-
Notifications
You must be signed in to change notification settings - Fork 3.8k
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
release-21.2: colbuilder: fall back to row-by-row processor wrapping for many renders #85860
Merged
yuzefovich
merged 2 commits into
cockroachdb:release-21.2
from
yuzefovich:backport21.2-85822
Aug 15, 2022
Merged
release-21.2: colbuilder: fall back to row-by-row processor wrapping for many renders #85860
yuzefovich
merged 2 commits into
cockroachdb:release-21.2
from
yuzefovich:backport21.2-85822
Aug 15, 2022
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
This commit adds a microbenchmark of queries with many render expressions. It'll be used in the following commit to tune when we fall back to wrapping a row-by-row processor to handle those renders. Release note: None
This commit introduces a mechanism to handle render expressions by wrapping a row-by-row processor into the vectorized flow when 1. the estimated number of rows to go through the renders is relatively small 2. the number of renders is relatively high. The idea is that the vectorized projection operators have higher overhead when many of them are planned AND there is not enough data to amortize the overhead, so to improve the performance in those cases we'll use the row-by-row noop processor. Both of the thresholds are controlled by cluster settings and the defaults were chosen based on a representative microbenchmark. It's worth pointing out that we only have the estimated row count for the scan operators, so the change has limited applicability. ``` RenderPlanning/rows=1/renders=1-24 407µs ± 2% 408µs ± 2% ~ (p=0.684 n=10+10) RenderPlanning/rows=1/renders=8-24 516µs ± 1% 537µs ± 1% +4.05% (p=0.000 n=10+10) RenderPlanning/rows=1/renders=32-24 832µs ± 1% 811µs ± 1% -2.59% (p=0.000 n=10+10) RenderPlanning/rows=1/renders=64-24 1.22ms ± 0% 1.14ms ± 1% -6.62% (p=0.000 n=9+10) RenderPlanning/rows=1/renders=128-24 2.02ms ± 0% 1.80ms ± 1% -11.18% (p=0.000 n=8+9) RenderPlanning/rows=1/renders=512-24 7.75ms ± 1% 5.75ms ± 1% -25.77% (p=0.000 n=10+9) RenderPlanning/rows=1/renders=4096-24 160ms ± 1% 62ms ± 1% -61.51% (p=0.000 n=10+9) RenderPlanning/rows=4/renders=1-24 438µs ± 2% 438µs ± 1% ~ (p=0.853 n=10+10) RenderPlanning/rows=4/renders=8-24 603µs ± 1% 633µs ± 2% +5.06% (p=0.000 n=10+10) RenderPlanning/rows=4/renders=32-24 1.08ms ± 1% 1.08ms ± 1% ~ (p=0.105 n=10+10) RenderPlanning/rows=4/renders=64-24 1.72ms ± 0% 1.62ms ± 0% -5.83% (p=0.000 n=9+9) RenderPlanning/rows=4/renders=128-24 3.01ms ± 1% 2.75ms ± 1% -8.78% (p=0.000 n=10+10) RenderPlanning/rows=4/renders=512-24 11.6ms ± 1% 9.6ms ± 2% -17.58% (p=0.000 n=10+10) RenderPlanning/rows=4/renders=4096-24 192ms ± 2% 91ms ± 2% -52.58% (p=0.000 n=10+10) RenderPlanning/rows=16/renders=1-24 494µs ± 1% 499µs ± 1% +1.03% (p=0.006 n=10+8) RenderPlanning/rows=16/renders=8-24 855µs ± 1% 901µs ± 1% +5.37% (p=0.000 n=10+10) RenderPlanning/rows=16/renders=32-24 2.03ms ± 1% 2.04ms ± 1% ~ (p=0.190 n=10+10) RenderPlanning/rows=16/renders=64-24 3.58ms ± 1% 3.42ms ± 1% -4.56% (p=0.000 n=10+9) RenderPlanning/rows=16/renders=128-24 6.74ms ± 1% 6.31ms ± 1% -6.37% (p=0.000 n=10+10) RenderPlanning/rows=16/renders=512-24 26.9ms ± 1% 24.7ms ± 1% -8.24% (p=0.000 n=9+10) RenderPlanning/rows=16/renders=4096-24 329ms ± 2% 218ms ± 2% -33.66% (p=0.000 n=10+10) RenderPlanning/rows=64/renders=1-24 666µs ± 1% 659µs ± 2% -1.07% (p=0.007 n=10+10) RenderPlanning/rows=64/renders=8-24 1.79ms ± 1% 1.84ms ± 1% +3.01% (p=0.000 n=10+10) RenderPlanning/rows=64/renders=32-24 5.53ms ± 1% 5.79ms ± 2% +4.74% (p=0.000 n=10+10) RenderPlanning/rows=64/renders=64-24 10.8ms ± 1% 11.0ms ± 1% +1.87% (p=0.000 n=10+9) RenderPlanning/rows=64/renders=128-24 21.2ms ± 1% 21.7ms ± 1% +2.71% (p=0.000 n=10+10) RenderPlanning/rows=64/renders=512-24 83.6ms ± 0% 84.9ms ± 0% +1.47% (p=0.000 n=10+7) RenderPlanning/rows=64/renders=4096-24 824ms ± 1% 751ms ± 2% -8.88% (p=0.000 n=10+10) RenderPlanning/rows=128/renders=1-24 853µs ± 1% 851µs ± 1% ~ (p=0.481 n=10+10) RenderPlanning/rows=128/renders=8-24 2.98ms ± 1% 3.11ms ± 1% +4.32% (p=0.000 n=10+10) RenderPlanning/rows=128/renders=32-24 10.4ms ± 1% 10.9ms ± 1% +5.44% (p=0.000 n=10+10) RenderPlanning/rows=128/renders=64-24 20.1ms ± 1% 21.3ms ± 1% +5.99% (p=0.000 n=10+10) RenderPlanning/rows=128/renders=128-24 39.7ms ± 1% 42.1ms ± 2% +5.98% (p=0.000 n=10+10) RenderPlanning/rows=128/renders=512-24 160ms ± 1% 168ms ± 2% +5.13% (p=0.000 n=9+10) RenderPlanning/rows=128/renders=4096-24 1.44s ± 1% 1.48s ± 2% +3.15% (p=0.000 n=9+10) RenderPlanning/rows=256/renders=1-24 1.22ms ± 1% 1.21ms ± 1% -1.01% (p=0.000 n=10+10) RenderPlanning/rows=256/renders=8-24 5.22ms ± 0% 5.19ms ± 1% -0.54% (p=0.011 n=8+9) RenderPlanning/rows=256/renders=32-24 19.9ms ± 1% 20.0ms ± 1% ~ (p=0.182 n=9+10) RenderPlanning/rows=256/renders=64-24 39.0ms ± 0% 38.9ms ± 0% -0.33% (p=0.023 n=10+10) RenderPlanning/rows=256/renders=128-24 76.8ms ± 1% 76.7ms ± 1% ~ (p=0.739 n=10+10) RenderPlanning/rows=256/renders=512-24 316ms ± 1% 319ms ± 1% +1.15% (p=0.001 n=9+10) RenderPlanning/rows=256/renders=4096-24 2.75s ± 1% 2.73s ± 1% -0.64% (p=0.002 n=8+9) ``` Release note: None
Thanks for opening a backport. Please check the backport criteria before merging:
If some of the basic criteria cannot be satisfied, ensure that the exceptional criteria are satisfied within.
Add a brief release justification to the body of your PR to justify this backport. Some other things to consider:
|
DrewKimball
approved these changes
Aug 10, 2022
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reviewable status: complete! 1 of 0 LGTMs obtained
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.
Backport 2/2 commits from #85822.
/cc @cockroachdb/release
colbuilder: add a microbenchmark for running many render expressions
This commit adds a microbenchmark of queries with many render
expressions. It'll be used in the following commit to tune when we fall
back to wrapping a row-by-row processor to handle those renders.
Release note: None
colbuilder: fall back to row-by-row processor wrapping for many renders
This commit introduces a mechanism to handle render expressions by
wrapping a row-by-row processor into the vectorized flow when
small
The idea is that the vectorized projection operators have higher
overhead when many of them are planned AND there is not enough data to
amortize the overhead, so to improve the performance in those cases
we'll use the row-by-row noop processor. Both of the thresholds are
controlled by cluster settings and the defaults were chosen based on
a representative microbenchmark.
It's worth pointing out that we only have the estimated row count for
the scan operators, so the change has limited applicability.
Fixes: #85632.
Release note: None
Release justification: performance regression fix.