MQE: subqueries

[charleskorn]

What this PR does

This PR adds support for subqueries to MQE.

Our benchmarks show mild latency improvements in most cases over Prometheus' engine, and improvements in peak memory utilisation in most cases:

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/sum_over_time(a_1[10m:3m]),_instant_query-10                         156.0µ ± 4%   152.3µ ± 4%   -2.36% (p=0.015 n=6)
Query/sum_over_time(a_1[10m:3m]),_range_query_with_100_steps-10            165.6µ ± 2%   154.6µ ± 1%   -6.63% (p=0.002 n=6)
Query/sum_over_time(a_1[10m:3m]),_range_query_with_1000_steps-10           250.2µ ± 1%   245.1µ ± 2%   -2.03% (p=0.041 n=6)
Query/sum_over_time(a_100[10m:3m]),_instant_query-10                       1.116m ± 2%   1.075m ± 8%        ~ (p=0.065 n=6)
Query/sum_over_time(a_100[10m:3m]),_range_query_with_100_steps-10          1.882m ± 4%   1.691m ± 1%  -10.18% (p=0.002 n=6)
Query/sum_over_time(a_100[10m:3m]),_range_query_with_1000_steps-10         8.662m ± 0%   9.212m ± 1%   +6.36% (p=0.002 n=6)
Query/sum_over_time(a_2000[10m:3m]),_instant_query-10                      16.02m ± 1%   15.38m ± 1%   -3.99% (p=0.002 n=6)
Query/sum_over_time(a_2000[10m:3m]),_range_query_with_100_steps-10         30.11m ± 1%   26.48m ± 1%  -12.06% (p=0.002 n=6)
Query/sum_over_time(a_2000[10m:3m]),_range_query_with_1000_steps-10        160.5m ± 7%   171.7m ± 3%   +6.95% (p=0.015 n=6)
Query/sum_over_time(nh_1[10m:3m]),_instant_query-10                        202.3µ ± 1%   193.9µ ± 1%   -4.18% (p=0.002 n=6)
Query/sum_over_time(nh_1[10m:3m]),_range_query_with_100_steps-10           260.3µ ± 5%   246.1µ ± 2%   -5.45% (p=0.002 n=6)
Query/sum_over_time(nh_1[10m:3m]),_range_query_with_1000_steps-10          829.5µ ± 0%   800.1µ ± 2%   -3.54% (p=0.002 n=6)
Query/sum_over_time(nh_100[10m:3m]),_instant_query-10                      5.656m ± 3%   5.558m ± 3%   -1.73% (p=0.041 n=6)
Query/sum_over_time(nh_100[10m:3m]),_range_query_with_100_steps-10         10.78m ± 1%   10.32m ± 0%   -4.29% (p=0.002 n=6)
Query/sum_over_time(nh_100[10m:3m]),_range_query_with_1000_steps-10        62.86m ± 1%   58.10m ± 4%   -7.57% (p=0.002 n=6)
Query/sum_over_time(nh_2000[10m:3m]),_instant_query-10                     105.3m ± 0%   103.4m ± 0%   -1.77% (p=0.002 n=6)
Query/sum_over_time(nh_2000[10m:3m]),_range_query_with_100_steps-10        204.7m ± 1%   194.1m ± 5%   -5.21% (p=0.015 n=6)
Query/sum_over_time(nh_2000[10m:3m]),_range_query_with_1000_steps-10        1.265 ± 3%    1.154 ± 1%   -8.81% (p=0.002 n=6)
Query/sum(sum_over_time(a_1[10m:3m])),_instant_query-10                    158.2µ ± 2%   150.4µ ± 2%   -4.93% (p=0.002 n=6)
Query/sum(sum_over_time(a_1[10m:3m])),_range_query_with_100_steps-10       169.8µ ± 2%   157.0µ ± 2%   -7.53% (p=0.002 n=6)
Query/sum(sum_over_time(a_1[10m:3m])),_range_query_with_1000_steps-10      279.2µ ± 1%   250.4µ ± 2%  -10.33% (p=0.002 n=6)
Query/sum(sum_over_time(a_100[10m:3m])),_instant_query-10                  1.135m ± 2%   1.071m ± 3%   -5.63% (p=0.002 n=6)
Query/sum(sum_over_time(a_100[10m:3m])),_range_query_with_100_steps-10     1.929m ± 1%   1.688m ± 3%  -12.50% (p=0.002 n=6)
Query/sum(sum_over_time(a_100[10m:3m])),_range_query_with_1000_steps-10    9.272m ± 1%   9.539m ± 1%   +2.88% (p=0.002 n=6)
Query/sum(sum_over_time(a_2000[10m:3m])),_instant_query-10                 16.19m ± 4%   15.58m ± 3%   -3.75% (p=0.002 n=6)
Query/sum(sum_over_time(a_2000[10m:3m])),_range_query_with_100_steps-10    31.18m ± 1%   27.01m ± 1%  -13.36% (p=0.002 n=6)
Query/sum(sum_over_time(a_2000[10m:3m])),_range_query_with_1000_steps-10   183.2m ± 1%   178.0m ± 1%   -2.82% (p=0.002 n=6)
geomean                                                                    4.550m        4.329m        -4.87%

                                                                         │  Prometheus   │                Mimir                │
                                                                         │       B       │      B        vs base               │
Query/sum_over_time(a_1[10m:3m]),_instant_query-10                          73.68Mi ± 1%   73.67Mi ± 1%        ~ (p=0.784 n=6)
Query/sum_over_time(a_1[10m:3m]),_range_query_with_100_steps-10             73.23Mi ± 1%   73.53Mi ± 1%        ~ (p=0.132 n=6)
Query/sum_over_time(a_1[10m:3m]),_range_query_with_1000_steps-10            70.84Mi ± 1%   71.88Mi ± 1%   +1.48% (p=0.009 n=6)
Query/sum_over_time(a_100[10m:3m]),_instant_query-10                        67.51Mi ± 1%   66.95Mi ± 1%        ~ (p=0.132 n=6)
Query/sum_over_time(a_100[10m:3m]),_range_query_with_100_steps-10           67.80Mi ± 1%   67.02Mi ± 0%   -1.15% (p=0.002 n=6)
Query/sum_over_time(a_100[10m:3m]),_range_query_with_1000_steps-10          69.89Mi ± 1%   69.70Mi ± 1%        ~ (p=1.000 n=6)
Query/sum_over_time(a_2000[10m:3m]),_instant_query-10                       68.45Mi ± 2%   69.11Mi ± 1%        ~ (p=0.132 n=6)
Query/sum_over_time(a_2000[10m:3m]),_range_query_with_100_steps-10          75.21Mi ± 1%   77.66Mi ± 1%   +3.26% (p=0.002 n=6)
Query/sum_over_time(a_2000[10m:3m]),_range_query_with_1000_steps-10         133.3Mi ± 1%   127.6Mi ± 0%   -4.26% (p=0.002 n=6)
Query/sum_over_time(nh_1[10m:3m]),_instant_query-10                         78.29Mi ± 1%   78.73Mi ± 1%        ~ (p=0.394 n=6)
Query/sum_over_time(nh_1[10m:3m]),_range_query_with_100_steps-10            73.31Mi ± 1%   72.88Mi ± 0%        ~ (p=0.065 n=6)
Query/sum_over_time(nh_1[10m:3m]),_range_query_with_1000_steps-10           72.63Mi ± 1%   71.93Mi ± 1%   -0.97% (p=0.015 n=6)
Query/sum_over_time(nh_100[10m:3m]),_instant_query-10                       70.20Mi ± 1%   70.10Mi ± 1%        ~ (p=1.000 n=6)
Query/sum_over_time(nh_100[10m:3m]),_range_query_with_100_steps-10          73.68Mi ± 1%   73.95Mi ± 1%        ~ (p=0.180 n=6)
Query/sum_over_time(nh_100[10m:3m]),_range_query_with_1000_steps-10         120.0Mi ± 1%   121.8Mi ± 1%   +1.44% (p=0.004 n=6)
Query/sum_over_time(nh_2000[10m:3m]),_instant_query-10                      76.34Mi ± 1%   73.81Mi ± 1%   -3.31% (p=0.002 n=6)
Query/sum_over_time(nh_2000[10m:3m]),_range_query_with_100_steps-10         180.4Mi ± 1%   184.1Mi ± 1%   +2.09% (p=0.002 n=6)
Query/sum_over_time(nh_2000[10m:3m]),_range_query_with_1000_steps-10        630.1Mi ± 0%   726.8Mi ± 7%  +15.35% (p=0.002 n=6)
Query/sum(sum_over_time(a_1[10m:3m])),_instant_query-10                     73.16Mi ± 1%   73.48Mi ± 1%        ~ (p=0.485 n=6)
Query/sum(sum_over_time(a_1[10m:3m])),_range_query_with_100_steps-10        72.80Mi ± 2%   73.14Mi ± 1%        ~ (p=0.394 n=6)
Query/sum(sum_over_time(a_1[10m:3m])),_range_query_with_1000_steps-10       70.45Mi ± 0%   71.66Mi ± 1%   +1.72% (p=0.004 n=6)
Query/sum(sum_over_time(a_100[10m:3m])),_instant_query-10                   67.38Mi ± 2%   66.86Mi ± 1%        ~ (p=0.260 n=6)
Query/sum(sum_over_time(a_100[10m:3m])),_range_query_with_100_steps-10      67.45Mi ± 1%   66.66Mi ± 1%   -1.17% (p=0.026 n=6)
Query/sum(sum_over_time(a_100[10m:3m])),_range_query_with_1000_steps-10     69.45Mi ± 2%   66.91Mi ± 1%   -3.66% (p=0.002 n=6)
Query/sum(sum_over_time(a_2000[10m:3m])),_instant_query-10                  69.12Mi ± 1%   68.40Mi ± 2%        ~ (p=0.065 n=6)
Query/sum(sum_over_time(a_2000[10m:3m])),_range_query_with_100_steps-10     76.16Mi ± 1%   68.52Mi ± 2%  -10.03% (p=0.002 n=6)
Query/sum(sum_over_time(a_2000[10m:3m])),_range_query_with_1000_steps-10   132.88Mi ± 2%   71.20Mi ± 1%  -46.41% (p=0.002 n=6)
geomean                                                                     85.72Mi        83.71Mi        -2.34%

It is possible to improve the performance of the 1000 step cases, however to do this without affecting the performance of range vector selectors requires a bit of refactoring I'd prefer to do in a separate PR.

Which issue(s) this PR fixes or relates to

(none)

Checklist

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

[tacole02]

Looks good!