Reduce memory allocations during tag and metric serialization #294
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Co-authored-by: Arthur Schreiber <arthurschreiber@github.com>
During memory profiling we noticed a high number of allocations
originating from the `Tag`- and `StatSerializer`.
This change reduces the amount of allocations on the happy path
by applying conditions to return early if the tag or metric names
don’t include edge case scenarios.
Please see the benchmark comparison below to see the effect
<details>
<summary>Before</summary>
```
rspec spec/statsd/serialization/tag_serializer_spec.rb:219
Run options: include {:locations=>{"./spec/statsd/serialization/tag_serializer_spec.rb"=>[219]}}
Datadog::Statsd::Serialization::TagSerializer
#format
benchmark
ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [x86_64-linux]
Warming up --------------------------------------
no tags 2.992M i/100ms
global tags 2.999M i/100ms
tags Array 161.844k i/100ms
tags Hash 92.146k i/100ms
tags Array + global tags
127.560k i/100ms
tags Hash + global tags
94.333k i/100ms
Calculating -------------------------------------
no tags 29.033M (± 4.4%) i/s - 146.589M in 5.059500s
global tags 28.592M (± 4.0%) i/s - 143.975M in 5.044502s
tags Array 1.690M (± 8.4%) i/s - 8.416M in 5.017852s
tags Hash 1.083M (± 6.2%) i/s - 5.437M in 5.042581s
tags Array + global tags
1.222M (± 8.2%) i/s - 6.123M in 5.044381s
tags Hash + global tags
831.802k (±12.1%) i/s - 4.151M in 5.074196s
Comparison:
no tags: 29033428.5 i/s
global tags: 28592301.5 i/s - same-ish: difference falls within error
tags Array: 1690469.5 i/s - 17.17x slower
tags Array + global tags: 1222161.7 i/s - 23.76x slower
tags Hash: 1082883.0 i/s - 26.81x slower
tags Hash + global tags: 831802.3 i/s - 34.90x slower
measure IPS
Calculating -------------------------------------
no tags 0.000 memsize ( 0.000 retained)
0.000 objects ( 0.000 retained)
0.000 strings ( 0.000 retained)
global tags 0.000 memsize ( 0.000 retained)
0.000 objects ( 0.000 retained)
0.000 strings ( 0.000 retained)
tags Array 240.000 memsize ( 0.000 retained)
5.000 objects ( 0.000 retained)
4.000 strings ( 0.000 retained)
tags Hash 400.000 memsize ( 0.000 retained)
8.000 objects ( 0.000 retained)
4.000 strings ( 0.000 retained)
tags Array + global tags
392.000 memsize ( 200.000 retained)
6.000 objects ( 5.000 retained)
4.000 strings ( 3.000 retained)
tags Hash + global tags
552.000 memsize ( 200.000 retained)
9.000 objects ( 5.000 retained)
4.000 strings ( 3.000 retained)
Comparison:
no tags: 0 allocated
global tags: 0 allocated - same
tags Array: 240 allocated - Infx more
tags Array + global tags: 392 allocated - Infx more
tags Hash: 400 allocated - Infx more
tags Hash + global tags: 552 allocated - Infx more
measure memory
Finished in 42.34 seconds (files took 0.09172 seconds to load)
2 examples, 0 failures
```
</details>
<details>
<summary>After</summary>
```
rspec spec/statsd/serialization/tag_serializer_spec.rb:219
Run options: include {:locations=>{"./spec/statsd/serialization/tag_serializer_spec.rb"=>[219]}}
Datadog::Statsd::Serialization::TagSerializer
#format
benchmark
ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [x86_64-linux]
Warming up --------------------------------------
no tags 3.014M i/100ms
global tags 2.956M i/100ms
tags Array 243.821k i/100ms
tags Hash 124.236k i/100ms
tags Array + global tags
165.977k i/100ms
tags Hash + global tags
97.519k i/100ms
Calculating -------------------------------------
no tags 29.620M (± 2.4%) i/s - 150.715M in 5.091350s
global tags 29.984M (± 2.1%) i/s - 150.757M in 5.030275s
tags Array 2.487M (± 2.1%) i/s - 12.435M in 5.003250s
tags Hash 1.330M (± 2.5%) i/s - 6.709M in 5.046724s
tags Array + global tags
1.663M (± 3.7%) i/s - 8.465M in 5.097958s
tags Hash + global tags
1.050M (± 3.5%) i/s - 5.266M in 5.021279s
Comparison:
global tags: 29984432.6 i/s
no tags: 29620311.3 i/s - same-ish: difference falls within error
tags Array: 2486521.6 i/s - 12.06x slower
tags Array + global tags: 1662829.5 i/s - 18.03x slower
tags Hash: 1330273.7 i/s - 22.54x slower
tags Hash + global tags: 1050157.0 i/s - 28.55x slower
measure IPS
Calculating -------------------------------------
no tags 0.000 memsize ( 0.000 retained)
0.000 objects ( 0.000 retained)
0.000 strings ( 0.000 retained)
global tags 0.000 memsize ( 0.000 retained)
0.000 objects ( 0.000 retained)
0.000 strings ( 0.000 retained)
tags Array 120.000 memsize ( 0.000 retained)
2.000 objects ( 0.000 retained)
1.000 strings ( 0.000 retained)
tags Hash 280.000 memsize ( 0.000 retained)
5.000 objects ( 0.000 retained)
4.000 strings ( 0.000 retained)
tags Array + global tags
272.000 memsize ( 80.000 retained)
3.000 objects ( 2.000 retained)
1.000 strings ( 0.000 retained)
tags Hash + global tags
432.000 memsize ( 240.000 retained)
6.000 objects ( 5.000 retained)
4.000 strings ( 3.000 retained)
Comparison:
no tags: 0 allocated
global tags: 0 allocated - same
tags Array: 120 allocated - Infx more
tags Array + global tags: 272 allocated - Infx more
tags Hash: 280 allocated - Infx more
tags Hash + global tags: 432 allocated - Infx more
measure memory
Finished in 42.27 seconds (files took 0.09182 seconds to load)
2 examples, 0 failures
```
</details>
<details>
<summary>Before</summary>
```
rspec spec/statsd/serialization/stat_serializer_spec.rb:97
Run options: include {:locations=>{"./spec/statsd/serialization/stat_serializer_spec.rb"=>[97]}}
Datadog::Statsd::Serialization::StatSerializer
benchmark
ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [x86_64-linux]
Warming up --------------------------------------
no tags 241.660k i/100ms
no tags + sample rate
134.558k i/100ms
with tags 101.980k i/100ms
with tags + sample rate
76.582k i/100ms
Calculating -------------------------------------
no tags 2.444M (± 1.0%) i/s - 12.325M in 5.043100s
no tags + sample rate
1.361M (± 1.5%) i/s - 6.862M in 5.043675s
with tags 1.020M (± 0.8%) i/s - 5.099M in 5.001806s
with tags + sample rate
763.286k (± 2.3%) i/s - 3.829M in 5.019571s
Comparison:
no tags: 2444105.8 i/s
no tags + sample rate: 1360915.8 i/s - 1.80x slower
with tags: 1019507.1 i/s - 2.40x slower
with tags + sample rate: 763286.0 i/s - 3.20x slower
measure IPS
Calculating -------------------------------------
no tags 160.000 memsize ( 0.000 retained)
4.000 objects ( 0.000 retained)
3.000 strings ( 0.000 retained)
no tags + sample rate
240.000 memsize ( 0.000 retained)
5.000 objects ( 0.000 retained)
4.000 strings ( 0.000 retained)
with tags 480.000 memsize ( 0.000 retained)
8.000 objects ( 0.000 retained)
7.000 strings ( 0.000 retained)
with tags + sample rate
520.000 memsize ( 0.000 retained)
9.000 objects ( 0.000 retained)
8.000 strings ( 0.000 retained)
Comparison:
no tags: 160 allocated
no tags + sample rate: 240 allocated - 1.50x more
with tags: 480 allocated - 3.00x more
with tags + sample rate: 520 allocated - 3.25x more
measure memory
Finished in 28.13 seconds (files took 0.09106 seconds to load)
2 examples, 0 failure
```
</details>
<details>
<summary>After</summary>
```
rspec spec/statsd/serialization/stat_serializer_spec.rb:97
Run options: include {:locations=>{"./spec/statsd/serialization/stat_serializer_spec.rb"=>[97]}}
Datadog::Statsd::Serialization::StatSerializer
benchmark
ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [x86_64-linux]
Warming up --------------------------------------
no tags 343.036k i/100ms
no tags + sample rate
159.735k i/100ms
with tags 140.044k i/100ms
with tags + sample rate
89.899k i/100ms
Calculating -------------------------------------
no tags 3.401M (± 3.8%) i/s - 17.152M in 5.052869s
no tags + sample rate
1.596M (± 0.4%) i/s - 7.987M in 5.004351s
with tags 1.400M (± 1.6%) i/s - 7.002M in 5.004013s
with tags + sample rate
960.181k (± 0.9%) i/s - 4.855M in 5.056339s
Comparison:
no tags: 3400803.1 i/s
no tags + sample rate: 1595983.7 i/s - 2.13x slower
with tags: 1399691.8 i/s - 2.43x slower
with tags + sample rate: 960181.4 i/s - 3.54x slower
measure IPS
Calculating -------------------------------------
no tags 120.000 memsize ( 0.000 retained)
3.000 objects ( 0.000 retained)
2.000 strings ( 0.000 retained)
no tags + sample rate
200.000 memsize ( 0.000 retained)
4.000 objects ( 0.000 retained)
3.000 strings ( 0.000 retained)
with tags 320.000 memsize ( 0.000 retained)
4.000 objects ( 0.000 retained)
3.000 strings ( 0.000 retained)
with tags + sample rate
360.000 memsize ( 0.000 retained)
5.000 objects ( 0.000 retained)
4.000 strings ( 0.000 retained)
Comparison:
no tags: 120 allocated
no tags + sample rate: 200 allocated - 1.67x more
with tags: 320 allocated - 2.67x more
with tags + sample rate: 360 allocated - 3.00x more
measure memory
Finished in 28.13 seconds (files took 0.08992 seconds to load)
2 examples, 0 failures
```
</details>
jhawthorn
reviewed
Sep 10, 2024
rayz
approved these changes
Sep 13, 2024
You're welcome. Thanks for the review! Could I bother you to take a look at #295 too? Do you know already when the next release will be cut? |
|
The new release should be published now 🙂 |
|
Hi Everyone, I guess this line introduced regression: |
|
Hi team, |
|
|
|
Fix WIP in #297 |
|
Hi, sorry for the delay. I've published a new release |
This was referenced Dec 13, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
During memory profiling we noticed a high number of allocations originating from the
Tag- andStatSerializer.This change reduces the amount of allocations on the happy path by applying conditions to return early if the tag or metric names don’t include edge case scenarios.
Please see the benchmark comparison below to see the effect
Before
After
Before
After