Björn Rabenstein

@stuartnelson3 and I have been working on replacing the deprecated `prometheus.InstrumentHandler` for a while now. Server middleware is already in head (package `promhttp`), while roundtripper-ware is being reviewed in https://github.com/prometheus/client_golang/pull/295...

Hi @mwitkow , Finally, I had a closer look. The selling point of this package is certainly the integration with the http_ctxtags, which are used in all the related package...

OK, makes sense. About my naming remark: You actually _do_ let the user pass in the "namespace", so my example about layering will work fine by providing different namespaces.

I can confirm that order of rules within a group matters because later rules may depend on the output of earlier rules. Having multiple rules funneling their results into different...

To avoid misunderstandings, I should mention that custom bucket layouts are a niche feature that is expected to be used very rarely for very specific use cases. Guaranteed mergeability across...

Weird, so your watermark is still progressing most of the time, and the delta between high and low watermark, while on the larger side, isn't ginormous. Still you see many...

Both. Or anything that is unusually expensive, long-running, or large. Frequent large scrapes could lead to many entries in txRing, especially if combined with frequent long-running queries. However, if that...

@marquis-wang what @bboreham said. It would be extremely surprising if the block duration settings change anything in terms of the tsdb.txRing size. (But if it actually did, it would be...

@ilixiaocui as @bboreham pointed out, your ingest load seems to be around 1 million samples per second. That's a bit on the high side but should be no problem in...

Right. I guess the thought was if it goes up that far once, we'll need it again in that size, so there is no point in shrinking and then regrowing...