influxdb-relay
influxdb-relay copied to clipboard
the relay leak client sockets in TIME_WAIT state when QPS is high
Seems the sockets connecting to backends are not closed in time. this causes socket port exhausted on the machine. I guess re-using the the http transport to backends should resolve the issue, it is not necessary to create new one for each write request.
BTW, I found it on windows.
Is there a open bug on this issue? Also has anyone provided a concrete work around to prevent this from happening?
Fixed on https://github.com/vente-privee/influxdb-relay.