Vladimir Smirnov

"Connection refused" means that there was no one listening on that port, and not that there were "too many connections". If we'll be talking about what happens on TCP layer,...

At least on Linux this is not a default behavior. Unless you set `tcp_abort_on_overflow` sysctl, which by default is set to 0: ``` tcp_abort_on_overflow - BOOLEAN If listening service is...

We've triggered this problem again but now with maxGlobs=100 but 20M metrics per server. Symptoms and trace pretty much the same. Maybe we should add some special cases, store metrics...

One of new ideas is to try to optimize go-trigrams with either fastbit (http://crd-legacy.lbl.gov/~kewu/fastbit/compression.html) or with roaring bitmaps (e.x. https://github.com/RoaringBitmap/roaring).

Problem with kafka listener is that it was never used in production (or at least I'm not aware of that) so it might contain bugs, but I (AFAIR I was...

I think it's not really related to cache, but just related to unmarshal. He's trying to fetch 90k metrics for some significant amount of time and that makes Unmarshal to...

@deniszh if go-carbon would have significantly more than max-creates-per-second new metrics, they will be delayed. Also by the value (1000) it looks like underlying storage is rather slow. And carbon...

@deniszh oh, I missed that PR. Thanks.

I would add that in default setup of carbon-cache it will just silently ignore all "NaN" values. If I'm correct that happens exactly here: https://github.com/graphite-project/carbon/blob/master/lib/carbon/protocols.py#L174-L175 So the only difference in...

Can you provide more details about that?