"docker.sock" is used to access Docker API which is required to:
1. subscribe to Docker events to configure-arp,
2. query Docker network info to detect whether multicast should be
enabled.
3. get list of Swarm peers before starting the plugin.
(1) is currently not implemented.
(2) is replaced by WEAVE_MULTICAST env var which
can enable multicast for all Weave networks created by
the plugin (not ideal, but better than nothing).
(3) is fixed by not removing "restart.sentinel".
- "noDNS" controls whether upon ContainerStarted event a container
should be registered within the Weave DNS.
- "isPluginV2" will be used to determine a path to host's /proc.
This simplifies the following:
- plugin-v2 does not need to start "plugin" as a separate process, thus
Docker is able to properly restart plugin-v2 in a case of any child
process failing.
- No need for weaveplugin container.
plugin is enabled by default when running "weave launch". when starting
only the router, you should pass "--plugin" parameter to enable it.
The env var WEAVE_NO_PLUGIN is no more.
doing it from the script
Note we need to run weaver in the host pid namespace, so it can search
for existing container IPs. Also now we are firing off sub-processes
it's better to not be PID 1 so Linux doesn't expect us to reap zombies.
Enabled hairpin mode on vethwe-bridge was one of the root causes
for installing looping flows in the fastdp mode.
Also, it confused the weave custom L2 bridge, as the bridge treats
a bounced back packet, e.g. ARP brodcast request, as a local one
(the case for "Captured a packet associated with remote host").
Serve metrics+status on status socket and api+status+metrics on api
socket; the reasoning is that metrics or status are read-only hence
may be scraped from an off-machine connection, while we don't want to
open the control port up unprotected.
Otherwise, we risk to lose peer list when any failure happens after
the creation but before storing the list and as a consequence,
the weave container gets restarted.