MQE: add support for quantile_over_time

[charleskorn]

What this PR does

This PR adds support in MQE for quantile_over_time.

Which issue(s) this PR fixes or relates to

#10067

Checklist

• Tests updated.
• [n/a] Documentation added.
• [covered by Mimir Query Engine #10067] CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX].
• [n/a] about-versioning.md updated with experimental features.

[jhesketh]

perhaps types.EmitAnnotationFunc could take in a parameter for argument index?

[charleskorn]

The difficulty with this is that FunctionOverRangeVector doesn't know which order the arguments were in originally (inner operator first, or scalars first?), so I think what's here is the least-confusing option.

[jhesketh]

Any chance of a benchmark for this one please? I'd also like to see the effect on other range-vector queries against main.

[charleskorn]

Any chance of a benchmark for this one please? I'd also like to see the effect on other range-vector queries against main.

tl;dr: MQE is far faster than Prometheus' engine for quantile_over_time (up to 50% faster), and there seems to be little impact on other functions over range vectors.

quantile_over_time: MQE vs Prometheus

goos: darwin
goarch: arm64
pkg: github.com/grafana/mimir/pkg/streamingpromql/benchmarks
cpu: Apple M1 Pro
                                                                          │ Prometheus  │               Mimir                │
                                                                          │   sec/op    │   sec/op     vs base               │
Query/quantile_over_time(0.3,_a_1[1m]),_instant_query-10                    140.8µ ± 2%   135.0µ ± 1%   -4.14% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_1[1m]),_range_query_with_100_steps-10       182.6µ ± 2%   156.3µ ± 2%  -14.39% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_1[1m]),_range_query_with_1000_steps-10      540.3µ ± 4%   264.9µ ± 1%  -50.97% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_100[1m]),_instant_query-10                  514.9µ ± 2%   411.7µ ± 2%  -20.06% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_100[1m]),_range_query_with_100_steps-10     3.810m ± 1%   1.989m ± 1%  -47.79% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_100[1m]),_range_query_with_1000_steps-10    27.36m ± 1%   11.80m ± 1%  -56.86% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_2000[1m]),_instant_query-10                 5.707m ± 1%   4.433m ± 3%  -22.33% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_2000[1m]),_range_query_with_100_steps-10    63.39m ± 0%   33.72m ± 1%  -46.81% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_2000[1m]),_range_query_with_1000_steps-10   453.5m ± 1%   221.8m ± 0%  -51.09% (p=0.002 n=6)
geomean                                                                     3.675m        2.299m       -37.46%

                                                                          │   Prometheus   │                Mimir                │
                                                                          │      B/op      │     B/op      vs base               │
Query/quantile_over_time(0.3,_a_1[1m]),_instant_query-10                      24.64Ki ± 0%   20.96Ki ± 0%  -14.91% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_1[1m]),_range_query_with_100_steps-10         58.69Ki ± 0%   22.06Ki ± 0%  -62.42% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_1[1m]),_range_query_with_1000_steps-10       360.49Ki ± 0%   25.96Ki ± 0%  -92.80% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_100[1m]),_instant_query-10                    165.2Ki ± 0%   120.1Ki ± 0%  -27.26% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_100[1m]),_range_query_with_100_steps-10      3323.0Ki ± 0%   212.6Ki ± 0%  -93.60% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_100[1m]),_range_query_with_1000_steps-10    32041.6Ki ± 0%   595.0Ki ± 0%  -98.14% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_2000[1m]),_instant_query-10                   2.799Mi ± 0%   1.991Mi ± 0%  -28.88% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_2000[1m]),_range_query_with_100_steps-10     66.827Mi ± 0%   3.889Mi ± 1%  -94.18% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_2000[1m]),_range_query_with_1000_steps-10    636.69Mi ± 0%   11.12Mi ± 1%  -98.25% (p=0.002 n=6)
geomean                                                                       2.141Mi        294.9Ki       -86.55%

                                                                          │  Prometheus   │               Mimir                │
                                                                          │   allocs/op   │  allocs/op   vs base               │
Query/quantile_over_time(0.3,_a_1[1m]),_instant_query-10                       448.0 ± 0%    367.0 ± 0%  -18.08% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_1[1m]),_range_query_with_100_steps-10          663.0 ± 0%    377.0 ± 0%  -43.14% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_1[1m]),_range_query_with_1000_steps-10        2491.0 ± 0%    400.0 ± 0%  -83.94% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_100[1m]),_instant_query-10                    2.172k ± 0%   1.884k ± 0%  -13.26% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_100[1m]),_range_query_with_100_steps-10      23.010k ± 0%   2.695k ± 0%  -88.29% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_100[1m]),_range_query_with_1000_steps-10    205.518k ± 0%   5.016k ± 0%  -97.56% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_2000[1m]),_instant_query-10                   34.89k ± 0%   30.78k ± 0%  -11.78% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_2000[1m]),_range_query_with_100_steps-10     454.60k ± 0%   46.93k ± 0%  -89.68% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_2000[1m]),_range_query_with_1000_steps-10   4101.18k ± 0%   93.00k ± 0%  -97.73% (p=0.002 n=6)
geomean                                                                       19.92k        3.858k       -80.63%

                                                                          │  Prometheus  │               Mimir                │
                                                                          │      B       │      B        vs base              │
Query/quantile_over_time(0.3,_a_1[1m]),_instant_query-10                    66.53Mi ± 0%   66.02Mi ± 2%       ~ (p=0.087 n=6)
Query/quantile_over_time(0.3,_a_1[1m]),_range_query_with_100_steps-10       63.64Mi ± 1%   65.41Mi ± 1%  +2.77% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_1[1m]),_range_query_with_1000_steps-10      62.57Mi ± 1%   64.51Mi ± 2%  +3.10% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_100[1m]),_instant_query-10                  61.46Mi ± 1%   61.68Mi ± 2%       ~ (p=0.851 n=6)
Query/quantile_over_time(0.3,_a_100[1m]),_range_query_with_100_steps-10     63.55Mi ± 1%   61.12Mi ± 1%  -3.82% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_100[1m]),_range_query_with_1000_steps-10    67.44Mi ± 1%   63.07Mi ± 1%  -6.48% (p=0.002 n=6)
Query/quantile_over_time(0.3,_a_2000[1m]),_instant_query-10                 63.33Mi ± 2%   63.78Mi ± 1%       ~ (p=0.132 n=6)
Query/quantile_over_time(0.3,_a_2000[1m]),_range_query_with_100_steps-10    71.98Mi ± 1%   70.32Mi ± 1%  -2.30% (p=0.004 n=6)
Query/quantile_over_time(0.3,_a_2000[1m]),_range_query_with_1000_steps-10   129.1Mi ± 1%   122.6Mi ± 1%  -5.09% (p=0.002 n=6)
geomean                                                                     70.14Mi        69.21Mi       -1.33%

Impact on other functions over range vectors, using rate as a representative example

Given the small absolute differences and the movement both up and down, I'm putting these differences down to noise rather than any systematic change.

                                                                      │ rate-before.txt │          rate-after.txt           │
                                                                      │     sec/op      │   sec/op     vs base              │
Query/rate(a_1[1m]),_instant_query/engine=Mimir-10                          134.2µ ± 3%   138.9µ ± 4%       ~ (p=0.065 n=6)
Query/rate(a_1[1m]),_range_query_with_100_steps/engine=Mimir-10             151.5µ ± 1%   150.6µ ± 2%       ~ (p=0.310 n=6)
Query/rate(a_1[1m]),_range_query_with_1000_steps/engine=Mimir-10            213.2µ ± 2%   226.1µ ± 6%  +6.06% (p=0.002 n=6)
Query/rate(a_100[1m]),_instant_query/engine=Mimir-10                        403.8µ ± 1%   407.5µ ± 3%       ~ (p=0.240 n=6)
Query/rate(a_100[1m]),_range_query_with_100_steps/engine=Mimir-10           1.532m ± 1%   1.538m ± 2%       ~ (p=0.310 n=6)
Query/rate(a_100[1m]),_range_query_with_1000_steps/engine=Mimir-10          6.999m ± 0%   7.086m ± 2%  +1.25% (p=0.002 n=6)
Query/rate(a_2000[1m]),_instant_query/engine=Mimir-10                       4.339m ± 1%   4.307m ± 2%       ~ (p=0.937 n=6)
Query/rate(a_2000[1m]),_range_query_with_100_steps/engine=Mimir-10          23.97m ± 1%   24.00m ± 2%       ~ (p=0.310 n=6)
Query/rate(a_2000[1m]),_range_query_with_1000_steps/engine=Mimir-10         128.3m ± 1%   128.7m ± 1%       ~ (p=0.065 n=6)
Query/rate(a_1[1m]),_range_query_with_10000_steps/engine=Mimir-10           894.0µ ± 1%   901.3µ ± 1%       ~ (p=0.093 n=6)
Query/rate(a_100[1m]),_range_query_with_10000_steps/engine=Mimir-10         65.19m ± 0%   65.22m ± 1%       ~ (p=0.818 n=6)
Query/rate(a_2000[1m]),_range_query_with_10000_steps/engine=Mimir-10         1.300 ± 2%    1.298 ± 1%       ~ (p=0.394 n=6)
Query/rate(nh_1[1m]),_instant_query/engine=Mimir-10                         139.1µ ± 2%   143.8µ ± 1%  +3.42% (p=0.002 n=6)
Query/rate(nh_1[1m]),_range_query_with_100_steps/engine=Mimir-10            262.5µ ± 2%   263.5µ ± 2%       ~ (p=0.818 n=6)
Query/rate(nh_1[1m]),_range_query_with_1000_steps/engine=Mimir-10           688.4µ ± 1%   660.8µ ± 2%  -4.00% (p=0.002 n=6)
Query/rate(nh_100[1m]),_instant_query/engine=Mimir-10                       582.7µ ± 5%   583.6µ ± 1%       ~ (p=0.699 n=6)
Query/rate(nh_100[1m]),_range_query_with_100_steps/engine=Mimir-10          11.39m ± 0%   11.39m ± 0%       ~ (p=1.000 n=6)
Query/rate(nh_100[1m]),_range_query_with_1000_steps/engine=Mimir-10         44.34m ± 1%   44.14m ± 3%       ~ (p=0.093 n=6)
Query/rate(nh_2000[1m]),_instant_query/engine=Mimir-10                      6.494m ± 1%   6.495m ± 1%       ~ (p=0.818 n=6)
Query/rate(nh_2000[1m]),_range_query_with_100_steps/engine=Mimir-10         210.6m ± 1%   210.7m ± 1%       ~ (p=0.394 n=6)
Query/rate(nh_2000[1m]),_range_query_with_1000_steps/engine=Mimir-10        850.4m ± 1%   852.1m ± 1%       ~ (p=0.589 n=6)
Query/rate(nh_1[1m]),_range_query_with_10000_steps/engine=Mimir-10          4.790m ± 1%   4.843m ± 1%  +1.10% (p=0.026 n=6)
Query/rate(nh_100[1m]),_range_query_with_10000_steps/engine=Mimir-10        417.1m ± 3%   417.2m ± 1%       ~ (p=0.589 n=6)
Query/rate(nh_2000[1m]),_range_query_with_10000_steps/engine=Mimir-10        8.644 ± 1%    8.699 ± 2%       ~ (p=0.065 n=6)