.apache.org/docs/installation.html ). The setup has one node running master and tablet server and another node running the tablet server only. I am having issues installing impala-kudu without Cloudera Manager on the node running kudu master. I have followed CDH installation instructions on this (see http://www.cloudera.com/documentation/enterprise/latest/topics/cdh_ig_cdh5_install.html ) page until Step 3. I have avoided installing CDH with YARN and MRv1 as I don’t need to run any mapreduce jobs and will not be using hadoop. Impala-kudu and impala-kudu-shell installed without errors. When I launch the impala-shell it returns:
Starting Impala Shell without Kerberos authentication
Error connecting: TTransportException, Could not connect to kudu_test:21000
***********************************************************************************
Welcome to the Impala shell. Copyright (c) 2015 Cloudera, Inc. All rights reserved.
(Impala Shell v2.7.0-cdh5-IMPALA_KUDU-cdh5 (48f1ad3) built on Thu Aug 18 12:15:44 PDT 2016)Want to know what version of Impala you're connected to? Run the VERSION command to
find out!
***********************************************************************************
[Not connected] >
I have tried to use the CONNECT option to connect to the kudu-master node without success. Both imapala-kudu and kudu are running on the same machine. Are there additional configuration settings which need to be changed or is hadoop and YARN a strict requirement to make impala-kudu work?
After running ps -ef | grep -i impalad I can confirm the impala daemon is not running. After navigating to the impala logs at ~/var/log/impala I find a few errors and warning files. Here is the output of impalad.ERROR:
Log file created at: 2016/09/13 13:26:24
Running on machine: kudu_test
Log line format: [IWEF]mmdd hh:mm:ss.uuuuuu threadid file:line] msg
E0913 13:26:24.084389 3021 logging.cc:118] stderr will be logged to this file.
E0913 13:26:25.406966 3021 impala-server.cc:249] Currently configured default filesystem: LocalFileSystem. fs.defaultFS (file:///) is not supported.ERROR: block location tracking is not properly enabled because
- dfs.datanode.hdfs-blocks-metadata.enabled is not enabled.
- dfs.client.file-block-storage-locations.timeout.millis is too low. It should be at least 10 seconds.
E0913 13:26:25.406990 3021 impala-server.cc:252] Aborting Impala Server startup due to improper configuration. Impalad exiting.
Maybe I need to revisit HDFS and the Hive Metastore to ensure I have these services configured properly?
According to the log, impalad quits because the default filesystem is configured to be LocalFileSystem, which is not supported. You have to set a distributed filesystem, such as HDFS as the default.
Although Kudu is a separate storage system and does not rely on HDFS, Impala still seems to require a non-local default FS even when using with Kudu. The Impala_Kudu documentation explicitly lists the following requirement:
Before installing Impala_Kudu, you must have already installed and configured services for HDFS (though it is not used by Kudu), the Hive Metastore (where Impala stores its metadata), and Kudu.
I can even imagine that HDFS may not really be needed for any other reason than to make Impala happy, but this is just speculation from my side. Update: Found IMPALA-1850 which confirms my suspicion that HDFS should not be needed for Impala any more, but it's not just a single check that has to be removed.

running Hadoop with HBase: org.apache.hadoop.hbase.client.HTable.<init>(Lorg/apache/hadoop/conf/Configuration;Ljava/lang/String

I'm trying to make a mapreduce program on Hadoop using HBase. I'm using Hadoop 2.5.1 with HBase 0.98.10.1.
The program can be compiled successfully and being made into a jar file. But, when I try to run the jar using "hadoop jar" the program shows error says:
"org.apache.hadoop.hbase.client.HTable.(Lorg/apache/hadoop/conf/Configuration;Ljava/lang/String".
Here is the line of code I used to initiate the HTable.
HBaseConfiguration config = new HBaseConfiguration();
HTable table = new HTable(config, "Test");
What should I change to overcome this error?
I've somehow solved this error by just re-installing both Hadoop 2.5.1 and HBase 0.98.10

Apache oozie sharedlib is showing a blank list

Relatively new to Apache OOZIE and did an installation on Ubuntu 14.04, Hadoop 2.6.0, JDK 1.8. I was able to install oozie and the web console is visible at the 11000 port of my server.
Now while i copied the examples bundled with oozie and tried to run them i am running into an error which says no sharedlib exists.
Installed the sharedlib as below-
bin/oozie-setup.sh sharelib create -fs hdfs://localhost:54310
(my namenode is running on localhost 54310 and JT on localhost 54311)
hadoop fs -ls /user/hduser/share/lib is showing shared library created as per the oozie-site.xml file. However when i check the shared library using the command -
oozie admin -oozie http://localhost:11000/oozie -shareliblist the list is blank and also jobs are failing for the same reason.
Any clues on how should i approach this problem?
Thanks.
The sharelib create command looks fine.
If you havent done so already copy the core-site.xml from your hadoop installation folder into $OOZIE_HOME/conf/hadoop-conf/.
There might already be a "placeholder" core-site.xml in the hadoop-conf folder, delete or rename that one. Oozie doesnt get its hadoop configuration directly from your hadoop install (like hive for example) but from the core-site.xml you place in that hadoop-conf folder.
Okay i got a solution for this.
So when i was trying to create the sharedlib directory it was doing on HDFS but while running the job local path was being refereed. So i extracted the oozie-sharedlib tar.gz file in my local /user/hduser/share/lib directory and its working now.
But did not get the reason so its still an open question.
I have encountered the same issue and it turned out that
oozie was not able to communicate with hdfs, as it was not able to find the location for core-site.xml or any other hadoop configuration which has to be declared inside oozie-site.xml.
Corresponding property in oozie-site.xml is oozie.service.HadoopAccessorService.hadoop.configurations
this property was defined wrongly in my case.
changed it to point to where my Hadoop configuration xmls are present and then it started communicating with hdfs and hence was able to locate the sharelib on hdfs