Nomad version

Nomad v0.9.3 (c5e8b66c3789e4e7f9a83b4e188e9a937eea43ce)

Operating system and Environment details

Amazon Linux 2

Issue

Non-leader task in taskgroup receives no interrupt when Nomad decides an allocation should be stopped, say, from a count decrement.

The leader task, which is the task where all the services are registered to Consul, properly handles the shutdown_delay and kill_timeout, but the secondary task receives no interrupt until the leader task dies. Then it receives an interrupt because the leader task is dead.

This is a problem because the leader task is the sidecar proxy which is the source network for the secondary task, the app itself. Since it receives no interrupts, it is unable to perform graceful shutdown actions until it receives its interrupt, which is when the network is gone since it was attached to the leader task.

Reproduction steps

1. Create a TaskGroup with two tasks, both using the docker driver
2. Set one task as leader
3. Set "network_mode": "container:<task_name>-${NOMAD_ALLOC_ID}" for the secondary
4. Register the healthchecks on the leader
5. Submit the job
6. Reduce the count of the job
7. Verify the leader task gets an interrupt, verify the secondary task gets no interrupt
8. After the shutdown_delay and leader terminates, verify the secondary task detects leader is gone
9. Verify the secondary task is sent its interrupt

Leader Task/Proxy:

Secondary Task/App

I believe the expected behavior is every task in a taskgroup should receive an interrupt at the same time.

I am seeing this as well - with the latest Nomad version too (Nomad v0.12.4 (8efaee4ba5e9727ab323aaba2ac91c2d7b572d84)). This issue should not have been closed by the bot. Is anybody looking at this?