[MESOS-9298] Task failures sometimes can't be understood without looking into agent logs.

• Type: Epic
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: scheduler api
• Labels: integration mesosphere observability
• Epic Name: task-failure-reasons
• Assignee: Unassigned
• Reporter: Alex R
• Created: 07/Oct/18 14:18
• Updated: 29/Apr/19 09:27

Description

Mesos communicates task state transitions via task status updates. They often include a reason, which aims to hint what exactly went wrong. However, these reasons are often:

• misleading
• vague
• generic.
  Needless to say, this complicates triaging why the task has actually failed and hence is a bad user experience. The failures can come from a bunch of different sources: fetcher, isolators (including custom ones!), namespace setup, etc.

This epic aims to improve the UX by providing detailed, ideally typed, information about task failures.

Issue Links

• is related to: MESOS-8531 Some task status updates sent by the default executor don't contain a REASON.
• is related to: MESOS-2077 Ensure that TASK_LOSTs for a hard slave drain (SIGUSR1) include a Reason.