Tornado and Concurrency

Reading the Tornado documents did not grand me basic understanding of asynchronous web service calls. After referring back and forth external examples and blogs, I summarize my brief understanding of tornado’s concurrency model.

  1. Common multi-thread web service frameworks, where one user request is given one dedicated thread, controls the number of concurrent threads running on a web server. The thread is blocked once the request hits some blocking calls (e.g. database access or an external blocking http call). The thread will wait until a result is returned and the control is returned to the thread.
  2. Different from above, Tornado makes use of IOLoop, a single thread to handle all requests, though a server may have multiple IO Loops, we did not discuss this case here. The design of IOLoop makes the design of asynchronous calls clearer, i.e. every asynchronous calls is spawned or yielded  from this single IOLoop, and eventually return controls to this single IOLoop. IOLoop is created by Application which is a singleton to maintain global status.
  3. Tornado making asynchronous easier does not mean by default Tornado is asynchronous. By default, handlers are synchronous, and it means other user requests will be blocked severely since IOLoop has only one thread, IF the blocking factor is non-negligible, like database calls, aws calls, long-polling (websocket maybe). In order to make handler asynchronous for those blocking monsters, first we need to annotate handler method with @tornado.web.asynchronous using callbacks, or annotate with @tornado.gen.coroutine using yield. Second, you need to make your calls asynchronous or in a executor pool: the design consideration will follow recommendations like this,
    1.  Try make the calling component asynchronous by using asynchronous library (database asynchronous drivers, e.g.) or AsynchHttpClient for external calls.
    2. Make synchronous call faster by using faster local database etc.
    3. If remote synchronous (blocking) calls have to be used, then make use of ThreadPoolExecutor.

So what to do for blocking database drivers like redshift?  Yes, redshift does not have asynchronous query driver support. And, via AWS we can’t safely assume the synchronous connection is fast (DynamoDB maybe fine).

Then we need to make use of ThreadPoolExecutor.

 

 

Tagged ,

Avro Parquet Redshift and S3 with Spark

Here I summarize what I found through the journey, while I have time when my cluster is busy running.

What I have:

Raw Input: in Avro with many schemas mixed in s3 bucket (this is bad, later I will tell you why)

Target: store date in redshift for query (this is ok. well I can work on parquet directly, storing in a db-like place can save me a lot of trouble when exporting and ad-hoc queries are badly needed in no time!)

So choices of approaches:

  1. parse avro to parquet and make use of spark parquet package to write into a redshift.
  2. load avro directly to redshift via COPY command

Choice 2 is better than Choice 1, because parquet to redshift actually is converted to avro and written into s3. Choice 1 requires two rounds of network io.

Choice 1 required only once, but you need to create your table first (which is simple and should be done for better control over schema).

What made me crazy:

  1. mix different avro schemas into one S3 bucket is a bad bad idea: currently library lacks of loading different schemas in one call. They failed to identify the difference of schemas. Instead, spark parquet package will throw schema column is not found error. And COPY command will falsely match different schemas into a strange result.
  2. avro needs to be checked extensively. Corrupted data will make program cry. Redshift COPY’s maxerror has no effects over corrupted data. Two types of errors were seen: 1. Cannot init avro reader from s3 file Incorrect sync bytes. 2. Invalid AVRO file found. Unexpected end of AVRO file. The two errors corresponds to spark parquet packages 1. Invalid Sync! 2. gives you empty avro data.
  3. For COPY command, I use manifest in S3. My advice is to use different key for different manifest, don’t try to overwrite existing manifest. There are cases you did not overwrite but append! Once manifest appended, redshift will append the records as well as duplicates.

 

 

 

Walk Through of Mahout LDA 0.9 againt Tweets on CDH 5.2.+

For the time being, only key points are listed. Best Tutorial Ever Follow Tom’s book on develop MapReduce Programs in a regression way, unit test -> local file -> hdfs file -> multiple hdfs files Package dependencies with maven shade plugin Flume generates too many small files 1170+ files, on my single CDH machine MR takes longer time 1 hour+ to process  (about 23k tweets) MR produces one 2.2 MB sequence file with key tweetID and value text. Create vectors using mahout seq2sparse (stopwords can be filtered out by the maxDFPercentage and using TF, ?? unknown). Create rowid for mahout LDA to work (convert into matrix ?? unknown).Run mahout cvb. Beware that if you don’t want to hit hdfs, remember to set sth. to MAHOUT_LOCAL so that mahout will run only in local mode. Applying seqdumper to read intermediate sequence file and using vectordump to display LDA output (topic term is the one displaying terms for a topic, whereas topic document is for document and topic matching) Yarn Related:

  • unset HADOOP_CLASSPATH (make sure Yarn will see the exact path as you are using. Your local Hadoop Classpath will not be able to transfer to Yarn)
  • sudo -u hdfs hadoop jar thePackagedJar.jar DriverClass
  • -fs hdfs://cdh-manager.localdomain (file:/// means local file system with respect to the current directory or absolute path)
  • -jt cdh-manager.localdomain:8032 (local means using local runner; this url pointing to Yarn’s resource manager)
  • /user/flume/tweets/2014/12/22/{13,14,15}” (input arg[0] taking a list of directories as set. For full list of file path pattern supported by hadoop, search “hadoop fs path pattern”)
  • /out-seq (output directory with respect to hdfs:/// root)

CDH may not be able to run your application for above mahout, because mahout may contains several application in a pipeline within the mahout scripts. Cases may happen when job is stuck at the point of requiring resources. My intake is that browsing the configuration of Yarn, and set all the memory and vcore related configuration to the default (without worrying about the physical limitations).

  • mahout seqdumper -i <input file name, address depending on MAHOUT_HOME local or hdfs>
  • -o <a local file>
  • sudo -u hdfs mahout vectordump
  • -i /tweet-lda
  • -o /tmp/topic_terms
  • -p true (Print out the key as well, delimited by tab (or the value if useKey is true)
  • -d /seqdir-sparse-lda/dictionary.file-* (the dictionary used for interpreting term)
  • -dt sequencefile (The dictionary file type (text|seqfile) is sequencefile)
  • -sort /tweet-lda (Sort output key/value pairs of the vector entries in abs magnitude of ?? unknown)
  • -vs 10 (each topic to display the top 10 terms associated with this topic)
  • -ni 1000000 (max number of items, in this case the max number of topics)

Mahout driver classes use a lot of shorthand. Check what they mean by going to file driver.classes.default.props located under conf/  

1. sudo -u hdfs mahout seq2sparse -i /demo-out-seq -o /demo-seq2sparse -ow --maxDFPercent 85 --namedVector -wt TF 2. sudo -u hdfs mahout rowid -i /demo-seq2sparse/tf-vectors -o /demo-rowid-matrix 3. sudo -u hdfs mahout cvb -i /demo-rowid-matrix/matrix -o /demo-tweet-cvb -k 10 -ow -x 5 -dict /demo-seq2sparse/dictionary.file-0 -dt /demo-cvb-topics -mt /demo-cvb-model 4. sudo -u hdfs mahout vectordump -i /demo-tweet-cvb -o /tmp/demo-topic_terms -p true -d /demo-seq2sparse/dictionary.file-* -dt sequencefile -sort /demo-tweet-cvb -vs 10 -ni 1000000

When writing a MapReduce Application, following these steps.

  1. Unit test your Mapper and Reducer class. First create unit test on the individual class methods. Second use mrunit to write drive test on mapper and reducer. Driver’s configuration can be set to test various situations.
  2. Write your application driver test. Of course you gonna need your driver class first by extending Configured and implementing Tool. In your test, invoke driver’s run method to actually run your applications. It is somehow like the integration test, since it is going to hit some file system (either local or hdfs). The purpose here is to test out a smaller real data set to see if your application is working.
  3. The last testing is really the cluster deployment testing. You can involve a relatively larger dataset and actually submit your driver application to the cluster.

Following above steps allow you to have confidence when debugging. Messing up the sequence will give you no idea where the possible error is. Passing external files (runtime) to Map Reduce. It is common that map reduce application may depend on some external configurations submitted during job submission time. There are several approaches to achieve it.

  • Map Reduce configuration is a channel for job to communicate with mapper and reducer tasks. You can specify a key in mr tasks, and the task at setup time simply do a checking on the keys. Let’s say if “external.filename” has a filename pointing to it, the task knows it should check out some file. But how to let the file accessible? We can use console’s -files options which would add any local files to the task’s classpath. At runtime the file name can be specified through console option -D property=value.

sudo -u hdfs hadoop jar target/your.jar DriverClass -fs hdfs://cdh-manager.localdomain -jt cdh-manager.localdomain:8032 -files target/classes/stopwords.data -Dcustom.stopwords.filename=stopwords.data /user/flume/tweets/2015/01/09/11 /out-seq

  • Another method is related to doing some work on the application driver class, by reading the argument list. If a relevant argument is detected, call MapReduce Job API to “addFileToClassPath” or  “addCacheFile”. But it may require a hdfs path visible (i.e. the file has been uploaded to hdfs or a URI is available for hadoop to fetch the file). I am still investigating and the useful link would be latter part of this tutorial.

Quick Run Spark with Docker in Mac OS

For the time being, I am just listing a brief guide.

1. install boot2docker, which is a linux core required to run docker.

2. install spark docker by https://github.com/sequenceiq/docker-spark (just pull docker image). The docker image has a hadoop yarn inside as well.

3. in MacOs, start boot2docker first, note down and export the $DOCKER_HOST environment. In MacOS, any docker client command requires this environment to connect to boot2docker’s host.

4. after export the docker environment in MacOS, investigate docker containers by `docker ps` with -a or -l(astest), or `docker inspect CONTAINER_HASH

5. now in Mac OS, it is safe to run

docker run  -p 4040:4040 -p 8030:8030 -p 49707:49707 -p 50020:50020 -p 8042:8042 -p 50070:50070 -p 8033:8033 -p 8032:8032 -p 50075:50075 -p 22:22 -p 8031:8031 -p 8040:8040 -p 50010:50010 -p 50090:50090 -p 8088:8088 -i -t -h sand    box sequenceiq/spark:1.2.0 /etc/bootstrap.sh -bash

I open the ports so that MacOS can access directly.

6. you can exam the ports opened in boot2docker by `sudo iptables -t nat -L -n` from boot2docker. Enter boot2docker by `boot2docker ssh`, and make hostname `sandbox` to be reflected in Mac OS by adding it in /etc/hosts

Despite the learning curve in docker, I have to say so far this way is the most convenient for deploying/testing a spark application and less computational power consuming.

Downside: It is not a cluster, but one node only.

Learning Notes: Software Architecture – Taking A Deeper Dive

Trade Off and ATAM:

Fulfilling all stakeholder’s requirements are impossible

Agile ATAM: Design conceptual architecture -> Bring stakeholder to define scenarios -> prioritize stakeholder’s requirements -> help in negotiating the priority -> redesign -> ask relevant stakeholder representatives including programmers to negotiate/evalute -> evolve the design and involve stakeholders early and often (dealing with scenarios/techniques changes)

Further reading: Software Architecture in Practice 3rd Edition. SEI digital library.

Continuous Delivery:

Water Fall approach is not getting quick feedback, but do continuous delivery to get fast feedback by releasing new features constantly.

holistic engineering: friction, e.g. database schema from source code.??

Automation in your development, testing and deployment; use automated tools or developing your own tools, e.g. use selenium, cucumber-puppet.

When automating, do consider the time constrain by using a timer, and avoid shaving the yak. Good enough for now is better than perfect in future.

Differentiate aspects among static, dynamic and production, developing working software as soon as possible, building vertical slides over time,

hide new feature until it finishes: feature toggle over feature branch (sync as often as necessary to avoid conflicts- merge ambush), togglez platform, google chunk-based development

Each change should be small and releasable: users don’t like giant changes, but gradually changes like in Facebook; Canary Releasing using router to choose users to move to newer version;  dark releasing using timezone to gradually release like Facebook chat; strangler pattern: new changes obsoletes a small percent of old features and live together by allowing users to change to new feature by themselves, like cloudera’s switching to new configuration UI.

Walk Through of Installing Cloudera Manager on a Single Node

I am following the official guide of auto installation, while it seems easy at first glance. I had a rough time on it.

Before going on, CDH is considered kind of heavy distribution of hadoop. What I am doing is to use CDH as dev, and so far my experiences are (1) VMWorkstation on a 16G RAM windows host machine, with CDH taking up to 8G single node or (2) Mac 8G RAM host, with vagrant-enabled virtual box taking up to 4G and 2 vcores for a single node (more than one node will cause non-functional CDH)

Make sure if you are behind a proxy by enabling proxy as stated in the guide. In addition, once you are in the phase of downloading parcels, remember to configure parcel downloading proxy setting through web browser. Confirm on the downloading issue of parcel by checking ‘/var/log/cloudera-scm-server/cloudera-scm-server.log‘.

Don’t need to deal with java and just let it go with default oracle-j2sdk.

Disable IPv6 by following the blog guide.

Find ip address of the machine node and create FQDN as suggested in the comments. And Cloudera requires more and more:

The hosts in a Cloudera Manager deployment must satisfy the following networking and security requirements:

  • Cluster hosts must have a working network name resolution system and correctly formatted /etc/hosts file. All cluster hosts must have properly configured forward and reverse host resolution through DNS. The /etc/hosts files must
    • Contain consistent information about hostnames and IP addresses across all hosts
    • Not contain uppercase hostnames
    • Not contain duplicate IP addresses

    A properly formatted /etc/hosts file should be similar to the following example:

    127.0.0.1	localhost.localdomain	localhost
    192.168.1.1	cluster-01.example.com	cluster-01
    192.168.1.2	cluster-02.example.com	cluster-02
    192.168.1.3	cluster-03.example.com	cluster-03

My intake is that 1) avoid using 127.0.0.1 from loopback interface, using IPs assigned through eth0 interface. 2) make hostname is FQDN as well, by using `sudo hostname <FQDN>` and saving the name in `/etc/hostname` for reboot.

Allow host to be resolvable from local /etc/hosts files by this:

cat /etc/NetworkManager/dnsmasq.d/hosts.conf 
addn-hosts=/etc/host

Disable firewall and iptable as in the guide.

CDH requires root access using password or private key. My take is using password for root user is easier. Do following,

Make sure openssh-server is installed and started

sudo apt-get install openssh-server

Give root a password, enable ssh root access

sudo passwd root;

vi /etc/ssh/sshd_config;

PermitRootLogin yes

service ssh restart

Test ssh access as root

ssh localhost

With above configuration, install cloudera manager and installation cluster should be working.

During cluster installation, if you need to retry from web browser, you may need to manually remove the lock by:

sudo rm /tmp/.scm_prepare_node.lock

If encountering any problem, you can always uninstall and get back to a clean state by following the uninstallation guide.

Note: during cluster installation, if the web browser does not show any progress bar, that means something wrong. Check the root access listed above.

======= after a running CDH, configurations to be continue ======
CDH calculates the settings (like memory location) for the host, but sometimes the configuration is not checked against the minimum requirement of installed components.

For example, the test installation with estimating PI does not work, unless increasing following memory settings in Yarn as (wired enough, the log in yarn does not point anything useful),

– Set the Container Memory (yarn.nodemanager.resource.memory-mb) to 4GB
– Set the Java Heap Size of ResourceManager to 1GB

The most useful way is to check the non-default settings by switching to the new view.

======= running mahout example ======
When trying to execute `mahout seq2sparse -i reuters-out-seqdir/ -o reuters-out-seqdir-lda -ow –maxDFPercent 85 –namedVector` with `MAHOUT_LOCAL` set to “something not null”, meaning running on local, guava library version mismatch.

Exception in thread “main” java.lang.NoSuchMethodError: com.google.common.base.Stopwatch.elapsedMillis()J

Mahout requires guava 16.0, while hadoop V2 uses guava 11.0.

The solution is quite wired. I was simply going to turn on the log by reading CDH 5’s mahout script, which is pointing mahout conf directory to /etc/mahout/conf.dist. In the conf directory, I put a simple log4j properties under it. Surprisingly, the guava problem is gone.

Learning Notes of Scala

Learning Scala is not easy, but it is interesting. I followed the Coursera course to understand scala. The purpose to learn scala is

1. to understand how functional could be helpful for my daily tasks.

2. to make use of spark better and to be able to write spark code.

3. to make use of Akka for concurrency programming, which is an advanced stage for java programmers like me.

The fact is,

Non-deterministic = parallel processing (concurrency and parallel) + mutable states

People want to write codes that are within their control, while controlling time (sync over time and remove unwanted scenarios) is difficult. So imperative programming is from the machine perspective, while not for the sake of programmers, and the main purpose of scala for me is to learn how to write concurrent codes easily.

Get Your Own Playframework Working

I found there is a lack on how to run customised play framework in the internet. This is what I found,

Definition:

Base Play version: bv; The play version which your play project based on: pv; Sbt-version specified in your play project’s build.properties: sv; your customised play version, cv;

0. Download the play binary’s and dependencies. (not required maybe)

1. Download the source of base versioned as bv;

2. Change the source code of bv as whatever you want;

3. Compile your play as specified in play docs (must be the same bv); sth. like ./build; public-local etc. your customised play will be available in .ivy2/local

4. Make sure your pv is the same as bv. Otherwise do a upgrade according to the docs. Note, if pv is different from bv, some manual work needs to be done such as change sv, different syntax in build.sbt, different play commands activator /play run, etc.

5. Make your project depends on your customised play, by adding

resolvers ++= Seq(
Resolver.file(“Local Repository”, file(“/YOUR PATH/.ivy2/local”))(Resolver.ivyStylePatterns),

)

//// Use the Play sbt plugin for Play projects
addSbtPlugin(“com.typesafe.play” % “sbt-plugin” % “cv”)

Note the cv is used for sbtplugin to identify the customised play build.

Install Apache Hadoop Cluster on one physical machine with Puppet and Vagrant

Speak At First

In this post, I will walk through the installation and configuration processes of my development environment including Apache Hadoop 1.2.1, Hive 0.13 and Presto 0.69 with the support of automate deployment and configuration management. The whole walkthrough was inspired by the post and this post with more slaves.

Short about Vagrant

I choose vagrant over docker, simply because vagrant gives me the feel of working on machines, instead of application environment. Vagrant is heavier than docker, since it takes time to build guest os on top of host os, while docker simulates environments by transmitting dependencies to containers. Vagrant gave me the feeling of working directly on virtual machines as what I do for everyday.

Useful commands for vagrants include

  • vagrant init: which gives me an initial Vagrantfile to work on
  • vagrant up –provision: builds up guest os with provision such as puppet; it took a while for first time and the guest addition version may need to be built as well for sharing folders that is needed for puppet later.
  • vagrant reload –provision: Once I change some configuration and want vagrant to re-apply my settings for guest os, this saves my a lot of time to avoid overhead as in the first time.
  • vagrant halt: for me to gracefully shut down my virtual machines before close my laptop.
  • vagrant destroy: oh, before doing that knowing you are going to take the overhead for the first time. But this cleans up all the messy guest os.

Get to know Puppet Basic

I am new to puppet, so I installed a test script for quick understanding of puppet first. Even before that, for me it is still a must-read beginner tutorial before writing the script.

Puppet Structure for vagrant

At vagrant root folder (the one with Vagrantfile), create folder “modules” and “manifests”. manifests folder contains the entry point for puppet to start. modules contain multiple module, i.e. a set of files for installing a library (like java, mysql for the guest os). A module is included in the overall puppet installation by “include [a module name]” in the entry point file like “test_vm.pp”.

So here it is, 1) a puppet script to create a test file (manifests/test_vm.pp, no modules)

2) the updated vagrant file, which defines a network ip and use vm provider to customize memory size.

Note: the syntax for private network and change memory size are specific to the latest vagrant version.

Problems even for this simple test puppet

  • Guest Addition Version mismatch issue: solution
  • Warning: Could not retrieve fact fqdn: This means the hostname is not in full qualified domain name. So either edit “domain my.com” to /etc/resolv.conf or do hostname my.com for current session or specify vm.hostname in vagrant in full
  • Warning: Config file /etc/puppet/hiera.yaml not found, using Hiera defaults: solutions. Note: one should look for the file in VM not the host machine, since this is not the problem for host but for the guest os.
  • Do puppet syntax check with puppet parser validate or do style checkup with puppet-lint; this saves your time by really deploying the script.

By now, the vagrant should be able to provide a nice vm with the specified test file, and we are ready to move on for more meaningful tasks, like install a java.

Install java in puppet

This post is a great reference to me.

  • puppet file with content => causing jitters for export PATH variables: in my case it is due to the whitespaces problem; simply make sure tabs are converted to spaces before debug further.

Install Hadoop with Puppet

test run and issues

  • create and grant access to tmp.dir
  • create masters and slaves.xml, using 0 secondary namenode, 2 additional slave nodes.
  • hadoop-env.sh.erb with $JAVA_HOME
  • Host Key Verification Fail, add to known_host key or specify hadoop ssh option as `UserKnownHostsFile=/dev/null` and `StrictHostKeyChecking=no`
  • org.apache.hadoop.mapred.ReduceTask: java.net.UnknownHostException: slave1.gp.netThe short answer is that Hadoop performs reverse hostname lookups even if you specify IP addresses in your configuration files. In your environment, in order for you to make Hadoop work, SSP-SANDBOX-1.mysite.com must resolve to the IP address of that machine, and the reverse lookup for that IP address must resolve to SSP-SANDBOX-1.mysite.com.
  • Do dfs format in master for enabling future hive/warehouse directory. Before doing that, I need to use puppet stage; a note between defining a class and use a class for stage to work
  • err: /Stage[final]/Hdfsrun/Exec[format hdfs]/returns: change from notrun to 0 failed: /usr/bin/env: bash: No such file or directory. It caused by 1. no path variable set or not full path (for hadoop script use full path, set path won’t work for me. Why?); 2. the script hadoop namenode -format includes an interactive yes/no portion. use option -force to bypass. However, it should be only for 1st time, otherwise all previous data is going to be lost. 3. specify user => vagrant, otherwise root by default will cause error

Install Hive with Puppet

Install mysql with puppet modules

I struggled to figure out this: VagrantFile has a field to specify module path in my host os. I specified as “modules” which is relative to my vagrant script starting point; In order to use puppetlabs modules which handles dependencies and configurations nicely, I need to 1) install puppet modules in my host os `puppet module install puppetlabs-mysql` that downloads puppet scripts under my home directory (of course –modulepath will overwrite this) 2) point my VagrantFile’s modulepath field to this. 3) Vagrant up each time will help me to copy all the necessary modules from my host to the guest tmp folder, and basically do puppet apply in the guest os. Therefore, no need to do puppet module install in the guest os at all.

  • Note: same user in puppet mysql script can only appear once otherwise duplicated declaration of the same user will occur.
  • Problem: Hive throws above exception when old version of MySQL is used as hive metastore.
    Solution: Set Latin1 as the charset for metastore
    mysql> alter database metastore_db character set latin1;

test run and issues

  • add mysql to rc.d or ensure => running
  • hive: get hdfs running first; type hive in to command to try first. Hive different modes: embeded-derby, local, remote. Presto only for remote. So creating remote hive server. Understand hive: http://www.cloudera.com/content/cloudera-content/cloudera-docs/CDH4/4.2.0/CDH4-Installation-Guide/cdh4ig_topic_18_4.html, but don’t 100% follow that for hive-site.xml. combined reference: http://rajnishbigdata.wordpress.com/mysql-with-hive/ ; http://www.thecloudavenue.com/2013/11/differentWaysOfConfiguringHiveMetastore.html; NOTE: mysql create user hive for localhost and create user hive for remote, and if needed grant privileges to local or remote. To test hive server, must get dfs running, and start hive –service metastore. then the metastore uri with thrift can be called by others like hiveserver2, impala or presto. must run hive –serivce hiveserver -p 10000 for presto? (presto does not require hiveserver, simply metastore server)
  • download movie lens data, unzip; to start hive database management (not query) 1. start-dfs, 2. mysql 3. metastore 9083, 4.hiveserver2 if from program otherwise only CLI.> hive> create tables.For hive query to work, must start map-r by start-mapred. Then select count(*) now works.
  • in order to configure hdfs for hive (creating hive folder and grant access), I prefer to write a shell script rather than puppet script, since a lot of condition checking with loop such as waiting name node leaves safe mode, checking existence of hive folders, etc. However, since puppet has a limited shell, I use command => bash -c to source .bashrc before running bash scripts since limited shell does not support regex etc.

Install Presto with Puppet

test run: A single presto node

  • start presto-server by launcher and start cli> now show tables not workv0.69 single presto node requires to set scheduler on.
  • presto log: /var/presto/data; set hive to /var/hive/hive.log
  • set discovery-service and worker node

test run: Multiple presto nodes

HOw to check workers have been discovered ?  https://groups.google.com/forum/#!searchin/presto-users/discovery/presto-users/Q6gmcI1Uo1s/EfeILTq1y5YJ

Make your VMs accessible to you

  • apt-get update, install vim, personal .vimrc, system time to SG

Writing a JDBC Client to Hive

  • Connection Refused: if hive.server2 host is set to localhost, then client program in host os cannot access through 10.17.3.10:10000; set hive.server2 host to fqdn, which matches /etc/hosts; then host os client can connect
  • Cannot execute select query: Error while processing statement: FAILED: Execution Error, return code 1 from org.apache.hadoop.hive.ql.exec.mr.MapRedTask. Go to hive server stderr it reads org.apache.hadoop.security.AccessControlException: org.apache.hadoop.security.Ac
    cessControlException: Permission denied: user=anonymous, access=WRITE, … The issue was caused by the jdbc client DriverManager.getConnection(“jdbc:hive2://10.17.3.10:10000/default”, “”, “”); previously provides an anonymous user name, which according to dfs the user has no write access. The solution is to simply put the user name on the hdfs who has the access. But the password is left empty which is very insecure —Unsolved

Deploy to Real world

NOTE!! apt-get install puppet-master only, not puppet passenger, otherwise, conflicts cause cert versification get revoked each time.

 

bamboo-dep:

0. vi /etc/hosts
10.110.254.128  citymodel-001.hpls.local citymodel-001
1. install ruby
sudo apt-get update
curl -L https://get.rvm.io | bash -s stable
source ~/.rvm/scripts/rvm
rvm install ruby
rvm rubygems curret
2. uninstall puppet
sudo apt-get –purge remove puppet
sudo apt-get autoremove
rm -rf /var/lib/puppet
rm -rf /etc/puppet
3. install
sudo apt-get install puppet
sudo puppet resource service puppet ensure=running enable=true
4 vi /etc/puppet/puppet.conf
[main]
server=citymodel-001.hpls.local
environment=try
5. on master puppet.conf
Set environmentpath = $confdir/environments in the puppet master’s puppet.conf
dns_alt_names=citymodel-001,citymodel-001.hpls.local
run:
sudo puppet master –verbose –no-daemonize
—————-
sudo puppet agent -t
6. master sign
sudo puppet cert sign “bambo-dep”
7. on agent
sudo puppet agent -t
Error: Could not request certificate: SSL_connect returned=1 errno=0 state=SSLv3 read server certificate B: certificate verify failed: [certificate revoked for /CN=citymodel-001.hpls.local]
make sure agent is running.
ps aux| grep puppet
8
Error: Could not retrieve catalog from remote server: Error 400 on SERVER: Could not find class first for bamboo-dep on node bamboo-dep
Warning: Not using cache on failed catalog
Error: Could not retrieve catalog; skipping run
# puppet agent --test --noop --debug



Play Use Bonecp connected to HiveServer2 failed "cannot connect database [datasource name]"
Debug into Hive.log


2014-06-16 06:20:28,153 INFO [pool-2-thread-1]: thrift.ThriftCLIService (ThriftCLIService.java:OpenSession(188
)) - Client protocol version: HIVE_CLI_SERVICE_PROTOCOL_V6
2014-06-16 06:20:28,167 INFO [pool-2-thread-1]: hive.metastore (HiveMetaStoreClient.java:open(297)) - Trying t
o connect to metastore with URI thrift://10.17.3.10:9083
2014-06-16 06:20:28,168 INFO [pool-2-thread-1]: hive.metastore (HiveMetaStoreClient.java:open(385)) - Connecte
d to metastore.
2014-06-16 06:20:28,169 INFO [pool-2-thread-1]: session.SessionState (SessionState.java:start(360)) - No Tez s
ession required at this point. hive.execution.engine=mr.
2014-06-16 06:20:28,211 INFO [pool-2-thread-1]: session.SessionState (SessionState.java:start(360)) - No Tez s
ession required at this point. hive.execution.engine=mr.
2014-06-16 06:20:28,276 INFO [pool-2-thread-2]: thrift.ThriftCLIService (ThriftCLIService.java:OpenSession(188
)) - Client protocol version: HIVE_CLI_SERVICE_PROTOCOL_V6
2014-06-16 06:20:28,279 INFO [pool-2-thread-2]: hive.metastore (HiveMetaStoreClient.java:open(297)) - Trying t
o connect to metastore with URI thrift://10.17.3.10:9083

repeated for three times.


Play's Bonecp cannot establish reliable connections with Hive. After debug, I found the cause is by hive, java.sql.SQLException: enabling autocommit is not supported

Crazy IntelliJ with Play Integration

I am always a fun of Intellij until it throws at me a bomb when working with play framework.

This post is served at the purpose of solving (at least listing) some of the problems I encountered.

To reproduce:

  1. Play Console [p>]: play create myapp with Java by `play new [a project name]`
  2. Import play project using IntelliJ through play console by `idea with-sources=yes` and then import from intelliJ
  3. Follow the “todo” play tutorial

Problem List:

    • “cannot resolve symbol” for reverse routing: As in return redirect(routes.Application.tasks());. It is related to the fact that part of folder target, i.e. target/scala-2.10/src_managed (some scala code compiled by play compile) is not included as the source. So do “play compile” before importing into IntelliJ, and IntelliJ will notify you that

You have useless source roots which may corrupt resolve for play 2 framework in your project. Don’t delete but ignore the warnning.

Execute the following steps to update your dependencies [Solution provided in Github]:

    1. Run the update task from your play console
    2. Remove the .idea_modules and .idea/libraries directories
    3. Run the idea with-sources=yes command in the play console
  • “Cannot find Tools | Play with Playframework”: for Play 2.0 Support with version 0.36.431, the play console is not seen under tools, but accessible from context menu of play-based modules like controller application, by Run Play 2 App after installing play 2 plugin.
  • When running test from intellij, an error the underlying test class is not found: check the folder test-classes, it is possible that the underlying test class never get compiled before running, so do test:compile without actual running the test. It is a bug reported here. So for now, do play>test and then run test. — Unsolved. Relative background of compiler process
  • “cannot resolve symbol” for reverse routing in “index.scala.html”: @form(routes.Application.newTask()): simply add class_managed to the class path.
  • to run test when accessing db is needed: put it with in a fakeApplication() and Runnable. Refer here
  • type mismatch @inputText(taskForm("label"))Unsolved

Working on play 2.3.0 with IntelliJ

Install plugin, scala, Play 2.0 support, SBT(optional)

activator> compile > test:compile > idea withsources=yes

open build.sbt > IntelliJ asks me to associate this build.sbt with project, do import

change IDE language levels if necessary.

Check IDE preference for play, sbt, scala.

Run a JUnit method test by IDE; Debug with Play 2 better than activator -jvm-debug 9999 run.

IDE will automatically create root and root-build in project modules.

NOTE: 2.10.4 is used instead of scalaVersion specified in build.sbt (Unsolved)

Install https://github.com/jrudolph/sbt-dependency-graph following the readme ‘s two steps, and within play run the tasks.

Warning: Class path contains multiple SLF4J bindings. slf4j-log4j12 conflicts with logback. Use dependency-graph to identify who is using log4j.

Follow

Get every new post delivered to your Inbox.