[HADOOP-14383] Implement FileSystem that reads from HTTP / HTTPS endpoints

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.9.0, 3.0.0-alpha4
• Component/s: fs
• Hadoop Flags: Reviewed
• Assignee: Haohui Mai
• Reporter: Haohui Mai
• Created: 04/May/17 01:11
• Updated: 18/Apr/19 16:07
• Resolved: 08/May/17 22:36

Description

We have a use case where YARN applications would like to localize resources from Artifactory. Putting the resources on HDFS itself might not be ideal as we would like to leverage Artifactory to manage different versions of the resources.

It would be nice to have something like HttpFileSystem that implements the Hadoop filesystem API and reads from a HTTP endpoint.

Note that Samza has implemented the proposal by themselves:

https://github.com/apache/samza/blob/master/samza-yarn/src/main/scala/org/apache/samza/util/hadoop/HttpFileSystem.scala

The downside of this approach is that it requires the YARN cluster to put the Samza jar into the classpath for each NM.

It would be much nicer for Hadoop to have this feature built-in.

Issue Links

• breaks: HADOOP-14598 Blacklist Http/HttpsFileSystem in FsUrlStreamHandlerFactory
• relates to: HADOOP-16256 HTTPFileSystem disallows users from configuring the connectionTimeout and readTimeout of HttpClient.