[MESOS-4255] Add mechanism for testing recovery of HTTP based executors

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.28.0
• Labels: mesosphere
• Target Version/s: 0.28.0
• Epic Link: Executor API v1
• Sprint: Mesosphere Sprint 28
• Story Points: 5
• Assignee: Anand Mazumdar
• Reporter: Anand Mazumdar
• Shepherd: Vinod Kone
• Created: 29/Dec/15 03:51
• Updated: 26/Nov/18 12:20
• Resolved: 26/Nov/18 12:20

Description

Currently, the slave process generates a process ID every time it is initialized via process::ID::generate function call. This is a problem for testing HTTP executors as it can't retry if there is a disconnection after an agent restart since the prefix is incremented.

Agent PID before:
slave(1)@127.0.0.1:43915

Agent PID after restart:
slave(2)@127.0.0.1:43915

There are a couple of ways to fix this:

• Add a constructor to Slave exclusively for testing that passes on a fixed ID instead of relying on ID::generate.
• Currently we delegate to slave(1)@ i.e. (1) when nothing is specified as the URL in libprocess i.e. 127.0.0.1:43915/api/v1/executor would delegate to slave(1)@127.0.0.1:43915/api/v1/executor. Instead of defaulting to (1), we can default to the last known active ID.

Issue Links

• is blocked by: MESOS-4425 Introduce filtering test abstractions for HTTP events to libprocess