[HADOOP-9629] Support Windows Azure Storage - Blob as a file system in Hadoop

• Type: New Feature
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.7.0
• Component/s: tools
• Target Version/s: 2.7.0
• Hadoop Flags: Reviewed
• Release Note: Hadoop now supports integration with Azure Storage as an alternative Hadoop Compatible File System.
• Assignee: Chris Nauroth
• Reporter: Mostafa Elhemali
• Created: 07/Jun/13 15:47
• Updated: 23/Jul/18 21:39
• Resolved: 18/Dec/14 00:18

Description

Description

This JIRA incorporates adding a new file system implementation for accessing Windows Azure Storage - Blob from within Hadoop, such as using blobs as input to MR jobs or configuring MR jobs to put their output directly into blob storage.

High level design

At a high level, the code here extends the FileSystem class to provide an implementation for accessing blob storage; the scheme wasb is used for accessing it over HTTP, and wasbs for accessing over HTTPS. We use the URI scheme:

wasb[s]://<container>@<account>/path/to/file

to address individual blobs. We use the standard Azure Java SDK (com.microsoft.windowsazure) to do most of the work. In order to map a hierarchical file system over the flat name-value pair nature of blob storage, we create a specially tagged blob named path/to/dir whenever we create a directory called path/to/dir, then files under that are stored as normal blobs path/to/dir/file. We have many metrics implemented for it using the Metrics2 interface. Tests are implemented mostly using a mock implementation for the Azure SDK functionality, with an option to test against a real blob storage if configured (instructions provided inside in README.txt).

Credits and history

This has been ongoing work for a while, and the early version of this work can be seen in HADOOP-8079. This JIRA is a significant revision of that and we'll post the patch here for Hadoop trunk first, then post a patch for branch-1 as well for backporting the functionality if accepted. Credit for this work goes to the early team: minwei, davidlao, lengningliu and stojanovic as well as multiple people who have taken over this work since then (hope I don't forget anyone): dexterb, Johannes Klein, ivanmi, Michael Rys, mostafae, brian_swan, mikelid, xifang, and chuanliu.

Test

Besides unit tests, we have used WASB as the default file system in our service product. (HDFS is also used but not as default file system.) Various different customer and test workloads have been run against clusters with such configurations for quite some time. The current version reflects to the version of the code tested and used in our production environment.

Issue Links

• is related to: HADOOP-10840 Fix OutOfMemoryError caused by metrics system in Azure File System
• is related to: HADOOP-11248 Add hadoop configuration to disable Azure Filesystem metrics collection
• is related to: HADOOP-10689 InputStream is not closed in AzureNativeFileSystemStore#retrieve()
• is related to: HADOOP-10690 Lack of synchronization on access to InputStream in NativeAzureFileSystem#NativeAzureFsInputStream#close()
• is related to: HADOOP-10728 Metrics system for Windows Azure Storage Filesystem
• is related to: HADOOP-10809 hadoop-azure: page blob support
• is related to: HADOOP-11188 hadoop-azure: automatically expand page blobs when they become full
• is related to: HADOOP-11395 Add site documentation for Azure Storage FileSystem integration.