tree fafe3258a74a97171a9d11d917a19c2e7387db73
parent 1f9a03b3f952320824b1ae49e56da3cb814cd5b0
author Serge Bazanski <serge@monogon.tech> 1629716724 +0200
committer Sergiusz Bazanski <serge@monogon.tech> 1629825995 +0000

m/n/core/curator: fix listener stuckness on restarts

This makes both gRPC listener runnables (local and public) manage their
own listening sockets, allowing them to restart independently of
eachother, and making sure that any listening sockets are cleaned up.

We also fix the existing curator test (which does not exercise the
listeners, just leadership election) to place the curators and their
local sockets in /tmp instead of the default bazel tempdir (as a path
based on that is longer than the maximum domain socket path). This makes
these tests slightly less noisy (as they kept crashing while not being
able to listen to the local socket).

This should've been caught by a curator listener test, if we had one
(other than the e2e test). I'm growing keen on spending some time
writing enough of a harness to actually do that. Maybe once we have a
follower implementation ready…

Change-Id: I0267292781b6ee8aff1d0557d420bbaa3c3d79f6
Reviewed-on: https://review.monogon.dev/c/monogon/+/304
Reviewed-by: Lorenz Brun <lorenz@monogon.tech>