Hari Sekhon - DevOps Python Tools

Hadoop, Spark / PySpark, HBase, Pig, Ambari, IPython and Linux Tools

A few of the Big Data, NoSQL & Linux tools I've written over the years. All programs have --help to list the available options.

For many more tools see the DevOps Perl Tools and Advanced Nagios Plugins Collection repos which contains many Hadoop, NoSQL, Web and infrastructure tools and Nagios plugins.

Hari Sekhon

Big Data Contractor, United Kingdom

https://www.linkedin.com/in/harisekhon

Make sure you run make update if updating and not just git pull as you will often need the latest library submodule and possibly new upstream libraries.

Quick Start

Ready to run Docker image

All programs and their pre-compiled dependencies can be found ready to run on DockerHub.

List all programs:

docker run harisekhon/pytools

Run any given program:

docker run harisekhon/pytools <program> <args>

Automated Build from source

git clone https://github.com/harisekhon/devops-python-tools pytools
cd pytools
make

Make sure to read Detailed Build Instructions further down for more information.

Some Hadoop tools with require Jython, see Jython for Hadoop Utils for details.

Usage

All programs come with a --help switch which includes a program description and the list of command line options.

Environment variables are supported for convenience and also to hide credentials from being exposed in the process list eg. $PASSWORD, $TRAVIS_TOKEN. These are indicated in the --help descriptions in brackets next to each option and often have more specific overrides with higher precedence eg. $AMBARI_HOST, $HBASE_HOST take priority over $HOST.

DevOps Python Tools

• Linux:
  • anonymize.py - anonymizes your configs / logs from files or stdin (for pasting to Apache Jira tickets or mailing lists)
    • anonymizes:
      • hostnames / domains / FQDNs
      • email addresses
      • IP + MAC addresses
      • Kerberos principals
      • LDAP sensitive fields (eg. CN, DN, OU, UID, sAMAccountName, member, memberOf...)
      • Cisco & Juniper ScreenOS configurations passwords, shared keys and SNMP strings
    • anonymize_custom.conf - put regex of your Name/Company/Project/Database/Tables to anonymize to <custom>
    • placeholder tokens indicate what was stripped out (eg. <fqdn>, <password>, <custom>)
    • --ip-prefix leaves the last IP octect to aid in cluster debugging to still see differentiated nodes communicating with each other to compare configs and log communications
    • --hash-hostnames - hashes hostnames to look like Docker temporary container ID hostnames so that vendors support teams can differentiate hosts in clusters
    • anonymize_parallel.sh - splits files in to multiple parts and runs anonymize.py on each part in parallel before re-joining back in to a file of the same name with a .anonymized suffix. Preserves order of evaluation important for anonymization rules, as well as maintaining file content order. On servers this parallelization can result in a 30x speed up for large log files
  • find_duplicate_files.py - finds duplicate files in one or more directory trees via multiple methods including file basename, size, MD5 comparison of same sized files, or bespoke regex capture of partial file basename
  • welcome.py - cool spinning welcome message greeting your username and showing last login time and user to put in your shell's .profile (there is also a perl version in my DevOps Perl Tools repo)
• Hadoop & NoSQL:
  • Spark & Data Format Converters:
    • spark_avro_to_parquet.py - PySpark Avro => Parquet converter
    • spark_parquet_to_avro.py - PySpark Parquet => Avro converter
    • spark_csv_to_avro.py - PySpark CSV => Avro converter, supports both inferred and explicit schemas
    • spark_csv_to_parquet.py - PySpark CSV => Parquet converter, supports both inferred and explicit schemas
    • spark_json_to_avro.py - PySpark JSON => Avro converter
    • spark_json_to_parquet.py - PySpark JSON => Parquet converter
    • json_to_xml.py - JSON to XML converter
    • xml_to_json.py - XML to JSON converter
    • json_docs_to_bulk_multiline.py - converts json files to bulk multi-record one-line-per-json-document format for pre-processing and loading to big data systems like Hadoop and MongoDB, can recurse directory trees, and mix json-doc-per-file / bulk-multiline-json / directories / standard input, combines all json documents and outputs bulk-one-json-document-per-line to standard output for convenient command line chaining and redirection, optionally continues on error, collects broken records to standard error for logging and later reprocessing for bulk batch jobs, even supports single quoted json while not technically valid json is used by MongoDB and even handles embedded double quotes in 'single quoted json'
    • see also validate_*.py further down for all these formats and more
  • Ambari:
    • ambari_blueprints.py - Blueprint cluster templating and deployment tool using Ambari API
      • list blueprints
      • fetch all blueprints or a specific blueprint to local json files
      • blueprint an existing cluster
      • create a new cluster using a blueprint
      • sorts and prettifies the resulting JSON template for deterministic config and line-by-line diff necessary for proper revision control
      • optionally strips out the excessive and overly specific configs to create generic more reusable templates
      • see the ambari_blueprints/ directory for a variety of Ambari blueprint templates generated by and deployable using this tool
    • ambari_ams_*.sh - query the Ambari Metrics Collector API for a given metrics, list all metrics or hosts
    • ambari_cancel_all_requests.sh - cancel all ongoing operations using the Ambari API
    • ambari_trigger_service_checks.py - trigger service checks using the Ambari API
  • Hadoop HDFS:
    • hadoop_hdfs_time_block_reads.jy - HDFS per-block read timing debugger with datanode and rack locations for a given file or directory tree. Reports the slowest Hadoop datanodes in descending order at the end. Helps find cluster data layer bottlenecks such as slow datanodes, faulty hardware or misconfigured top-of-rack switch ports.
    • hadoop_hdfs_files_native_checksums.jy - fetches native HDFS checksums for quicker file comparisons (about 100x faster than doing hdfs dfs -cat | md5sum)
    • hadoop_hdfs_files_stats.jy - fetches HDFS file stats. Useful to generate a list of all files in a directory tree showing block size, replication factor, underfilled blocks and small files
  • HBase:
    • hbase_generate_data.py - inserts random generated data in to a given HBase table, with optional skew support with configurable skew percentage. Useful for testing region splitting, balancing, CI tests etc. Outputs stats for number of rows written, time taken, rows per sec and volume per sec written.
    • hbase_show_table_region_ranges.py - dumps HBase table region ranges information, useful when pre-splitting tables
    • hbase_table_region_row_distribution.py - calculates the distribution of rows across regions in an HBase table, giving per region row counts and % of total rows for the table as well as median and quartile row counts per regions
    • hbase_table_row_key_distribution.py - calculates the distribution of row keys by configurable prefix length in an HBase table, giving per prefix row counts and % of total rows for the table as well as median and quartile row counts per prefix
    • hbase_compact_tables.py - compacts HBase tables (for off-peak compactions). Defaults to finding and iterating on all tables or takes an optional regex and compacts only matching tables.
    • hbase_flush_tables.py - flushes HBase tables. Defaults to finding and iterating on all tables or takes an optional regex and flushes only matching tables.
    • hbase_regions_by_*size.py - queries given RegionServers JMX to lists topN regions by storeFileSize or memStoreSize, ascending or descending
    • hbase_region_requests.py - calculates requests per second per region across all given RegionServers or average since RegionServer startup, configurable intervals and count, can filter to any combination of reads / writes / total requests per second. Useful for watching more granular region stats to detect region hotspotting
    • hbase_regionserver_requests.py - calculates requests per regionserver second across all given regionservers or average since regionserver(s) startup(s), configurable interval and count, can filter to any combination of read, write, total, rpcScan, rpcMutate, rpcMulti, rpcGet, blocked per second. Useful for watching more granular RegionServer stats to detect RegionServer hotspotting
    • hbase_regions_least_used.py - finds topN biggest/smallest regions across given RegionServers than have received the least requests (requests below a given threshold)
  • OpenTSDB:
    • opentsdb_import_metric_distribution.py - calculates metric distribution in bulk import file(s) to find data skew and help avoid HBase region hotspotting
    • opentsdb_list_metrics*.sh - lists OpenTSDB metric names, tagk or tagv via OpenTSDB API or directly from HBase tables with optionally their created date, sorted ascending
  • Pig
    • pig-text-to-elasticsearch.pig - bulk index unstructured files in Hadoop to Elasticsearch
    • pig-text-to-solr.pig - bulk index unstructured files in Hadoop to Solr / SolrCloud clusters
    • pig_udfs.jy - Pig Jython UDFs for Hadoop
  • ipython-notebook-pyspark.py - per-user authenticated IPython Notebook + PySpark integration to allow each user to auto-create their own password protected IPython Notebook running Spark
  • find_active_server.py - returns first available healthy server or active master in high availability deployments, useful for chaining with single argument tools. Configurable tests include socket, http, https, ping, url and/or regex content match, multi-threaded for speed. Designed to extend tools that only accept a single --host option but for which the technology has later added multi-master support or active-standby masters (eg. Hadoop, HBase) or where you want to query cluster wide information available from any online peer (eg. Elasticsearch)
    • The following are simplified specialisations of the above program, just pass host arguments, all the details have been baked in, no switches required
      • find_active_hadoop_namenode.py - returns active Hadoop Namenode in HDFS HA
      • find_active_hadoop_resource_manager.py - returns active Hadoop Resource Manager in Yarn HA
      • find_active_hbase_master.py - returns active HBase Master in HBase HA
      • find_active_hbase_thrift.py - returns first available HBase Thrift Server (run multiple of these for load balancing)
      • find_active_hbase_stargate.py - returns first available HBase Stargate rest server (run multiple of these for load balancing)
      • find_active_apache_drill.py - returns first available Apache Drill node
      • find_active_cassandra.py - returns first available Apache Cassandra node
      • find_active_impala*.py - returns first available Impala node of either Impalad, Catalog or Statestore
      • find_active_presto_coordinator.py - returns first available Presto Coordinator
      • find_active_kubernetes_api.py - returns first available Kubernetes API server
      • find_active_oozie.py - returns first active Oozie server
      • find_active_solrcloud.py - returns first available Solr / SolrCloud node
      • find_active_elasticsearch.py - returns first available Elasticsearch node
      • see also: Advanced HAProxy configurations which are part of the Advanced Nagios Plugins Collection
• Docker:
  • docker_registry_show_tags.py / dockerhub_show_tags.py / quay_show_tags.py - shows tags for docker repos in a docker registry or on DockerHub or Quay.io - Docker CLI doesn't support this yet but it's a very useful thing to be able to see live on the command line or use in shell scripts (use -q/--quiet to return only the tags for easy shell scripting). You can use this to pre-download all tags of a docker image before running tests across versions in a simple bash for loop, eg. docker_pull_all_tags.sh
  • dockerhub_search.py - search DockerHub with a configurable number of returned results (official docker search is limited to only 25 results), using --verbose will also show you how many results were returned to the termainal and how many DockerHub has in total (use -q / --quiet to return only the image names for easy shell scripting). This can be used to download all of my DockerHub images in a simple bash for loop eg. docker_pull_all_images.sh and can be chained with dockerhub_show_tags.py to download all tagged versions for all docker images eg. docker_pull_all_images_all_tags.sh
  • dockerfiles_check_git*.py - check Git tags & branches align with the containing Dockerfile's ARG *_VERSION
• Travis CI:
  • travis_last_log.py - fetches Travis CI latest running / completed / failed build log for given repo - useful for quickly getting the log of the last failed build when CCMenu or BuildNotify applets turn red
  • travis_debug_session.py - launches a Travis CI interactive debug build session via Travis API, tracks session creation and drops user straight in to the SSH shell on the remote Travis build, very convenient one shot debug launcher for Travis CI
• Data Validation (useful in CI):
  • validate_*.py - validate files, directory trees and/or standard input streams
    • supports the following file formats:
      • Avro
      • CSV
      • INI / Java Properties (also detects duplicate sections and duplicate keys within sections)
      • JSON (both normal and json-doc-per-line bulk / big data format as found in MongoDB and Hadoop json data files)
      • LDAP LDIF
      • Parquet
      • XML
      • YAML
    • directories are recursed, testing any files with relevant matching extensions (.avro, .csv, json, parquet, .ini/.properties, .ldif, .xml, .yml/.yaml)
    • used for Continuous Integration tests of various adjacent Spark data converters as well as configuration files for things like Presto, Ambari, Apache Drill etc found in my DockerHub images Dockerfiles master repo which contains docker builds and configurations for many open source Big Data & Linux technologies

Detailed Build Instructions

Python VirtualEnv localized installs

The automated build will use 'sudo' to install required Python PyPI libraries to the system unless running as root or it detects being inside a VirtualEnv. If you want to install some of the common Python libraries using your OS packages instead of installing from PyPI then follow the Manual Build section below.

Manual Setup

Enter the pytools directory and run git submodule init and git submodule update to fetch my library repo:

git clone https://github.com/harisekhon/devops-python-tools pytools
cd pytools
git submodule init
git submodule update
sudo pip install -r requirements.txt

Offline Setup

Download the DevOps Python Tools and Pylib git repos as zip files:

https://github.com/HariSekhon/devops-python-tools/archive/master.zip

https://github.com/HariSekhon/pylib/archive/master.zip

Unzip both and move Pylib to the pylib folder under DevOps Python Tools.

unzip devops-python-tools-master.zip
unzip pylib-master.zip

mv -v devops-python-tools-master pytools
mv -v pylib-master pylib
mv -vf pylib pytools/

Proceed to install PyPI modules for whichever programs you want to use using your usual procedure - usually an internal mirror or proxy server to PyPI, or rpms / debs (some libraries are packaged by Linux distributions).

All PyPI modules are listed in the requirements.txt file.

Internal Mirror example (JFrog Artifactory or similar):

sudo pip install --index https://host.domain.com/api/pypi/repo/simple --trusted host.domain.com -r requirements.txt

Proxy example:

sudo pip install --proxy hari:mypassword@proxy-host:8080 -r requirements.txt

Mac OS X

The automated build also works on Mac OS X but you'll need to download and install Apple XCode. I also recommend you get HomeBrew to install other useful tools and libraries you may need like OpenSSL for development headers and tools such as wget (these are installed automatically if Homebrew is detected on Mac OS X):

brew install openssl wget

If failing to build an OpenSSL lib dependency, just prefix the build command like so:

sudo OPENSSL_INCLUDE=/usr/local/opt/openssl/include OPENSSL_LIB=/usr/local/opt/openssl/lib ...

You may get errors trying to install to Python library paths even as root on newer versions of Mac, sometimes this is caused by pip 10 vs pip 9 and downgrading will work around it:

sudo pip install --upgrade pip==9.0.1
make
sudo pip install --upgrade pip
make

Jython for Hadoop Utils

The 3 Hadoop utility programs listed below require Jython (as well as Hadoop to be installed and correctly configured)

hadoop_hdfs_time_block_reads.jy
hadoop_hdfs_files_native_checksums.jy
hadoop_hdfs_files_stats.jy

Run like so:

jython -J-cp $(hadoop classpath) hadoop_hdfs_time_block_reads.jy --help

The -J-cp $(hadoop classpath) part dynamically inserts the current Hadoop java classpath required to use the Hadoop APIs.

See below for procedure to install Jython if you don't already have it.

Automated Jython Install

This will download and install jython to /opt/jython-2.7.0:

make jython

Manual Jython Install

Jython is a simple download and unpack and can be fetched from http://www.jython.org/downloads.html

Then add the Jython install bin directory to the $PATH or specify the full path to the jython binary, eg:

/opt/jython-2.7.0/bin/jython hadoop_hdfs_time_block_reads.jy ...

Configuration for Strict Domain / FQDN validation

Strict validations include host/domain/FQDNs using TLDs which are populated from the official IANA list is done via my PyLib library submodule - see there for details on configuring this to permit custom TLDs like .local, .intranet, .vm, .cloud etc. (all already included in there because they're common across companies internal environments).

Python SSL certificate verification problems

If you end up with an error like:

./dockerhub_show_tags.py centos ubuntu
[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed (_ssl.c:765)

It can be caused by an issue with the underlying Python + libraries due to changes in OpenSSL and certificates. One quick fix is to do the following:

sudo pip uninstall -y certifi &&
sudo pip install certifi==2015.04.28

Updating

Run make update. This will git pull and then git submodule update which is necessary to pick up corresponding library updates.

If you update often and want to just quickly git pull + submodule update but skip rebuilding all those dependencies each time then run make update-no-recompile (will miss new library dependencies - do full make update if you encounter issues).

Testing

Continuous Integration is run on this repo with tests for success and failure scenarios:

• unit tests for the custom supporting python library
• integration tests of the top level programs using the libraries for things like option parsing
• functional tests for the top level programs using local test data and Docker containers

To trigger all tests run:

make test

which will start with the underlying libraries, then move on to top level integration tests and functional tests using docker containers if docker is available.

Contributions

Patches, improvements and even general feedback are welcome in the form of GitHub pull requests and issue tickets.

See Also

• DevOps Perl Tools - 25+ tools for Hadoop, NoSQL, Solr, Elasticsearch, Pig, Hive, Web URL + Nginx stats watchers, SQL and NoSQL syntax recasers, various Linux CLI tools

• The Advanced Nagios Plugins Collection - 400+ programs for Nagios monitoring your Hadoop & NoSQL clusters. Covers every Hadoop vendor's management API and every major NoSQL technology (HBase, Cassandra, MongoDB, Elasticsearch, Solr, Riak, Redis etc.) as well as message queues (Kafka, RabbitMQ), continuous integration (Jenkins, Travis CI) and traditional infrastructure (SSL, Whois, DNS, Linux)

• PyLib - Python library leveraged throughout the programs in this repo as a submodule

• Perl Lib - Perl version of above library

• Spark Apps eg. Spark => Elasticsearch - Scala application to index from Spark to Elasticsearch. Used to index data in Hadoop clusters or local data via Spark standalone. This started as a Scala Spark port of pig-text-to-elasticsearch.pig from this repo.

You might also be interested in the following really nice Jupyter notebook for HDFS space analysis created by another Hortonworks guy Jonas Straub:

• https://github.com/mr-jstraub/HDFSQuota/blob/master/HDFSQuota.ipynb