[HDDS-1214] Enable tracing for the datanode read/write path

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Component/s: Ozone Datanode
• Labels: pull-request-available
• Target Version/s: 0.4.0
• Assignee: Marton Elek
• Reporter: Marton Elek
• Created: 04/Mar/19 14:23
• Updated: 12/Mar/19 10:49
• Resolved: 12/Mar/19 10:37

Description

HDDS-1150 introduced distributed for ozone components. But we have no trace context propagation between the clients and Ozone Datanodes.

As we use Grpc and Ratis on this RPC path the full tracing could be quite complex: we should propagate the trace id in Ratis and include it in all the log entries.

I propose a simplified solution here: to trace only the StateMachine operations.

As Ratis is a library we provide the implementation of the appropriate Raft elements especially the StateMachine and the raft messages. We can add the tracing information to the raft messages (in fact, we already have this field) and we can restore the tracing context during the StateMachine operations.

This approach is very simple (only a few lines of codes) and can show the time of the real write/read operations, but can't see the internals of the Ratis operations.

Issue Links

• links to: GitHub Pull Request #550