[HDDS-1017] Use distributed tracing to indentify performance problems in Ozone

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.4.0
• Labels: pull-request-available
• Target Version/s: 0.4.0
• Sprint: HDDS BadLands
• Assignee: Marton Elek
• Reporter: Marton Elek
• Created: 28/Jan/19 12:51
• Updated: 21/Feb/19 21:32
• Resolved: 11/Feb/19 11:30

Description

In the recent months multiple performance issues are resolved in OM/SCM and datanode sides. To identify the remaining problems a distributed tracing framework would help a lot.

In HADOOP-15566 there is an ongoing discussion to remove the discontinued HTrace and use something else instead. Until now without any conclusion, but

1). There is one existing poc in the jira which uses opentracing
2). It was suggested to "evaluating all the options" before a final decision

As an evaluation step we would like to investigate the performance of ozone components with opentracing. This patch can help us to find the performance problem but can be reverted when we will have a final solution in HADOOP-15566 about the common tracing library.

To make it lightweight we can use the ozone message level tracing identifier for context propagation instead of modifying the existing hadoop rpc framework.

Issue Links

• links to: GitHub Pull Request #472