[UIMA-3737] DUCC Agent should not try to remove cgroup when kill -9 fails to stop a process

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.0.0-Ducc
• Fix Version/s: 1.1.0-Ducc
• Component/s: DUCC
• Assignee: Jaroslaw Cwiklik
• Reporter: Jaroslaw Cwiklik
• Created: 10/Apr/14 15:31
• Updated: 30/Apr/14 17:45
• Resolved: 30/Apr/14 17:45

Description

When an agent starts up it tries to clean up the node. It first waits for the OR publications and then checks the node for running processes. It then reconciles what is running against what should be running (using OR state). Any processes not in the OR state should be evicted with kill -9 and its cgroup removed.
If there is a defunct process on a node, it cannot be killed. Such process still has associated cgroup which cannot be removed. Its been observed that the agent's attempt to remove the cgroup of a defunct process causes the CPU utilization to spike for a couple of minutes.
Agent should detect if a process is defunct and not try to remove its cgroup