(address . bug-guix@gnu.org)
871q8xh5ja.fsf@inria.fr
Hello,
Substitute servers discovered over Avahi don’t disappear when they go
off-line: they remain in /var/guix/discover/publish “forever” (?) and
‘guix substitute’ keeps trying to connect to them, in vain.
The problem AIUI is that the mDNS records are still valid according to
their TTL and thus cached by avahi-daemon; we can only notice that
things go wrong when trying to resolve:
Toggle snippet (7 lines)
$ avahi-browse _guix_publish._tcp -r
+ enp0s31f6 IPv6 guix-publish-xyz _guix_publish._tcp local
+ enp0s31f6 IPv4 guix-publish-xyz _guix_publish._tcp local
Failed to resolve service 'guix-publish-xyz' of type '_guix_publish._tcp' in domain 'local': Timeout reached
Failed to resolve service 'guix-publish-xyz' of type '_guix_publish._tcp' in domain 'local': Timeout reached
(One way to reproduce that is to ‘pkill -9 avahi-daemon’ on the machine
that advertises the service.)
Problem is that ‘avahi-browse-service-thread’ attempts to resolve
services once for all; thus it never notices that services went away, at
least not until some long TTL has expired.
Ludo’.