HELLO response with node missing IP address

antirez / disque

Disque is a distributed message broker

BSD 3-Clause "New" or "Revised" License

8.01k stars 538 forks source link

HELLO response with node missing IP address #201

Open dominics opened 7 years ago

dominics commented 7 years ago

In our testing environment, we saw a Disque node start to respond to HELLO with a response missing an IP address for its own node (it had neither 127.0.0.1 nor the local network IP - two interfaces available on this machine):

127.0.0.1:7711> hello
1) (integer) 1
2) "e4b6ad92530fa59da330ff921d9ef00ed1eea985"
3) 1) "e4b6ad92530fa59da330ff921d9ef00ed1eea985"
   2) ""
   3) "7711"
   4) "1"

This confused the client I was using to no end. The client shouldn't try to use the HELLO information for a cluster of one node, but it was trying to anyway, and not expecting the missing IP.

Uptime on this disque node was 7 days, with very low load over that time.

dominics commented 7 years ago

A few more details:

CLUSTER RESET didn't fix it, nor did DEBUG FLUSHALL, nor restarting the disque server, because...
nodes.conf is missing the IP address too: e4b6ad92530fa59da330ff921d9ef00ed1eea985 :7711 myself 0 0 connected
- Stopping the server, removing nodes.conf, restarting, will recreate this sort of entry missing the IP address (different ID obviously)
bind config option is unset on this machine, setting it didn't help

dominics commented 7 years ago

OK, finally found the IP address for this node updated to 127.0.0.1 message upon connecting any other node.

So, it seems like this is expected behaviour: HELLO responses won't have an IP address for a node that hasn't met a cluster yet. I think that should probably be documented explicitly somewhere? I can then take that to the client maintainer and ask them to filter out HELLO responses with a missing IP address from consideration.