weaver is using 40% CPU

[tomwilkie]

Looks like something in the gossip code, probably dns.

[tomwilkie]

Repro with the test on dns-gossip branch (210_dns_cross_host)

[tomwilkie]

vagrant@host1:~$ go tool pprof http://$(docker inspect --format='{{.NetworkSettings.IPAddress}}' weave):6784/debug/pprof/profile
Fetching profile from http://172.17.0.92:6784/debug/pprof/profile
Please wait... (30s)
Saved profile in /home/vagrant/pprof/pprof.172.17.0.92:6784.samples.cpu.001.pb.gz
Entering interactive mode (type "help" for commands)
(pprof) top10 -cum
0.08s of 12.91s total ( 0.62%)
Dropped 200 nodes (cum <= 0.06s)
Showing top 10 nodes out of 114 (cum >= 3.89s)
      flat  flat%   sum%        cum   cum%
         0     0%     0%      8.15s 63.13%  runtime.goexit
         0     0%     0%      5.46s 42.29%  github.com/weaveworks/weave/router.(*LocalConnection).receiveTCP
         0     0%     0%      4.97s 38.50%  encoding/gob.(*Decoder).Decode
     0.04s  0.31%  0.31%      4.97s 38.50%  encoding/gob.(*Decoder).DecodeValue
         0     0%  0.31%      4.81s 37.26%  github.com/weaveworks/weave/router.(*LocalConnection).handleProtocolMsg
         0     0%  0.31%      4.81s 37.26%  github.com/weaveworks/weave/router.(*Router).handleGossip
         0     0%  0.31%      4.31s 33.38%  github.com/weaveworks/weave/router.(*GossipChannel).deliver
     0.03s  0.23%  0.54%      4.24s 32.84%  encoding/gob.(*Decoder).decodeValue
         0     0%  0.54%      3.89s 30.13%  github.com/weaveworks/weave/nameserver.(*Nameserver).OnGossip
     0.01s 0.077%  0.62%      3.89s 30.13%  github.com/weaveworks/weave/nameserver.(*Nameserver).receiveGossip

[tomwilkie]

Looks like its definitely the new gossipdns, and come kind of non-converging gossip.