I set up 5 syncthing nodes each listening to different GUI addresses. I set globalAnnounceEnabled and relaysEnabled to False. Then I let the nodes run so that they may discover each other. Almost immediately, 1 node (let’s call this node 1) is fully connected to the rest of 4 nodes, with the remainder partially connected to each other. This keeps going indefinitely without extra pairing (I waited for half an hour) until I stop node 1, and also restart the remaining 4. Then another node becomes fully connected (say, node 3). I have to stop node 3 and restart the remainder again. Perhaps at any given time, a single node is absorbing all the multicast packets?
I expected the local discovery to happen immediately for the 5 nodes. But instead I had to manually restart the nodes.
To begin with you would to well to upgrade to a more recent release. After that, check on the non-connecting nodes in the device list on the right. If you see addresses with error messages, discovery works but connection doesn’t. If you don’t see any address either for the non-connected devices discovery isn’t working. Typically the firewall is the culprit, but if it’s working sometimes but not other times then I don’t know. There’s no “absorbing” the packets so what other devices are doing shouldn’t affect it.
I am building an environment to try the latest master of syncthing.
In the meantime let me provide you with some more precise information than my initial post. This is my testing process to produce the bug:
Install 5 syncthing nodes with custom listen address
Connect all the nodes
Share some folders
Inspect the log files
One node will always have 4 entries with “Secure connection established to XXXX” showing it is connected to all other nodes. All other nodes show a single entry of “Secure connection established {ID of node 1}”
Does this indicate failure of discovery or failure of connectivity or both?
I can also produce a simple bash script to reproduce the problem if this would help?
None of that indicates a failure at all, but the screenshots of UI with the parts noted above would clarify it. I have environments with tens and hundreds of devices using local discovery perfectly fine so I don’t think there is a conceptual problem once you reach five devices per subnet.