Fabian Groffen

So, you want each metric to end up at any set of two stores, but you don't care which one. Normally, people would use a consistent hash for this, but...

Yeah, I think I once meant to probe for a free port to use, but this is likely a TODO. So it uses a port that's probably in use by...

working theory at the moment is that for some reason the port increment isn't working, so the second server start uses the same port and thus fails update: scratch that,...

I'm wondering, is there anything specific towards the network stack in this env? e.g. disabled ipv6 or something?

Does that mean that any bind() is supposed to fail in the buildenv? The tests try to ensure data is being transferred across unix and TCP sockets.

Hmmmm, I'm a bit at loss here. Since I just use bind(), it seems to suggest this is glibc related (as only package that I can imagine influencing this, except...

``` [pid 1559539] socket(AF_INET, SOCK_STREAM, IPPROTO_TCP) = 1 [pid 1559539] setsockopt(1, SOL_SOCKET, SO_RCVTIMEO_OLD, "\0\0\0\0\0\0\0\0 \241\7\0\0\0\0\0", 16) = 0 [pid 1559539] setsockopt(1, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0 [pid 1559539] bind(1,...

I think what could happen is that tests execute too fast after each other or something and for some reason the connection is still lingering around. Oddly enough this shouldn't...

yes, but the amount of time it sleeps ... it should produce the aggregate on a window end You could try and increase the sleep, I wonder if that helps

That depends on the cluster type you're using. `any_of` or `failover` typically try not to queue metrics up, but any of the `xxxx_ch` clusters will queue up until overflow.