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.

Sign up for GitHub

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

Jump to bottom

[ENHANCEMENT] TSDB: Faster exemplars #14080

Merged
merged 6 commits into from
May 30, 2024
Merged

[ENHANCEMENT] TSDB: Faster exemplars #14080

merged 6 commits into from
May 30, 2024
+52 −49

Conversation

bboreham
Copy link
Member

Mostly to change from a slice of pointers to a slice of exemplar entries.
Since the data structure is a circular buffer, every entry will be in-use anyway, except at startup before the buffer fills for the first time.

Also:

  • Reduce map lookups on exemplar index
  • Re-use buffer in Resize(), saves having to zero it every time.
  • Save map lookup on validation.

Couple of changes to the benchmarks, to make them more realistic. One exemplar per series is not a typical workload.
BenchmarkResizeExemplars shows 4x more memory, because the memory before resize is not counted by the benchmark.

Benchmarks:

goos: darwin
goarch: arm64
pkg: github.com/prometheus/prometheus/tsdb
                                           │  before.txt   │             after.txt              │
                                           │    sec/op     │   sec/op     vs base               │
AddExemplar/10000/1000-8                     1317.5µ ±  2%   749.0µ ± 1%  -43.15% (p=0.002 n=6)
AddExemplar/100000/1000-8                    13.076m ±  1%   7.412m ± 2%  -43.32% (p=0.002 n=6)
AddExemplar/1000000/1000-8                   129.91m ±  0%   73.86m ± 2%  -43.15% (p=0.002 n=6)
AddExemplar/10000/10000-8                    1189.6µ ±  1%   707.9µ ± 2%  -40.50% (p=0.002 n=6)
AddExemplar/100000/10000-8                   13.000m ±  0%   7.449m ± 1%  -42.70% (p=0.002 n=6)
AddExemplar/1000000/10000-8                  132.26m ±  1%   75.30m ± 1%  -43.06% (p=0.002 n=6)
AddExemplar/10000/100000-8                   1201.4µ ±  1%   703.0µ ± 2%  -41.49% (p=0.002 n=6)
AddExemplar/100000/100000-8                  11.699m ±  0%   6.849m ± 2%  -41.45% (p=0.002 n=6)
AddExemplar/1000000/100000-8                 131.36m ±  0%   75.10m ± 2%  -42.83% (p=0.002 n=6)
ResizeExemplars/grow-100000-to-200000-8       2.858m ±  1%   2.152m ± 3%  -24.70% (p=0.002 n=6)
ResizeExemplars/shrink-100000-to-50000-8     1407.2µ ±  1%   795.8µ ± 2%  -43.45% (p=0.002 n=6)
ResizeExemplars/grow-1000000-to-2000000-8     33.62m ± 10%   21.65m ± 1%  -35.59% (p=0.002 n=6)
ResizeExemplars/shrink-1000000-to-500000-8   13.785m ±  1%   7.575m ± 0%  -45.05% (p=0.002 n=6)
geomean                                       10.36m         6.110m       -41.00%

                                           │   before.txt   │               after.txt               │
                                           │      B/op      │     B/op       vs base                │
AddExemplar/10000/1000-8                      72.235Ki ± 0%    9.734Ki ± 0%   -86.52% (p=0.002 n=6)
AddExemplar/100000/1000-8                     161.30Ki ± 0%    98.80Ki ± 0%   -38.75% (p=0.002 n=6)
AddExemplar/1000000/1000-8                    1075.4Ki ± 0%   1012.9Ki ± 0%    -5.81% (p=0.002 n=6)
AddExemplar/10000/10000-8                    634.743Ki ± 0%    9.740Ki ± 0%   -98.47% (p=0.002 n=6)
AddExemplar/100000/10000-8                    723.80Ki ± 0%    98.80Ki ± 0%   -86.35% (p=0.002 n=6)
AddExemplar/1000000/10000-8                   1637.9Ki ± 0%   1012.9Ki ± 0%   -38.16% (p=0.002 n=6)
AddExemplar/10000/100000-8                   634.741Ki ± 0%    9.734Ki ± 0%   -98.47% (p=0.002 n=6)
AddExemplar/100000/100000-8                  6348.85Ki ± 0%    98.80Ki ± 0%   -98.44% (p=0.002 n=6)
AddExemplar/1000000/100000-8                  7262.9Ki ± 0%   1012.9Ki ± 0%   -86.05% (p=0.002 n=6)
ResizeExemplars/grow-100000-to-200000-8        1.984Mi ± 0%   11.140Mi ± 0%  +461.49% (p=0.002 n=6)
ResizeExemplars/shrink-100000-to-50000-8       511.9Ki ± 0%   2855.9Ki ± 0%  +457.92% (p=0.002 n=6)
ResizeExemplars/grow-1000000-to-2000000-8      22.32Mi ± 0%   113.87Mi ± 0%  +410.07% (p=0.002 n=6)
ResizeExemplars/shrink-1000000-to-500000-8     5.623Mi ± 0%   28.506Mi ± 0%  +406.93% (p=0.002 n=6)
geomean                                        1.291Mi         493.3Ki        -62.69%

                                           │  before.txt   │              after.txt               │
                                           │   allocs/op   │  allocs/op   vs base                 │
AddExemplar/10000/1000-8                       1499.0 ± 0%    499.0 ± 0%  -66.71% (p=0.002 n=6)
AddExemplar/100000/1000-8                      5.999k ± 0%   4.999k ± 0%  -16.67% (p=0.002 n=6)
AddExemplar/1000000/1000-8                     51.00k ± 0%   50.00k ± 0%   -1.96% (p=0.002 n=6)
AddExemplar/10000/10000-8                     10499.0 ± 0%    499.0 ± 0%  -95.25% (p=0.002 n=6)
AddExemplar/100000/10000-8                    14.999k ± 0%   4.999k ± 0%  -66.67% (p=0.002 n=6)
AddExemplar/1000000/10000-8                    60.00k ± 0%   50.00k ± 0%  -16.67% (p=0.002 n=6)
AddExemplar/10000/100000-8                    10499.0 ± 0%    499.0 ± 0%  -95.25% (p=0.002 n=6)
AddExemplar/100000/100000-8                  104.999k ± 0%   4.999k ± 0%  -95.24% (p=0.002 n=6)
AddExemplar/1000000/100000-8                  150.00k ± 0%   50.00k ± 0%  -66.67% (p=0.002 n=6)
ResizeExemplars/grow-100000-to-200000-8        1.004k ± 0%   1.004k ± 0%        ~ (p=1.000 n=6) ¹
ResizeExemplars/shrink-100000-to-50000-8        504.0 ± 0%    504.0 ± 0%        ~ (p=1.000 n=6) ¹
ResizeExemplars/grow-1000000-to-2000000-8      10.00k ± 0%   10.00k ± 0%        ~ (p=1.000 n=6) ¹
ResizeExemplars/shrink-1000000-to-500000-8     5.004k ± 0%   5.004k ± 0%        ~ (p=1.000 n=6) ¹
geomean                                        10.47k        3.905k       -62.70%
¹ all samples are equal

@bboreham bboreham requested review from codesome and removed request for jesusvazquez May 11, 2024 16:53
beorn7
beorn7 approved these changes May 15, 2024
View reviewed changes
Copy link
Member

@beorn7 beorn7 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Very nice.

codesome
codesome approved these changes May 29, 2024
View reviewed changes
Copy link
Member

@codesome codesome left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM. Can you fix the conflict?

@bboreham
[Test] TSDB: let BenchmarkAddExemplar reuse slots
929fbf8 
Test with different amounts of capacity and exemplars, so that sometimes
new exemplars are evicting older exemplars.

Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
@bboreham
[Test] TSDB: BenchmarkResizeExemplar multiple per series
f0c50b5 
One exemplar per series is not a typical workload. Make it the same as
`BenchmarkAddExemplar`.

Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
@bboreham
[ENHANCEMENT] TSDB: Reduce map lookups on exemplar index
3eb5581 
In many cases we already have a pointer to the entry.

Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
@bboreham
[ENHANCEMENT] TSDB: Eliminate pointer when storing exemplars
c0bb156 
Saves memory and effort.

Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
@bboreham
[ENHANCEMENT] TSDB: let Resize re-use buffer
7d98487 
This saves having to zero the buffer every time.

Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
@bboreham
[ENHANCEMENT] TSDB: Save map lookup on validation
3ee52ab 
Goes faster.

Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
@bboreham
Copy link
Member Author

Rebased to fix conflict.

@codesome codesome merged commit e47474d into prometheus:main May 30, 2024
25 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@beorn7 beorn7 beorn7 approved these changes

@codesome codesome codesome approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet
3 participants
@bboreham @beorn7 @codesome