Bryan Boreham
Bryan Boreham
> does this clarify? Yes thank you; I think you switched names from `http_path` and `http_verb` to `path` and `verb` midway through your explanation but I get the general idea....
Tip: run your benchmarks several times to account for random slow-downs on your computer. For example `-count=6`. Then use the `benchstat` tool to analyze the timings and give you the...
> having it as a slice is +16 bytes for a slice. But now, memChunkList also adds +16 bytes for the 2 pointers. A slice header is 24 bytes so...
> pay more for linked list In what sense?
> Slice overhead is 24 bytes per each slice. plus 8 bytes per chunk, since the type was `[]*memChunk`. (plus some rounding up due to append behaviour and allocation span...
To zoom out a bit, the broader context is that `memSeries` is one of the larger consumers of memory in Prometheus, and we recently did a lot of work to...
Do you have a benchmark that illustrates the problem?
My constructive feedback is to show a benchmark.
> abnormal queries that remain active indefinitely Can you explain how this happens?
I think this proposed change exposes an implementation detail in TSDB that the `Appender` interface was designed to hide. How about a function similar to [`GetRef`](https://github.com/prometheus/prometheus/blob/30b31cae93b2fe66486f188c07e215b80ebc93f8/storage/interface.go#L245-L253) which adds the reference...