[REEF-1837] IsParentAliveCancellationDetector logs lack of HTTP endpoint excessively

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Invalid
• Component/s: REEF.NET Driver
• Assignee: Unassigned
• Reporter: Markus Weimer
• Created: 24/Jul/17 15:54
• Updated: 07/Aug/17 16:45
• Resolved: 07/Aug/17 16:45

Description

In IsParentAliveCancellationDetector, we generate a log line each time an upstream HTTP server cannot be found, e.g.:

INFO: Using Yarn Configuration to retrieve RM address.
Microsoft.MachineLearning.Distributed.Core.Common.IsParentAliveCancellationDetector Error: 0 : 2017-07-23T18:00:28.6024001-07:00 0004
ERROR: Failed to initialize yarn configuration
Encountered error [System.ApplicationException: Unable to find RM Webapp Address from yarn-site.xml.
   at Org.Apache.REEF.Utilities.Runtime.Yarn.YarnConfiguration.GetYarnRMWebappEndpoints()
   at Microsoft.MachineLearning.Distributed.Core.Common.IsParentAliveCancellationDetector.InitYarnConfiguration(String rmAddresses)]

This leads to excessively bloated logs. We should reduce the amount of logging here, e.g. by assuming that the state of the YARN deployment doesn't change during execution of one job.