Bryan Boreham

I think I was referring to the response when I said "Probably this is an extension to the Prometheus remote-write proto format.".

I am concerned that reporting all errors could be very computationally expensive, e.g. if a call contains 500 series and 500 of them error. After reflection I think something like...

>not many instant queries have a time range > 24h However the ones we most want to speed up do.

An option make it a hard error if HA labels are defined but not found in input? Also I often want to know what was the primary selected at points...

I was trying to remember why it wasn’t done this way in the first place, and I realised this is much more like query-sharding than slicing by day. Federated queries...

Note Mimir currently uses https://github.com/themihai/gomemcache fork, in order to add a circuit-breaker (only in query-frontend right now).

> To avoid copying every single line just to scan it. I wanted to check the size of this problem, so I grabbed another profile. ``` Total: 6.89TB ROUTINE ========================...

I coded `scanGetResponseLine()` using `IndexByte()` instead of `SScanf()`. https://github.com/bradfitz/gomemcache/compare/master...bboreham:faster-scanline ``` name old time/op new time/op delta ScanGetResponseLine-4 4.06µs ± 1% 0.13µs ± 4% -96.89% (p=0.008 n=5+5) name old alloc/op new...

I have created https://github.com/grafana/gomemcache/ with the above change.

I renamed this issue because the basic idea generalises to splitting by any label - for instance people might have a `team` or a `domain` label that can tie back...