Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Filter out read-only instances, but skip filtering when it's not required. #565

Merged
merged 18 commits into from
Aug 23, 2024

Conversation

pstibrany
Copy link
Member

@pstibrany pstibrany commented Aug 15, 2024

What this PR does:

This PR updates shuffle-sharding to 1) filter out read-only instances on shuffle shard with lookback, if instance changed its state before lookback window, 2) avoid computing shuffle shard if no filtering is required.

Benchmark compared to main (updated after 95e9564):

benchstat main-new-bench.txt final2.txt
goos: darwin
goarch: arm64
pkg: github.com/grafana/dskit/ring
                                                                                  │ main-new-bench.txt │              final2.txt              │
                                                                                  │       sec/op       │    sec/op     vs base                │
Ring_ShuffleShard/num_instances_=_50,_num_zones_=_1,_shard_size_=_0-10                657443.00n ± ∞ ¹   59.71n ± ∞ ¹   -99.99% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_50,_num_zones_=_1,_shard_size_=_3-10                    19.83µ ± ∞ ¹   19.34µ ± ∞ ¹         ~ (p=0.151 n=5)
Ring_ShuffleShard/num_instances_=_50,_num_zones_=_1,_shard_size_=_10-10                   53.21µ ± ∞ ¹   53.59µ ± ∞ ¹         ~ (p=0.841 n=5)
Ring_ShuffleShard/num_instances_=_50,_num_zones_=_1,_shard_size_=_30-10                   163.3µ ± ∞ ¹   163.5µ ± ∞ ¹         ~ (p=0.310 n=5)
Ring_ShuffleShard/num_instances_=_50,_num_zones_=_1,_shard_size_=_100-10                  651.8µ ± ∞ ¹   272.4µ ± ∞ ¹   -58.21% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_50,_num_zones_=_1,_shard_size_=_1000-10                 652.2µ ± ∞ ¹   271.9µ ± ∞ ¹   -58.31% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_50,_num_zones_=_3,_shard_size_=_0-10                680198.00n ± ∞ ¹   58.80n ± ∞ ¹   -99.99% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_50,_num_zones_=_3,_shard_size_=_3-10                    45.55µ ± ∞ ¹   45.19µ ± ∞ ¹    -0.78% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_50,_num_zones_=_3,_shard_size_=_10-10                   87.33µ ± ∞ ¹   89.49µ ± ∞ ¹    +2.47% (p=0.016 n=5)
Ring_ShuffleShard/num_instances_=_50,_num_zones_=_3,_shard_size_=_30-10                   200.8µ ± ∞ ¹   203.2µ ± ∞ ¹         ~ (p=0.222 n=5)
Ring_ShuffleShard/num_instances_=_50,_num_zones_=_3,_shard_size_=_100-10                  677.0µ ± ∞ ¹   293.9µ ± ∞ ¹   -56.59% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_50,_num_zones_=_3,_shard_size_=_1000-10                 671.4µ ± ∞ ¹   291.4µ ± ∞ ¹   -56.60% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_100,_num_zones_=_1,_shard_size_=_0-10              1395995.00n ± ∞ ¹   59.98n ± ∞ ¹  -100.00% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_100,_num_zones_=_1,_shard_size_=_3-10                   19.45µ ± ∞ ¹   19.35µ ± ∞ ¹         ~ (p=0.310 n=5)
Ring_ShuffleShard/num_instances_=_100,_num_zones_=_1,_shard_size_=_10-10                  53.32µ ± ∞ ¹   55.47µ ± ∞ ¹    +4.02% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_100,_num_zones_=_1,_shard_size_=_30-10                  162.9µ ± ∞ ¹   168.1µ ± ∞ ¹    +3.19% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_100,_num_zones_=_1,_shard_size_=_100-10                 654.3µ ± ∞ ¹   635.1µ ± ∞ ¹    -2.93% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_100,_num_zones_=_1,_shard_size_=_1000-10               1366.3µ ± ∞ ¹   628.9µ ± ∞ ¹   -53.97% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_100,_num_zones_=_3,_shard_size_=_0-10              1377139.00n ± ∞ ¹   60.41n ± ∞ ¹  -100.00% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_100,_num_zones_=_3,_shard_size_=_3-10                   45.29µ ± ∞ ¹   46.29µ ± ∞ ¹    +2.22% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_100,_num_zones_=_3,_shard_size_=_10-10                  86.79µ ± ∞ ¹   91.29µ ± ∞ ¹    +5.18% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_100,_num_zones_=_3,_shard_size_=_30-10                  200.1µ ± ∞ ¹   208.0µ ± ∞ ¹    +3.94% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_100,_num_zones_=_3,_shard_size_=_100-10                1147.1µ ± ∞ ¹   672.6µ ± ∞ ¹   -41.36% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_100,_num_zones_=_3,_shard_size_=_1000-10               1383.6µ ± ∞ ¹   671.2µ ± ∞ ¹   -51.49% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_1000,_num_zones_=_1,_shard_size_=_0-10            17568050.00n ± ∞ ¹   61.36n ± ∞ ¹  -100.00% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_1000,_num_zones_=_1,_shard_size_=_3-10                  17.58µ ± ∞ ¹   18.28µ ± ∞ ¹    +3.96% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_1000,_num_zones_=_1,_shard_size_=_10-10                 50.52µ ± ∞ ¹   52.53µ ± ∞ ¹    +3.99% (p=0.016 n=5)
Ring_ShuffleShard/num_instances_=_1000,_num_zones_=_1,_shard_size_=_30-10                 157.3µ ± ∞ ¹   163.3µ ± ∞ ¹    +3.82% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_1000,_num_zones_=_1,_shard_size_=_100-10                626.2µ ± ∞ ¹   651.5µ ± ∞ ¹    +4.04% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_1000,_num_zones_=_1,_shard_size_=_1000-10               8.996m ± ∞ ¹   8.596m ± ∞ ¹    -4.44% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_1000,_num_zones_=_3,_shard_size_=_0-10            17422654.00n ± ∞ ¹   61.11n ± ∞ ¹  -100.00% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_1000,_num_zones_=_3,_shard_size_=_3-10                  41.42µ ± ∞ ¹   43.73µ ± ∞ ¹    +5.59% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_1000,_num_zones_=_3,_shard_size_=_10-10                 80.49µ ± ∞ ¹   86.60µ ± ∞ ¹    +7.60% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_1000,_num_zones_=_3,_shard_size_=_30-10                 191.4µ ± ∞ ¹   200.6µ ± ∞ ¹    +4.80% (p=0.008 n=5)
Ring_ShuffleShard/num_instances_=_1000,_num_zones_=_3,_shard_size_=_100-10                692.7µ ± ∞ ¹   719.2µ ± ∞ ¹         ~ (p=0.151 n=5)
Ring_ShuffleShard/num_instances_=_1000,_num_zones_=_3,_shard_size_=_1000-10              17.892m ± ∞ ¹   8.968m ± ∞ ¹   -49.87% (p=0.008 n=5)
Ring_ShuffleShardCached/num_instances_=_50,_num_zones_=_1,_shard_size_=_0-10            1688.00n ± ∞ ¹   68.99n ± ∞ ¹   -95.91% (p=0.008 n=5)
Ring_ShuffleShardCached/num_instances_=_50,_num_zones_=_1,_shard_size_=_3-10              149.4n ± ∞ ¹   148.7n ± ∞ ¹         ~ (p=0.802 n=5)
Ring_ShuffleShardCached/num_instances_=_50,_num_zones_=_1,_shard_size_=_10-10             366.3n ± ∞ ¹   357.0n ± ∞ ¹    -2.54% (p=0.032 n=5)
Ring_ShuffleShardCached/num_instances_=_50,_num_zones_=_1,_shard_size_=_30-10             1.027µ ± ∞ ¹   1.008µ ± ∞ ¹         ~ (p=0.151 n=5)
Ring_ShuffleShardCached/num_instances_=_50,_num_zones_=_1,_shard_size_=_100-10            1.671µ ± ∞ ¹   1.693µ ± ∞ ¹         ~ (p=0.548 n=5)
Ring_ShuffleShardCached/num_instances_=_50,_num_zones_=_1,_shard_size_=_1000-10           1.700µ ± ∞ ¹   1.648µ ± ∞ ¹    -3.06% (p=0.008 n=5)
Ring_ShuffleShardCached/num_instances_=_50,_num_zones_=_3,_shard_size_=_0-10            1698.00n ± ∞ ¹   69.67n ± ∞ ¹   -95.90% (p=0.008 n=5)
Ring_ShuffleShardCached/num_instances_=_50,_num_zones_=_3,_shard_size_=_3-10              149.1n ± ∞ ¹   148.5n ± ∞ ¹         ~ (p=0.968 n=5)
Ring_ShuffleShardCached/num_instances_=_50,_num_zones_=_3,_shard_size_=_10-10             431.9n ± ∞ ¹   426.0n ± ∞ ¹         ~ (p=0.310 n=5)
Ring_ShuffleShardCached/num_instances_=_50,_num_zones_=_3,_shard_size_=_30-10             1.015µ ± ∞ ¹   1.009µ ± ∞ ¹         ~ (p=0.103 n=5)
Ring_ShuffleShardCached/num_instances_=_50,_num_zones_=_3,_shard_size_=_100-10            1.691µ ± ∞ ¹   1.648µ ± ∞ ¹         ~ (p=0.151 n=5)
Ring_ShuffleShardCached/num_instances_=_50,_num_zones_=_3,_shard_size_=_1000-10           1.630µ ± ∞ ¹   1.668µ ± ∞ ¹         ~ (p=0.421 n=5)
Ring_ShuffleShardCached/num_instances_=_100,_num_zones_=_1,_shard_size_=_0-10           3602.00n ± ∞ ¹   69.57n ± ∞ ¹   -98.07% (p=0.008 n=5)
Ring_ShuffleShardCached/num_instances_=_100,_num_zones_=_1,_shard_size_=_3-10             147.5n ± ∞ ¹   147.0n ± ∞ ¹         ~ (p=0.690 n=5)
Ring_ShuffleShardCached/num_instances_=_100,_num_zones_=_1,_shard_size_=_10-10            360.3n ± ∞ ¹   365.2n ± ∞ ¹         ~ (p=0.222 n=5)
Ring_ShuffleShardCached/num_instances_=_100,_num_zones_=_1,_shard_size_=_30-10            1.027µ ± ∞ ¹   1.031µ ± ∞ ¹         ~ (p=0.690 n=5)
Ring_ShuffleShardCached/num_instances_=_100,_num_zones_=_1,_shard_size_=_100-10           3.604µ ± ∞ ¹   3.577µ ± ∞ ¹         ~ (p=1.000 n=5)
Ring_ShuffleShardCached/num_instances_=_100,_num_zones_=_1,_shard_size_=_1000-10          3.570µ ± ∞ ¹   3.581µ ± ∞ ¹         ~ (p=1.000 n=5)
Ring_ShuffleShardCached/num_instances_=_100,_num_zones_=_3,_shard_size_=_0-10           3608.00n ± ∞ ¹   69.46n ± ∞ ¹   -98.07% (p=0.008 n=5)
Ring_ShuffleShardCached/num_instances_=_100,_num_zones_=_3,_shard_size_=_3-10             149.9n ± ∞ ¹   150.3n ± ∞ ¹         ~ (p=0.421 n=5)
Ring_ShuffleShardCached/num_instances_=_100,_num_zones_=_3,_shard_size_=_10-10            424.4n ± ∞ ¹   429.2n ± ∞ ¹         ~ (p=0.841 n=5)
Ring_ShuffleShardCached/num_instances_=_100,_num_zones_=_3,_shard_size_=_30-10            1.014µ ± ∞ ¹   1.028µ ± ∞ ¹         ~ (p=1.000 n=5)
Ring_ShuffleShardCached/num_instances_=_100,_num_zones_=_3,_shard_size_=_100-10           3.606µ ± ∞ ¹   3.580µ ± ∞ ¹    -0.72% (p=0.032 n=5)
Ring_ShuffleShardCached/num_instances_=_100,_num_zones_=_3,_shard_size_=_1000-10          3.682µ ± ∞ ¹   3.545µ ± ∞ ¹    -3.72% (p=0.016 n=5)
Ring_ShuffleShardCached/num_instances_=_1000,_num_zones_=_1,_shard_size_=_0-10         56902.00n ± ∞ ¹   69.58n ± ∞ ¹   -99.88% (p=0.008 n=5)
Ring_ShuffleShardCached/num_instances_=_1000,_num_zones_=_1,_shard_size_=_3-10            146.5n ± ∞ ¹   145.9n ± ∞ ¹         ~ (p=0.690 n=5)
Ring_ShuffleShardCached/num_instances_=_1000,_num_zones_=_1,_shard_size_=_10-10           357.9n ± ∞ ¹   358.4n ± ∞ ¹         ~ (p=0.421 n=5)
Ring_ShuffleShardCached/num_instances_=_1000,_num_zones_=_1,_shard_size_=_30-10           1.023µ ± ∞ ¹   1.021µ ± ∞ ¹         ~ (p=0.889 n=5)
Ring_ShuffleShardCached/num_instances_=_1000,_num_zones_=_1,_shard_size_=_100-10          3.604µ ± ∞ ¹   3.604µ ± ∞ ¹         ~ (p=1.000 n=5)
Ring_ShuffleShardCached/num_instances_=_1000,_num_zones_=_1,_shard_size_=_1000-10         56.92µ ± ∞ ¹   57.21µ ± ∞ ¹         ~ (p=0.841 n=5)
Ring_ShuffleShardCached/num_instances_=_1000,_num_zones_=_3,_shard_size_=_0-10         57481.00n ± ∞ ¹   69.66n ± ∞ ¹   -99.88% (p=0.008 n=5)
Ring_ShuffleShardCached/num_instances_=_1000,_num_zones_=_3,_shard_size_=_3-10            144.6n ± ∞ ¹   144.6n ± ∞ ¹         ~ (p=0.817 n=5)
Ring_ShuffleShardCached/num_instances_=_1000,_num_zones_=_3,_shard_size_=_10-10           423.3n ± ∞ ¹   417.6n ± ∞ ¹         ~ (p=0.310 n=5)
Ring_ShuffleShardCached/num_instances_=_1000,_num_zones_=_3,_shard_size_=_30-10           1.023µ ± ∞ ¹   1.032µ ± ∞ ¹         ~ (p=0.738 n=5)
Ring_ShuffleShardCached/num_instances_=_1000,_num_zones_=_3,_shard_size_=_100-10          3.747µ ± ∞ ¹   3.627µ ± ∞ ¹         ~ (p=0.151 n=5)
Ring_ShuffleShardCached/num_instances_=_1000,_num_zones_=_3,_shard_size_=_1000-10         57.83µ ± ∞ ¹   56.68µ ± ∞ ¹    -1.99% (p=0.032 n=5)
Ring_ShuffleShard_512Tokens-10                                                            173.9µ ± ∞ ¹   173.3µ ± ∞ ¹    -0.35% (p=0.032 n=5)
Ring_ShuffleShard_LargeShardSize-10                                                       8.231m ± ∞ ¹   8.217m ± ∞ ¹         ~ (p=0.095 n=5)
Ring_ShuffleShard_ShardSize_0-10                                                     5152882.00n ± ∞ ¹   61.14n ± ∞ ¹  -100.00% (p=0.008 n=5)
geomean                                                                                   26.13µ         6.128µ         -76.54%
¹ need >= 6 samples for confidence interval at level 0.95

Which issue(s) this PR fixes:

Follow up to #560.

Checklist

  • Tests updated
  • [na] CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX]

ring/ring.go Outdated Show resolved Hide resolved
@jhalterman jhalterman force-pushed the filter-out-old-read-only-instances-on-read branch 2 times, most recently from f973823 to 0187781 Compare August 16, 2024 21:59
ring/ring.go Outdated Show resolved Hide resolved
ring/ring.go Show resolved Hide resolved
@pstibrany
Copy link
Member Author

I've rerun go test -run TestRing_ShuffleShardWithLookback_CorrectnessWithFuzzy -count=1000 -parallel 4 -timeout 1h after latest changes, and it did not report any errors.

TODO: test for new filterOutReadOnlyInstances method.

@pstibrany
Copy link
Member Author

TODO: test for new filterOutReadOnlyInstances method.

I've added shardSize=0 cases to TestRing_ShuffleShard and TestRing_ShuffleShardWithLookback tests to cover filtering of read-only instances in this method.

@pstibrany pstibrany marked this pull request as ready for review August 20, 2024 13:08
ring/ring.go Outdated Show resolved Hide resolved
@pstibrany pstibrany merged commit 70f84e5 into main Aug 23, 2024
6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants