[MESOS-8828] Clock::advance can race with process::delay in tests.

• Type: Bug
• Status: Accepted
• Priority: Major
• Resolution: Unresolved
• Labels: flaky integration mesosphere
• Epic Link: test-harness
• Assignee: Unassigned
• Reporter: Andrei Budnik
• Created: 24/Apr/18 14:34
• Updated: 29/Apr/19 09:27

Description

There are lots of tests that use the following pattern:
1) Pause clocks
2) Start an agent
3) Advance clocks to trigger an event
4) Wait for the event

If an event is scheduled via `process::delay()` after advancing the clocks, then a test hangs in the endless wait for the event that is never triggered, because libprocess clocks are paused. For example, `DiskResource/PersistentVolumeTest.SharedPersistentVolumeRescindOnDestroy/0` test hangs at step 4, because the clocks at step 3 has been already advanced before the agent scheduled a call of Slave::authenticate() method. After a successful authentication with a master, the agent sends a UpdateSlaveMessage. But the authentication process never finishes because `Slave::authenticate()` is never called.

A list of tests that might be affected by the issue attached to this ticket.

Issue Links

• blocks: MESOS-9139 Use composing containerizer by default in tests.
• relates to: MESOS-8732 Use composing containerizer in some agent tests.