Xiaochao Dong

Replay WAL concurrently without blocking

12

The WAL replay is extremely slow maybe because we have tons of series, and I have found the `time.Sleep` comes up multiple times. After some learning, I think we better...

Validate the metric name and label mames

8

According to the documentation: https://prometheus.io/docs/concepts/data_model/ > The metric name must match the regex [a-zA-Z_:][a-zA-Z0-9_:]*. Label names must match the regex [a-zA-Z_][a-zA-Z0-9_]*. Prometheus isn't validating the metric name and label names...

Fix repeated resolved alerts

13

Assumptions: 1. Prometheus/XXRuler failed to send some active alerts, but succeeded to send all resolved alerts: alert-a: T1(active:notify-ok), T2(resolved:notify-ok), **T3(active:notify-failed)**, T4(resolved:notify-ok). 2. Or maybe the alertmanager is behind a proxy...

Users may construct a promQL like `metric{key=~"b|z|..tens of thousands..|a"}` with Grafana variables or code, which is extremely slow. In our environment, the index header is ~99MiB per block, the CPU...

Reject incoming requests on shutdown

2

**Checklist** - [ ] Tests updated - [ ] Documentation added - [x] `CHANGELOG.md` updated - the order of entries should be `[CHANGE]`, `[FEATURE]`, `[ENHANCEMENT]`, `[BUGFIX]`

Send Retry-After header with 429 response or 413 if samples exceed the burst of rate limiter

4

**What this PR does**: **Which issue(s) this PR fixes**: Fixes # **Checklist** - [x] Tests updated - [ ] Documentation added - [x] `CHANGELOG.md` updated - the order of entries...

**What this PR does**: **Which issue(s) this PR fixes**: Some providers may only support one of those bucket lookup styles. **Checklist** - [x] Tests updated - [ ] Documentation added...

Fix split range queries with wrong boundaries

2

**What this PR does**: The original splitQuery add a gap(step) between boundaries, in my point of view, we may lose some data points because of this. We should strictly align...

Since some implementations may not has a default error message.