Store Prometheus labels as a slice instead of a map

[swiatekm]

Description:
Store labels for targets in the target allocator the same way Prometheus itself does - as a key+value list of type labels.Labels instead of a model.Labelset map. The former is what relabelling acts on, so we get to skip the conversion there, and building the list is faster as well. This makes looking up values of specific labels slower, but we don't really do that outside of determining the Node name.

I've modified the benchmark to actually drop half the targets during relabelling to make it more realistic.

Benchstat shows some nice gains:

goos: linux
goarch: amd64
pkg: github.com/open-telemetry/opentelemetry-operator/cmd/otel-allocator
cpu: AMD Ryzen 9 7950X3D 16-Core Processor          
                                                      │ bench_targets_main.txt │      bench_targets_branch.txt       │
                                                      │         sec/op         │   sec/op     vs base                │
ProcessTargets/least-weighted-32                                   78.24m ± 1%   42.75m ± 1%  -45.36% (p=0.000 n=10)
ProcessTargets/consistent-hashing-32                               78.86m ± 0%   42.72m ± 1%  -45.83% (p=0.000 n=10)
ProcessTargets/per-node-32                                         78.71m ± 0%   42.86m ± 1%  -45.55% (p=0.000 n=10)
ProcessTargetsWithRelabelConfig/per-node-32                        78.69m ± 1%   43.15m ± 1%  -45.17% (p=0.000 n=10)
ProcessTargetsWithRelabelConfig/least-weighted-32                  79.11m ± 1%   43.25m ± 1%  -45.33% (p=0.000 n=10)
ProcessTargetsWithRelabelConfig/consistent-hashing-32              79.05m ± 1%   42.86m ± 2%  -45.79% (p=0.000 n=10)
geomean                                                            78.78m        42.93m       -45.50%

                                                      │ bench_targets_main.txt │       bench_targets_branch.txt       │
                                                      │          B/op          │     B/op      vs base                │
ProcessTargets/least-weighted-32                                  45.14Mi ± 0%   40.41Mi ± 0%  -10.48% (p=0.000 n=10)
ProcessTargets/consistent-hashing-32                              45.15Mi ± 0%   40.41Mi ± 0%  -10.49% (p=0.000 n=10)
ProcessTargets/per-node-32                                        45.15Mi ± 0%   40.41Mi ± 0%  -10.49% (p=0.000 n=10)
ProcessTargetsWithRelabelConfig/per-node-32                       45.38Mi ± 0%   40.44Mi ± 0%  -10.88% (p=0.000 n=10)
ProcessTargetsWithRelabelConfig/least-weighted-32                 45.38Mi ± 0%   40.44Mi ± 0%  -10.88% (p=0.000 n=10)
ProcessTargetsWithRelabelConfig/consistent-hashing-32             45.38Mi ± 0%   40.44Mi ± 0%  -10.88% (p=0.000 n=10)
geomean                                                           45.26Mi        40.43Mi       -10.68%

                                                      │ bench_targets_main.txt │      bench_targets_branch.txt       │
                                                      │       allocs/op        │  allocs/op   vs base                │
ProcessTargets/least-weighted-32                                  112.36k ± 0%   52.05k ± 0%  -53.67% (p=0.000 n=10)
ProcessTargets/consistent-hashing-32                              112.38k ± 0%   52.05k ± 0%  -53.68% (p=0.000 n=10)
ProcessTargets/per-node-32                                        112.38k ± 0%   52.06k ± 0%  -53.68% (p=0.000 n=10)
ProcessTargetsWithRelabelConfig/per-node-32                       112.73k ± 0%   52.26k ± 0%  -53.64% (p=0.000 n=10)
ProcessTargetsWithRelabelConfig/least-weighted-32                 112.73k ± 0%   52.26k ± 0%  -53.64% (p=0.000 n=10)
ProcessTargetsWithRelabelConfig/consistent-hashing-32             112.74k ± 0%   52.26k ± 0%  -53.64% (p=0.000 n=10)
geomean                                                            112.6k        52.16k       -53.66%

[nicolastakashi]

Amazing work on that @swiatekm 👏🏽

[nicolastakashi]

LGTM

[nicolastakashi]

Hey @swiatekm anything I can do to help with this change?

[swiatekm]

@nicolastakashi I'm just waiting for another approver review.

[yuriolisa]

LGTM