Closed azhiltsov closed 4 years ago
Can you please give a little bit more explanation on this?
So, the problem manifests as follows:
Right?
Here's a test trying to replicate this. Everything works if Nanotube is started from scratch with the default config. See the screenshot below.
A single point is sent and it arrives at both clusters.
Probably the problem happens only when Nanotube is in some specific state.
if you run go-carbon on the destination 10.10.10.20 then it will break connection after ~60 sec itself as it currently ignores keep-alive we are sending. After that you can retry sending metric from nanotube and should get a broken pipe, as nanotube does not notice that connection loss.
Thanks, I will try that.
submitting traffic to low traffic destination requires to open a connection, however it seems that nanotube is assuming that connection is already open and losing datapoints by writing to a broken pipe.