The genesis of the best SQL engine in Hadoop is not an overnight, “me too” product. It seems that it wasn’t too long ago the Hadoop vendors all but wrote off SQL but the demand just hasn’t gone away.
HAWQ is the result of many, many years of work leveraging open source software as well as contributing back to the open source community. I think a short history lesson is in order to fully understand how this product came to be.
Greenplum Database began life in 2003. The founders used open source PostgreSQL and released a commercial product soon after.
Bizgres, an open source version of Greenplum, was released in 2005. Very early on, the founders of Greenplum embraced contributing back to the open source community.
Madlib was released in 2010 as an open source project which later became an Apache Incubator project.
Greenplum was acquired by EMC in 2010 and almost immediately, EMC invested heavily into Hadoop. The Greenplum division was agile like a small startup company but with the deep pockets of a multi-billion dollar company.
Greenplum released a Hadoop distribution in 2011 and integration between Greenplum Database and HDFS got more robust with the introduction of “gphdfs”. Greenplum supported External Tables to read/write data in parallel to HDFS from several different distributions.
HAWQ, a fork of Greenplum Database, was released in 2013. HAWQ was immediately extremely performant and compliant with the newest SQL syntax. HAWQ borrowed from the 10 years experience of developing Greenplum to provide a robust optimizer designed for HDFS.
2013 also saw Pivotal become a company. EMC contributed Greenplum Database, VMWare contributed Gemfire and Cloud Foundry, and GE contributed capital as an active partner. Paul Maritiz became the CEO and the dedication to fully embrace open source became an integral part of the corporate culture.
During the last three years, HAWQ has become an Apache Incubator Project. The Pivotal product is now a rather boring name of “Pivotal HDB” while HAWQ is the name of the Apache project.
Pivotal also made Greenplum and Geode (Gemfire is the commercial product name) open source projects too. Clearly, Pivotal has embraced open source with probably more committers than those other “open source” data companies.
So what now? What is happening in 2016? Well, Pivotal is about to release Pivotal HDB (HAWQ) 2.0. I’ve been testing this product for months on various platforms and I keep getting amazed by the performance and ease of use.
HAWQ 2.0 embraces Hadoop fully. I believe the two biggest features are elasticity and performance. HAWQ now supports elasticity for growing or shrinking clusters without having to redistribute the data. The performance is also much improved as it better utilizes HDFS and YARN.
So you may be asking, “is it fast?” and the answer is yes! I haven’t found a SQL engine that is faster and I’ve been doing competitive analysis for months. The other question you may ask is, “can I run my SQL?” and the answer is yes! A major competitor in the SQL on Hadoop landscape requires more tweaking of SQL just to get the SQL to execute and more tuning to get decent performance.
That “other SQL on Hadoop” product can’t do the following things, as well as many more, that HAWQ can.
– Can’t handle a comment line at the end of SQL file (I found that rather strange)
– Won’t do partition elimination through a joined table (e.g. date dimension)
– Can’t get number of milliseconds when subtracting from a date. Only gets seconds.
– Has “interval” in SQL dialect but doesn’t have interval as a type.
– No time data type.
– Concatenating strings doesn’t use || or +. You must use concat() function.
– Doesn’t support intersect or except.
– Doesn’t support subqueries that return more than one row.
– Doesn’t support correlated subqueries.
– Doesn’t support group by rollup.
– Doesn’t support subqueries in having statement.
– Subqueries not supported in select list.
HAWQ is the real deal. It is an Apache project, going to be part of ODPi, faster than everyone else, integrated with Apache Ambari, certified with Hortonworks, and the most mature SQL engine for Hadoop.