Nyc hadoop meetup introduction to h base
6 likes1,878 views
Jonathan Gray gave an introduction to HBase at the NYC Hadoop Meetup. He began with an overview of HBase and why it was created to handle large datasets beyond what Hadoop could support alone. He then described what HBase is, as a distributed, column-oriented database management system. Gray explained how HBase works with its master and regionserver nodes and how it partitions data across tables and regions. He highlighted some key features of HBase and examples of companies using it in production. Gray concluded with what is planned for the future of HBase and contrasted it with relational database examples.
![HBase Tables
? An HBase cluster is made up of any number of user-
defined tables
? Table schema only defines it¡¯s column families
¨C Each family consists of any number of columns
¨C Each column consists of any number of versions
¨C Columns only exist when inserted, NULLs are free
¨C Everything except table/family names are byte[]
¨C Rows in a table are sorted and stored sequentially
¨C Columns in a family are sorted and stored sequentially
(Table, Row, Family, Column, Timestamp) ? Value](https://image.slidesharecdn.com/nychadoopmeetup-introductiontohbase-100928233306-phpapp02/85/Nyc-hadoop-meetup-introduction-to-h-base-8-320.jpg)
More Related Content
What's hot (20)
Similar to Nyc hadoop meetup introduction to h base (20)
Nyc hadoop meetup introduction to h base
- 1. Introduction to HBase NYC Hadoop Meetup Jonathan Gray February 11, 2010
- 2. About Me ? Jonathan Gray ¨C HBase Committer ¨C HBase User since early 2008 ¨C Migrated large PostgreSQL instance to HBase ? In production @ streamy.com since June 2008 ¨C Core contributor to performance improvements in HBase 0.20 ¨C Currently consulting around HBase ? As well as Hadoop/MR and Lucene/Katta
- 3. Overview ? Why HBase? ? What is HBase? ? How does HBase work? ? HBase Today and Tomorrow ? HBase vs. RDBMS Example ? HBase and ¡°NoSQL¡±
- 4. Why HBase? ? Same reasons we need Hadoop ¨C Datasets growing into Terabytes and Petabytes ¨C Scaling out is cheaper than scaling up ? Continue to grow just by adding commodity nodes ? But sometimes Hadoop is not enough ¨C Need to support random reads and random writes Traditional databases are expensive to scale and difficult to distribute
- 5. What is HBase? ? Distributed ? Column-Oriented ? Multi-Dimensional ? High-Availability ? High-Performance ? Storage System Project Goal Billions of Rows * Millions of Columns * Thousands of Versions Petabytes across thousands of commodity servers
- 6. HBase is not¡ ? A Traditional SQL Database ¨C No joins, no query engine, no types, no SQL ¨C Transactions and secondary indexing possible but these are add-ons, not part of core HBase ? A drop-in replacement for your RDBMS ? You must be OK with RDBMS anti-schema ¨C Denormalized data ¨C Wide and sparsely populated tables Just say ¡°no¡± to your inner DBA
- 7. How does HBase work? ? Two types of HBase nodes: Master and RegionServer ? Master (one at a time) ¨C Manages cluster operations ? Assignment, load balancing, splitting ¨C Not part of the read/write path ¨C Highly available with ZooKeeper and standbys ? RegionServer (one or more) ¨C Hosts tables; performs reads, buffers writes ¨C Clients talk directly to them for reads/writes
- 8. HBase Tables ? An HBase cluster is made up of any number of user- defined tables ? Table schema only defines it¡¯s column families ¨C Each family consists of any number of columns ¨C Each column consists of any number of versions ¨C Columns only exist when inserted, NULLs are free ¨C Everything except table/family names are byte[] ¨C Rows in a table are sorted and stored sequentially ¨C Columns in a family are sorted and stored sequentially (Table, Row, Family, Column, Timestamp) ? Value
- 9. HBase Table as Data Structures ? A table maps rows to its families ¨C SortedMap(Row ? List(ColumnFamilies)) ? A family maps column names to versioned values ¨C SortedMap(Column ? SortedMap(VersionedValues)) ? A column maps timestamps to values ¨C SortedMap(Timestamp ? Value) An HBase table is a three-dimensional sorted map (row, column, and timestamp)
- 10. HBase Regions ? Table is made up of any number of regions ? Region is specified by its startKey and endKey ¨C Empty table: (Table, NULL, NULL) ¨C Two-region table: (Table, NULL, ¡°MidKey¡±) and (Table, ¡°MidKey¡±, NULL) ? A region only lives on one RegionServer at a time ? Each region may live on a different node and is made up of several HDFS files and blocks, each of which is replicated by Hadoop
- 11. More HBase Architecture ? Region information and locations stored in special tables called catalog tables -ROOT- table contains location of meta table .META. table contains schema/locations of user regions ? Location of -ROOT- is stored in ZooKeeper ¨C This is the ¡°bootstrap¡± location ? ZooKeeper is used for coordination / monitoring ¨C Leader election to decide who is master ¨C Ephemeral nodes to detect RegionServer node failures
- 14. HBase Key Features ? Automatic partitioning of data ¨C As data grows, it is automatically split up ? Transparent distribution of data ¨C Load is automatically balanced across nodes ? Tables are ordered by row, rows by column ¨C Designed for efficient scanning (not just gets) ¨C Composite keys allow ORDER BY / GROUP BY ? Server-side filters ? No SPOF because of ZooKeeper integration
- 15. HBase Key Features (cont) ? Fast adding/removing of nodes while online ¨C Moving locations of data doesn¡¯t move data ? Supports creating/modifying tables online ¨C Both table-level and family-level configuration parameters ? Close ties with Hadoop MapReduce ¨C TableInputFormat/TableOutputFormat ¨C HFileOutputFormat
- 16. Connecting to HBase ? Native Java Client/API ¨C Get, Scan, Put, Delete classes ¨C HTable for read/write, HBaseAdmin for admin stuff ? Non-Java Clients ¨C Thrift server (Ruby, C++, PHP, etc) ¨C REST server (stargate contrib) ? HBase Shell ¨C Jruby shell supports put, delete, get, scan ¨C Also supports administrative tasks ? TableInputFormat/TableOutputFormat
- 17. HBase Add-ons ? MapReduce / Cascading / Hive / Pig ¨C Support for HBase as a data source or sink ? Transactional HBase ¨C Distributed transactions using OCC ? Indexed HBase ¨C Utilizes Transactional HBase for secondary indexing ? IHbase ¨C New contrib for in-memory secondary indexes ? HBql ¨C SQL syntax on top of HBase
- 18. HBase Today ? Latest stable release is HBase 0.20.3 ¨C Major improvement over HBase 0.19 ¨C Focus on performance improvement ¨C Add ZooKeeper, remove SPOF ¨C Expansion of in-memory and caching capabilities ¨C Compatible with Hadoop 0.20.x ¨C Recommend upgrading from earlier 0.20.x HBase releases as 0.20.3 includes some important fixes ? Improves logging, shell, faster cluster ops, stability
- 19. HBase in Production ? Streamy ? StumbleUpon ? Adobe ? Meetup ? Ning ? Openplaces ? Powerset ? SocialMedia.com ? TrendMicro
- 20. The Future of HBase ? Next release is HBase 0.21.0 ¨C Release date will be ~1 month after Hadoop 0.21 ? Data durability is fixed in this release ¨C HDFS append/sync finally works in Hadoop 0.21 ¨C This is implemented and working on TRUNK ¨C Have added group commit and knobs to adjust ? Other cool features ¨C Inter-cluster replication ¨C Master Rewrite ¨C Parallel Puts ¨C Co-processors
- 21. HBase Web Crawl Example ? Store web crawl data ¨C Table crawl with family content ¨C Row is URL with Columns ? content:data stores raw crawled data ? content:language stores http language header ? content:type stores http content-type header ¨C If processing raw data for hyperlinks and images, add families links and images ? links:<url> column for each hyperlink ? images:<url> column for each image
- 22. Web Crawl Example in RDBMS ? How would this look in a traditional DB? ¨C Table crawl with columns url, data, language, and type ¨C Table links with columns url and link ¨C Table images with columns url and image ? How will this scale? ¨C 10M documents w/ avg10 links and 10 images ¨C 210M total rows versus 10M total rows ¨C Index bloat with links/images tables
- 23. What is ¡°NoSQL¡±? ? Has little to do with not being SQL ¨C SQL is just a query language standard ¨C HBql is an attempt to add SQL syntax to HBase ¨C Millions are trained in SQL; resistance is futile! ? Popularity of Hive and Pig over raw MapReduce ? Has more to do with anti-RDBMS architecture ¨C Dropping the relational aspects ¨C Loosening ACID and transactional elements
- 24. NoSQL Types and Projects ? Column-oriented ¨C HBase, Cassandra, Hypertable ? Key/Value ¨C BerkeleyDB, Tokyo, Memcache, Redis, SimpleDB ? Document ¨C CouchDB, MongoDB ? Other differentiators as well¡ ¨C Strong vs. Eventual consistency ¨C Database replication vs. Filesystem replication