[AURORA-1533] Transient connection errors can leave client in irrecoverable state

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 0.10.0
• Assignee: Stephan Erb
• Reporter: Stephan Erb
• Created: 02/Nov/15 12:56
• Updated: 10/Nov/15 17:59
• Resolved: 02/Nov/15 18:54

Description

During a cluster update, some of our schedulers returned an unknown error to connecting clients (relevant code). Long running clients failed to recover from these errors as the code assumed the connection was already established. Subsequent scheduling calls thus failed with the following exception:

File  "venv/local/lib/python2.7/site-packages/apache/aurora/client/api/__init__.py"  in query_no_configs
  140.       raise self.ThriftInternalError(e.args[0])

Exception Type: ThriftInternalError
Exception Value: Error during thrift call getTasksWithoutConfigs to 
testcluster: 'NoneType' object has no attribute 'getTasksWithoutConfigs'

Background: We are using the python client to dispatch calls to Aurora from within a long-running web service. The connection is kept open as long as the web service is running.