send SIGHUP to docker daemon after change daemon.json , docker network doesn't reload correctly

[wangweihong]

environment:
Linux bogon 3.10.0-229.el7.x86_64 #1 SMP Fri Mar 6 11:36:42 UTC 2015 x86_64 x86_64 x86_64 GNU/Linux
docker Version:
1.11.2

1. I started a etcd server on my node1(192.168.14.130).and create /etc/docker/daemon.json in node2 like this:
   {
   "cluster-advertise": "192.168.14.87:2375",
   "cluster-store": "etcd://192.168.14.130:2379"
   }
   node2 is fresh, it didn't configure docker cluster configuration before.
2. Run " kill -SIGHUP pidof docker", docker info command get right infos:
   [root@bogon ~]# docker info
   ...
   Labels:
   com.example.environment="production"
   Cluster store: etcd://192.168.14.130:2379
   Cluster advertise: 192.168.14.87:2375
3. see what docker netowork ls shows
   [root@bogon ~]# docker network ls
   NETWORK ID NAME DRIVER
   fc1fa519d860 bridge bridge
   a34729eddc3e bridge10 overlay
   63795cfb1874 bridge11 overlay
   c8f6150341f0 bridge2 overlay
   c8f6150341f0 bridge2 overlay
   f7caa25698eb bridge5 overlay
   e21e3206baff bridge6 overlay
   4bfd9837011f bridge7 overlay
   d83655b1d8dc bridge9 overlay
   71f7ce2a1f59 host host
   864d033a3a1d isolated_Network overlay
   394cbced64d6 none null
   f44205d44b1b test1 overlay
   fe691a8573c8 test2 overlay
4. modify daemon.json ,change cluster-store
   {
   "cluster-advertise": "192.168.14.87:2375",
   "cluster-store": "etcd://192.168.14.130:2379**/test3**"
   }

5.run setp 2, 3 again
[root@bogon ~]# kill -SIGHUP pidof docker
[root@bogon ~]# docker info
...
Labels:
com.example.environment="production"
Cluster store: etcd://192.168.14.130:2379/test3
Cluster advertise: 192.168.14.87:2375

[root@bogon ~]# docker network ls
NETWORK ID NAME DRIVER
fc1fa519d860 bridge bridge
a34729eddc3e bridge10 overlay
63795cfb1874 bridge11 overlay
c8f6150341f0 bridge2 overlay

docker network ls show infos stored in etcd://192.168.14.130:2379 instead of etcd://192.168.14.130:2379/test3

1. docker network ls show correct infos after restart docker daemon
   [root@bogon ~]# systemctl restart docker
   [root@bogon ~]# docker network ls
   NETWORK ID NAME DRIVER
   fd8072220683 bridge bridge
   082f0f6947bd bridge3 overlay
   958c9f00d0e5 bridge4 overlay
   9f380b2f9829 bridge5 overlay
   bce851833ab5 docker_gwbridge bridge
   cc1c38e19959 host host
   b1f56b75853a network 1 overlay
   4d52f6e74d73 network 2 overlay
   4f74db38d8f2 network 3 overlay
   841cce023697 network 4 overlay
   182a7c53433d none null

Is this a bug?

[aboch]

See https://github.com/docker/libnetwork/blob/master/drivers/overlay/overlay.go#L292

Initial requirement was to only accept a datastore config if one was not already present.

[wangweihong]

@aboch so you mean, if I want to change a datastore config, I can't send SIGHUP to docker daemon to make it works, because it still hold the last datastore config.

[aboch]

yeah, libnetwork drivers will discard a new datastore config (received via SIGHUP daemon reload) if they already have a global scope datastore configured

[wangweihong]

@aboch Will it support reconfigure already configured datestore config without requiring to restart docker daemon in the future?
I think restart docker daemon to reconfigure is not a good experience.
Suppose situation like this: departmentA and departmentB both has 100 nodes. They have different datastore config . if I want to join 50 nodes in departmentA to departmentB . I have to change these nodes' datestore config, and restart their docker daemon one by one.What i am worry is directly restart docker daemon usually corrupt containers in it.

[aboch]

@wangweihong It is not on the plans at the moment.

Will go ahead and close this issue, as what can be done with SIGHUP was clarified.