There are cases where hab sup term stops working

[jaym]

I ran into a situation where hab sup term refuses to bring things down. It seems to be some sort of race condition with unloading a service. Here's the test script i've been using:

termtest() {
    if ! hab pkg path core/procps-ng; then
        hab pkg install --binlink core/procps-ng
    fi

    if ! hab pkg path "$HAB_ORIGIN/term-test"; then
        build .
    fi

    if ! hab svc load "$HAB_ORIGIN/term-test"; then
        echo "Failed to load term-test"
        return 1
    fi

    sleep 5

    hab svc unload "$HAB_ORIGIN/term-test" & hab sup term &

    sleep 3

    if pidof hab-launch; then
        echo "hab should have died but didn't"
        return 1
    fi

    return 0
}

You can clone https://github.com/jaym/hab-term-test and test it out by:

hab studio enter
termtest

You'll find that after running that (most of the time), hab-launch sticks around. It will not respond to hab sup term.

[stevendanna]

@jaym Do you know what versions of hab-launcher and hab-sup were in use here?

[jaym]

core/hab-launcher/8744/20181016214636
hab-sup/0.59.0/20180712161546

also tried hab-sup 0.66 and had the same problem

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. We value your input and contribution. Please leave a comment if this issue still affects you.

[stevendanna]

I'm running the test from @jaym and don't seem to be hitting this bug anymore. I wonder if this was fixed by the various deadlock fixes we've shipped since version 0.66. Specifically, we had the signal-handling deadlock that I think could be the cause of something like this. I'll leave my reproduction loop running, but so far it has only found false-positive (cases where the termination takes a bit more than 3 seconds)

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. We value your input and contribution. Please leave a comment if this issue still affects you.