Monday, 14 September 2015

Initial Support for Apache Avro and Gora

Avro and Gora are two Apache projects that belong to the Hadoop ecosystem. Avro is a data serialization framework that relies on JSON for defining data types and protocols, and serializes data in a compact binary format. Its primary use in Hadoop is to provide a serialization format for persistent data, and a wire format for communication between Hadoop nodes, and from client programs to the Hadoop services. Gora is an open-source software framework that provides an in-memory data model and persistence for big data. Gora supports persisting to column stores, key/value stores or databases, and analyzing the data with extensive Apache Hadoop MapReduce support.

As an effort to run Hadoop based applications atop Infinispan, the LEADS EU FP7 project has developed an Avro backend (infinispan-avro) and a Gora module (gora-infinispan). The former allows to store, retrieve and query Avro defined types via the HotRod protocol. The latter allows Gora-based applications to use Infinispan as a storage backend for their MapReduce jobs. In the current state of the implementation, the two modules make use of Infinispan 8.0.0.Final, Avro 1.7.6 and Gora 0.6

What’s in it for you Infinispan user

There are several use cases for which you can benefit from those modules.
  • With Infinispan’s Avro support, you can decide to persist your data in Infinispan using Avro’s portable format instead of Infinispan’s own format (or Java serialization’s format). This might help you standardize upon a common format for your data at rest. 
  • If you use Apache Gora to store/query some of your data in, or even out, of the Hadoop ecosystem, you can use Infinispan as the backend and benefit Infinispan’s features that you come to know like data distribution, partition handling, cross-site clustering. 
  • The last use case is to run legacy Hadoop applications, using Infinispan as the primary storage. For instance, it is possible to run the Apache Nutch web crawler atop Infinispan. A recent paper at IEEE Cloud 2015 gives a detailed description of such an approach in a geo-distributed environment (a preprint is available here).