Wasmtime GitHub notifications bot (Sep 16 2022 at 14:25): Wasmtime GitHub notifications bot (Sep 16 2022 at 14:25):alexcrichton commented on issue #4916:
I think this secret may not be on this repo yet, do you have access to add it or would you like me to add one?
Wasmtime GitHub notifications bot (Sep 16 2022 at 20:29):jlb6740 commented on issue #4916:
/bench_x64
Wasmtime GitHub notifications bot (Sep 19 2022 at 05:41):jlb6740 commented on issue #4916:
I think this secret may not be on this repo yet, do you have access to add it or would you like me to add one?
@alexcrichton
Forked branches don't have access to the secrets. Looks like a road block as neither clone or pushing will be authorized.https://github.com/actions/checkout/issues/298
I am wondering if the issue_comment trigger will allow reading the secret but the problem is the PR associated with the issue_comment is not made available via the event payload .. which is why we switched to using the pull_request_review event.
Wasmtime GitHub notifications bot (Sep 19 2022 at 14:20):alexcrichton commented on issue #4916:
I'm not sure I understand, why would we want forked branches to have access to anything? I think what's in the repo works, but https://github.com/bytecodealliance/wasmtime/pull/4919 needs to merge so the
mainbranch reflects correct secret configuration.
Wasmtime GitHub notifications bot (Sep 19 2022 at 17:27):jlb6740 commented on issue #4916:
I stumbled onto that link as the workflow/action was failing with the same error and I thought the conversation there made sense to apply here. However maybe I am wrong and that policy being referred to has nothing to do with the situation here. This is the line that I thought was relevant:
Secrets are not passed to workflows that are triggered by a pull request from a fork. Learn more.
It is written here: https://docs.github.com/en/actions/security-guides/encrypted-secrets#using-encrypted-secrets-in-a-workflow
Our workflow is triggered by pull_request_review and and the issue that I am seeing is that it appears that when the workflow, triggered that the secret is in fact not available. When I tested this I had to test on my own repo (forked from wasmtime), but the PR was not from a branch on a fork .. it was from a branch from the same repository that I was trying to merge as a PR into the default branch, so the secret was available. The patch is failing to run due to errors trying to access the private repo. Errors such as:
fatal: could not read Username for 'https://github.com/': No such device or address
Cloning into 'wasmtime-sightglass-benchmarking'...
fatal: could not read Password for 'https://***@github.com': No such device or address
Error: Process completed with exit code 128.Cloning into 'wasmtime-sightglass-benchmarking'...
fatal: could not read Username for 'https://github.com/': No such device or address
Error: Process completed with exit code 128.are seen when the token/secret is empty which is what I was seeing, or I am seeing:
remote: Invalid username or password.
fatal: Authentication failed for 'https://github.com/bytecodealliance/wasmtime-sightglass-benchmarking.git/'Which error depends on the various things I tried where I've lost track of which error is from which change I tried. So, I can't remember which one, but I believe one of these is consistent with the token/secret string being empty. The bottom line is we currently can't clone or push to the performance repo due to lack of permission. The secret is either not available or not working. I need to put some test in to confirm that it is simply not available, but the line "Secrets are not passed to workflows that are triggered by a pull request from a fork." I think makes sense as the culprit.
Wasmtime GitHub notifications bot (Sep 19 2022 at 17:27):jlb6740 edited a comment on issue #4916:
@alexcrichton I stumbled onto that link as the workflow/action was failing with the same error and I thought the conversation there made sense to apply here. However maybe I am wrong and that policy being referred to has nothing to do with the situation here. This is the line that I thought was relevant:
Secrets are not passed to workflows that are triggered by a pull request from a fork. Learn more.
It is written here: https://docs.github.com/en/actions/security-guides/encrypted-secrets#using-encrypted-secrets-in-a-workflow
Our workflow is triggered by pull_request_review and and the issue that I am seeing is that it appears that when the workflow, triggered that the secret is in fact not available. When I tested this I had to test on my own repo (forked from wasmtime), but the PR was not from a branch on a fork .. it was from a branch from the same repository that I was trying to merge as a PR into the default branch, so the secret was available. The patch is failing to run due to errors trying to access the private repo. Errors such as:
fatal: could not read Username for 'https://github.com/': No such device or address
Cloning into 'wasmtime-sightglass-benchmarking'...
fatal: could not read Password for 'https://***@github.com': No such device or address
Error: Process completed with exit code 128.Cloning into 'wasmtime-sightglass-benchmarking'...
fatal: could not read Username for 'https://github.com/': No such device or address
Error: Process completed with exit code 128.are seen when the token/secret is empty which is what I was seeing, or I am seeing:
remote: Invalid username or password.
fatal: Authentication failed for 'https://github.com/bytecodealliance/wasmtime-sightglass-benchmarking.git/'Which error depends on the various things I tried where I've lost track of which error is from which change I tried. So, I can't remember which one, but I believe one of these is consistent with the token/secret string being empty. The bottom line is we currently can't clone or push to the performance repo due to lack of permission. The secret is either not available or not working. I need to put some test in to confirm that it is simply not available, but the line "Secrets are not passed to workflows that are triggered by a pull request from a fork." I think makes sense as the culprit.
Wasmtime GitHub notifications bot (Sep 19 2022 at 17:50):alexcrichton commented on issue #4916:
I think this "just" needs to merge https://github.com/bytecodealliance/wasmtime/pull/4919, rebase this PR, and try again.
The
pull_request_reviewtrigger runs the workflow as-defined onmain. As-defined onmainthe workflow is accessingSIGHTGLASS_BENCHMARKING_TOKENwhich is not defined by the Wasmtime repo at this time. I believe this is why it's appearing that you don't have access to secrets.If my suspicions are correct then after merging #4919 this should work.
Wasmtime GitHub notifications bot (Sep 19 2022 at 18:05):jlb6740 commented on issue #4916:
I think this "just" needs to merge #4919, rebase this PR, and try again.
The
pull_request_reviewtrigger runs the workflow as-defined onmain. As-defined onmainthe workflow is accessingSIGHTGLASS_BENCHMARKING_TOKENwhich is not defined by the Wasmtime repo at this time. I believe this is why it's appearing that you don't have access to secrets.If my suspicions are correct then after merging #4919 this should work.
I wish it were more convenient to test these things with workflows and actions before merging but merging and just trying is probably the most efficient way to confirm. At least this doesn't run unless manually triggered so it should cause disruption even if it fails again.
Wasmtime GitHub notifications bot (Sep 19 2022 at 18:05):jlb6740 edited a comment on issue #4916:
I think this "just" needs to merge #4919, rebase this PR, and try again.
The
pull_request_reviewtrigger runs the workflow as-defined onmain. As-defined onmainthe workflow is accessingSIGHTGLASS_BENCHMARKING_TOKENwhich is not defined by the Wasmtime repo at this time. I believe this is why it's appearing that you don't have access to secrets.If my suspicions are correct then after merging #4919 this should work.
I wish it were more convenient to test workflows and actions before merging but merging and just trying is probably the most efficient way to confirm. At least this doesn't run unless manually triggered so it should cause disruption even if it fails again.
Wasmtime GitHub notifications bot (Sep 19 2022 at 18:06):jlb6740 edited a comment on issue #4916:
I think this "just" needs to merge #4919, rebase this PR, and try again.
The
pull_request_reviewtrigger runs the workflow as-defined onmain. As-defined onmainthe workflow is accessingSIGHTGLASS_BENCHMARKING_TOKENwhich is not defined by the Wasmtime repo at this time. I believe this is why it's appearing that you don't have access to secrets.If my suspicions are correct then after merging #4919 this should work.
I wish it were more convenient to test workflows and actions before merging but merging and just trying is probably the most efficient way to confirm. At least this doesn't run unless manually triggered so it shouldn't cause disruption even if it fails again.
Wasmtime GitHub notifications bot (Sep 19 2022 at 20:29):alexcrichton commented on issue #4916:
I used an equivalent token from the account that
PERSONAL_ACCESS_TOKENis connected to and I locally rangit clone ...and it worked, so I don't think it's token permissions at this point.That the run is still failing seems to mean that
pull_request_reviewactions don't have access to secrets. I... guess? That seems to confirm your hypothesis above which would unfortunately render most of this moot... To confirm that though, do you want to try sending a PR to this repo, from this repo, and see if the benchmarking works there?
Wasmtime GitHub notifications bot (Sep 20 2022 at 00:01):jlb6740 commented on issue #4916:
I used an equivalent token from the account that
PERSONAL_ACCESS_TOKENis connected to and I locally rangit clone ...and it worked, so I don't think it's token permissions at this point.That the run is still failing seems to mean that
pull_request_reviewactions don't have access to secrets. I... guess? That seems to confirm your hypothesis above which would unfortunately render most of this moot... To confirm that though, do you want to try sending a PR to this repo, from this repo, and see if the benchmarking works there?You mean to just push a branch from origin https://github.com/bytecodealliance/wasmtime.git and try from a PR on that branch? I don't have permission to push new branches .. maybe you could try? That would confirm things. If it is confirmed there are a couple of other ideas we can maybe try.
Wasmtime GitHub notifications bot (Sep 20 2022 at 14:05):alexcrichton commented on issue #4916:
I've opened https://github.com/bytecodealliance/wasmtime/pull/4928 to test.
Wasmtime GitHub notifications bot (Sep 20 2022 at 14:07):alexcrichton commented on issue #4916:
While waiting for the CI runner it says:
Job defined at: bytecodealliance/wasmtime/.github/workflows/performance.yml@refs/pull/4928/mergeso if it's using the
performance.ymlfrom the merge commit instead ofmainthen it definitely isn't going to have access to secrets. (since PRs could arbitrarily put whatever they want in workflow configuration files).Other events like
issue_commentmay guarantee a run frommainwhich would probably have access to secrets. Figuring out the PR context will be a bit nontrivial though.
Wasmtime GitHub notifications bot (Sep 20 2022 at 18:48):jlb6740 commented on issue #4916:
While waiting for the CI runner it says:
Job defined at: bytecodealliance/wasmtime/.github/workflows/performance.yml@refs/pull/4928/mergeso if it's using the
performance.ymlfrom the merge commit instead ofmainthen it definitely isn't going to have access to secrets. (since PRs could arbitrarily put whatever they want in workflow configuration files).Other events like
issue_commentmay guarantee a run frommainwhich would probably have access to secrets. Figuring out the PR context will be a bit nontrivial though.Looks like it ran (https://github.com/bytecodealliance/wasmtime/pull/4928) .. so that confirms things. Yes, it makes sense that forks should have access to those secrets. I was thinking either issue_comment and using github-script https://github.com/actions/github-script (such as in this example https://lightrun.com/answers/actions-checkout-any-way-to-checkout-pr-from-issue_comment-event) to figure it out, could work. Or setting up a workflow dispatch trigger may work. https://github.blog/changelog/2020-07-06-github-actions-manual-triggers-with-workflow_dispatch/ where the actor has to input their token or password into the gui and they'd need to have permission at the performance repo. I think the first option is the most desired.
Wasmtime GitHub notifications bot (Sep 20 2022 at 18:48):jlb6740 edited a comment on issue #4916:
While waiting for the CI runner it says:
Job defined at: bytecodealliance/wasmtime/.github/workflows/performance.yml@refs/pull/4928/mergeso if it's using the
performance.ymlfrom the merge commit instead ofmainthen it definitely isn't going to have access to secrets. (since PRs could arbitrarily put whatever they want in workflow configuration files).Other events like
issue_commentmay guarantee a run frommainwhich would probably have access to secrets. Figuring out the PR context will be a bit nontrivial though.Looks like it ran (https://github.com/bytecodealliance/wasmtime/pull/4928) .. so that confirms things. Yes, it makes sense that forks should not have access to those secrets. I was thinking either issue_comment and using github-script https://github.com/actions/github-script (such as in this example https://lightrun.com/answers/actions-checkout-any-way-to-checkout-pr-from-issue_comment-event) to figure it out, could work. Or setting up a workflow dispatch trigger may work. https://github.blog/changelog/2020-07-06-github-actions-manual-triggers-with-workflow_dispatch/ where the actor has to input their token or password into the gui and they'd need to have permission at the performance repo. I think the first option is the most desired.
Wasmtime GitHub notifications bot (Oct 04 2022 at 17:18):alexcrichton commented on issue #4916:
/bench_x64
Wasmtime GitHub notifications bot (Oct 04 2022 at 17:22):jlb6740 commented on issue #4916:
/bench_x64
Wasmtime GitHub notifications bot (Oct 04 2022 at 17:23):jlb6740 commented on issue #4916:
/bench_x64
@alexcrichton
This needed to be rebased in order to use the latest yml. I've done that and submitted another test trigger.
Wasmtime GitHub notifications bot (Oct 04 2022 at 17:23):jlb6740 commented on issue #4916:
Lol .. and accidently submitted another in my reply.
Wasmtime GitHub notifications bot (Oct 04 2022 at 17:44):jlb6740 commented on issue #4916:
Shows pct_change on x64 for the patch if merged compared to current head for main.
Pct_change is based on clocktick event cycles where the benchmarks are run with Sightglass.
A negative pct_change means clockticks are expected to be reduced for the benchmark,
for that phase, and by that factor, if the patch were merged (i.e. negative is good).
wasm arch phase pct_change benchmarks/blake3-scalar x86_64 Compilation -0.034287 benchmarks/blake3-scalar x86_64 Execution -0.011487 benchmarks/blake3-scalar x86_64 Instantiation 0.076832 benchmarks/blake3-simd x86_64 Compilation -0.022895 benchmarks/blake3-simd x86_64 Execution 0.001343 benchmarks/blake3-simd x86_64 Instantiation -0.134462 benchmarks/bz2 x86_64 Compilation -0.051869 benchmarks/bz2 x86_64 Execution 0.008466 benchmarks/bz2 x86_64 Instantiation -0.053237 benchmarks/intgemm-simd x86_64 Compilation 0.000813 benchmarks/intgemm-simd x86_64 Execution 0.000096 benchmarks/intgemm-simd x86_64 Instantiation 0.028721 benchmarks/meshoptimizer x86_64 Compilation -0.007054 benchmarks/meshoptimizer x86_64 Execution -0.001786 benchmarks/meshoptimizer x86_64 Instantiation 0.095982 benchmarks/noop x86_64 Compilation -0.009973 benchmarks/noop x86_64 Execution 0.403306 benchmarks/noop x86_64 Instantiation 0.007373 benchmarks/pulldown-cmark x86_64 Compilation 0.027008 benchmarks/pulldown-cmark x86_64 Execution -0.016214 benchmarks/pulldown-cmark x86_64 Instantiation -0.013706 benchmarks/shootout-ackermann x86_64 Compilation 0.035487 benchmarks/shootout-ackermann x86_64 Execution -0.053356 benchmarks/shootout-ackermann x86_64 Instantiation 0.001452 benchmarks/shootout-base64 x86_64 Compilation 0.013889 benchmarks/shootout-base64 x86_64 Execution 0.000187 benchmarks/shootout-base64 x86_64 Instantiation -0.006239 benchmarks/shootout-ctype x86_64 Compilation -0.034853 benchmarks/shootout-ctype x86_64 Execution -0.001521 benchmarks/shootout-ctype x86_64 Instantiation -0.014821 benchmarks/shootout-ed25519 x86_64 Compilation -0.007662 benchmarks/shootout-ed25519 x86_64 Execution -0.001859 benchmarks/shootout-ed25519 x86_64 Instantiation -0.125123 benchmarks/shootout-fib2 x86_64 Compilation 0.045457 benchmarks/shootout-fib2 x86_64 Execution 0.000295 benchmarks/shootout-fib2 x86_64 Instantiation -0.023764 benchmarks/shootout-gimli x86_64 Compilation 0.002144 benchmarks/shootout-gimli x86_64 Execution -0.018011 benchmarks/shootout-gimli x86_64 Instantiation -0.042722 benchmarks/shootout-heapsort x86_64 Compilation 0.019075 benchmarks/shootout-heapsort x86_64 Execution 0.000060 benchmarks/shootout-heapsort x86_64 Instantiation -0.061197 benchmarks/shootout-keccak x86_64 Compilation 0.003209 benchmarks/shootout-keccak x86_64 Execution 0.001870 benchmarks/shootout-keccak x86_64 Instantiation 0.020665 benchmarks/shootout-matrix x86_64 Compilation 0.012603 benchmarks/shootout-matrix x86_64 Execution 0.005769 benchmarks/shootout-matrix x86_64 Instantiation 0.004154 benchmarks/shootout-memmove x86_64 Compilation -0.020977 benchmarks/shootout-memmove x86_64 Execution 0.000839 benchmarks/shootout-memmove x86_64 Instantiation 0.007037 benchmarks/shootout-minicsv x86_64 Compilation 0.003152 benchmarks/shootout-minicsv x86_64 Execution 0.001506 benchmarks/shootout-minicsv x86_64 Instantiation -0.053419 benchmarks/shootout-nestedloop x86_64 Compilation -0.009232 benchmarks/shootout-nestedloop x86_64 Execution -0.023647 benchmarks/shootout-nestedloop x86_64 Instantiation -0.015899 benchmarks/shootout-random x86_64 Compilation -0.009856 benchmarks/shootout-random x86_64 Execution -0.000872 benchmarks/shootout-random x86_64 Instantiation -0.078992 benchmarks/shootout-ratelimit x86_64 Compilation -0.023974 benchmarks/shootout-ratelimit x86_64 Execution 0.012502 benchmarks/shootout-ratelimit x86_64 Instantiation -0.022845 benchmarks/shootout-seqhash x86_64 Compilation 0.004471 benchmarks/shootout-seqhash x86_64 Execution -0.000565 benchmarks/shootout-seqhash x86_64 Instantiation 0.000801 benchmarks/shootout-sieve x86_64 Compilation -0.065953 benchmarks/shootout-sieve x86_64 Execution 0.000828 benchmarks/shootout-sieve x86_64 Instantiation 0.030104 benchmarks/shootout-switch x86_64 Compilation -0.011299 benchmarks/shootout-switch x86_64 Execution 0.000224 benchmarks/shootout-switch x86_64 Instantiation 0.002284 benchmarks/shootout-xblabla20 x86_64 Compilation 0.034352 benchmarks/shootout-xblabla20 x86_64 Execution -0.015639 benchmarks/shootout-xblabla20 x86_64 Instantiation -0.158237 benchmarks/shootout-xchacha20 x86_64 Compilation 0.003882 benchmarks/shootout-xchacha20 x86_64 Execution -0.010573 benchmarks/shootout-xchacha20 x86_64 Instantiation 0.051459 benchmarks/spidermonkey x86_64 Compilation -0.005556 benchmarks/spidermonkey x86_64 Execution -0.001852 benchmarks/spidermonkey x86_64 Instantiation 0.004107
Wasmtime GitHub notifications bot (Oct 04 2022 at 17:46):jlb6740 commented on issue #4916:
Oh .. guess it did work
Wasmtime GitHub notifications bot (Oct 04 2022 at 18:03):jlb6740 commented on issue #4916:
Shows pct_change on x64 for the patch if merged compared to current head for main.
Pct_change is based on clocktick event cycles where the benchmarks are run with Sightglass.
A negative pct_change means clockticks are expected to be reduced for the benchmark,
for that phase, and by that factor, if the patch were merged (i.e. negative is good).
wasm arch phase pct_change benchmarks/blake3-scalar x86_64 Compilation -0.030310 benchmarks/blake3-scalar x86_64 Execution 0.023858 benchmarks/blake3-scalar x86_64 Instantiation 0.032645 benchmarks/blake3-simd x86_64 Compilation 0.251867 benchmarks/blake3-simd x86_64 Execution 0.111039 benchmarks/blake3-simd x86_64 Instantiation 0.096104 benchmarks/bz2 x86_64 Compilation -0.044554 benchmarks/bz2 x86_64 Execution -0.001013 benchmarks/bz2 x86_64 Instantiation 0.058163 benchmarks/intgemm-simd x86_64 Compilation -0.007175 benchmarks/intgemm-simd x86_64 Execution 0.001357 benchmarks/intgemm-simd x86_64 Instantiation -0.005574 benchmarks/meshoptimizer x86_64 Compilation 0.004291 benchmarks/meshoptimizer x86_64 Execution -0.000236 benchmarks/meshoptimizer x86_64 Instantiation -0.076082 benchmarks/noop x86_64 Compilation 0.051348 benchmarks/noop x86_64 Execution 0.074605 benchmarks/noop x86_64 Instantiation 0.104645 benchmarks/pulldown-cmark x86_64 Compilation 0.001547 benchmarks/pulldown-cmark x86_64 Execution -0.002676 benchmarks/pulldown-cmark x86_64 Instantiation 0.016475 benchmarks/shootout-ackermann x86_64 Compilation 0.077409 benchmarks/shootout-ackermann x86_64 Execution 0.164519 benchmarks/shootout-ackermann x86_64 Instantiation 0.038244 benchmarks/shootout-base64 x86_64 Compilation 0.018951 benchmarks/shootout-base64 x86_64 Execution -0.003764 benchmarks/shootout-base64 x86_64 Instantiation 0.059158 benchmarks/shootout-ctype x86_64 Compilation -0.044444 benchmarks/shootout-ctype x86_64 Execution 0.001686 benchmarks/shootout-ctype x86_64 Instantiation 0.019625 benchmarks/shootout-ed25519 x86_64 Compilation -0.007561 benchmarks/shootout-ed25519 x86_64 Execution -0.000823 benchmarks/shootout-ed25519 x86_64 Instantiation -0.032482 benchmarks/shootout-fib2 x86_64 Compilation 0.004821 benchmarks/shootout-fib2 x86_64 Execution 0.001347 benchmarks/shootout-fib2 x86_64 Instantiation 0.001049 benchmarks/shootout-gimli x86_64 Compilation 0.063835 benchmarks/shootout-gimli x86_64 Execution 0.004247 benchmarks/shootout-gimli x86_64 Instantiation 0.082987 benchmarks/shootout-heapsort x86_64 Compilation -0.032375 benchmarks/shootout-heapsort x86_64 Execution -0.000662 benchmarks/shootout-heapsort x86_64 Instantiation -0.028302 benchmarks/shootout-keccak x86_64 Compilation 0.012330 benchmarks/shootout-keccak x86_64 Execution 0.003932 benchmarks/shootout-keccak x86_64 Instantiation -0.046790 benchmarks/shootout-matrix x86_64 Compilation -0.029146 benchmarks/shootout-matrix x86_64 Execution -0.001596 benchmarks/shootout-matrix x86_64 Instantiation 0.022471 benchmarks/shootout-memmove x86_64 Compilation -0.004290 benchmarks/shootout-memmove x86_64 Execution -0.000002 benchmarks/shootout-memmove x86_64 Instantiation 0.006008 benchmarks/shootout-minicsv x86_64 Compilation 0.096753 benchmarks/shootout-minicsv x86_64 Execution -0.000926 benchmarks/shootout-minicsv x86_64 Instantiation 0.026605 benchmarks/shootout-nestedloop x86_64 Compilation 0.008242 benchmarks/shootout-nestedloop x86_64 Execution 0.017152 benchmarks/shootout-nestedloop x86_64 Instantiation 0.008496 benchmarks/shootout-random x86_64 Compilation -0.017752 benchmarks/shootout-random x86_64 Execution -0.000730 benchmarks/shootout-random x86_64 Instantiation -0.012865 benchmarks/shootout-ratelimit x86_64 Compilation -0.002059 benchmarks/shootout-ratelimit x86_64 Execution 0.002869 benchmarks/shootout-ratelimit x86_64 Instantiation 0.019441 benchmarks/shootout-seqhash x86_64 Compilation 0.030591 benchmarks/shootout-seqhash x86_64 Execution -0.002861 benchmarks/shootout-seqhash x86_64 Instantiation 0.002807 benchmarks/shootout-sieve x86_64 Compilation 0.001555 benchmarks/shootout-sieve x86_64 Execution -0.001753 benchmarks/shootout-sieve x86_64 Instantiation 0.049312 benchmarks/shootout-switch x86_64 Compilation -0.030542 benchmarks/shootout-switch x86_64 Execution -0.003779 benchmarks/shootout-switch x86_64 Instantiation 0.046580 benchmarks/shootout-xblabla20 x86_64 Compilation 0.026218 benchmarks/shootout-xblabla20 x86_64 Execution 0.001472 benchmarks/shootout-xblabla20 x86_64 Instantiation -0.101582 benchmarks/shootout-xchacha20 x86_64 Compilation 0.017953 benchmarks/shootout-xchacha20 x86_64 Execution 0.014455 benchmarks/shootout-xchacha20 x86_64 Instantiation 0.045538 benchmarks/spidermonkey x86_64 Compilation -0.025875 benchmarks/spidermonkey x86_64 Execution -0.000559 benchmarks/spidermonkey x86_64 Instantiation -0.007227
Wasmtime GitHub notifications bot (Oct 04 2022 at 18:09):jlb6740 commented on issue #4916:
Shows pct_change on x64 for the patch if merged compared to current head for main.
Pct_change is based on clocktick event cycles where the benchmarks are run with Sightglass.
A negative pct_change means clockticks are expected to be reduced for the benchmark,
for that phase, and by that factor, if the patch were merged (i.e. negative is good).
wasm arch phase pct_change benchmarks/blake3-scalar x86_64 Compilation -0.050030 benchmarks/blake3-scalar x86_64 Execution 0.010691 benchmarks/blake3-scalar x86_64 Instantiation -0.113561 benchmarks/blake3-simd x86_64 Compilation -0.018848 benchmarks/blake3-simd x86_64 Execution 0.024774 benchmarks/blake3-simd x86_64 Instantiation -0.038650 benchmarks/bz2 x86_64 Compilation -0.063704 benchmarks/bz2 x86_64 Execution -0.021275 benchmarks/bz2 x86_64 Instantiation 0.001148 benchmarks/intgemm-simd x86_64 Compilation -0.019306 benchmarks/intgemm-simd x86_64 Execution -0.006286 benchmarks/intgemm-simd x86_64 Instantiation -0.066674 benchmarks/meshoptimizer x86_64 Compilation -0.026582 benchmarks/meshoptimizer x86_64 Execution 0.000714 benchmarks/meshoptimizer x86_64 Instantiation 0.116023 benchmarks/noop x86_64 Compilation -0.042627 benchmarks/noop x86_64 Execution 0.083770 benchmarks/noop x86_64 Instantiation 0.012709 benchmarks/pulldown-cmark x86_64 Compilation -0.066239 benchmarks/pulldown-cmark x86_64 Execution 0.005557 benchmarks/pulldown-cmark x86_64 Instantiation -0.011457 benchmarks/shootout-ackermann x86_64 Compilation -0.041132 benchmarks/shootout-ackermann x86_64 Execution 0.003190 benchmarks/shootout-ackermann x86_64 Instantiation -0.049969 benchmarks/shootout-base64 x86_64 Compilation 0.013453 benchmarks/shootout-base64 x86_64 Execution 0.003252 benchmarks/shootout-base64 x86_64 Instantiation 0.022696 benchmarks/shootout-ctype x86_64 Compilation -0.037390 benchmarks/shootout-ctype x86_64 Execution 0.000918 benchmarks/shootout-ctype x86_64 Instantiation 0.027234 benchmarks/shootout-ed25519 x86_64 Compilation -0.012902 benchmarks/shootout-ed25519 x86_64 Execution -0.000566 benchmarks/shootout-ed25519 x86_64 Instantiation 0.003339 benchmarks/shootout-fib2 x86_64 Compilation 0.021018 benchmarks/shootout-fib2 x86_64 Execution 0.000560 benchmarks/shootout-fib2 x86_64 Instantiation -0.012342 benchmarks/shootout-gimli x86_64 Compilation 0.049840 benchmarks/shootout-gimli x86_64 Execution 0.021725 benchmarks/shootout-gimli x86_64 Instantiation 0.049550 benchmarks/shootout-heapsort x86_64 Compilation -0.008319 benchmarks/shootout-heapsort x86_64 Execution 0.001982 benchmarks/shootout-heapsort x86_64 Instantiation 0.043471 benchmarks/shootout-keccak x86_64 Compilation 0.011576 benchmarks/shootout-keccak x86_64 Execution 0.008654 benchmarks/shootout-keccak x86_64 Instantiation 0.003275 benchmarks/shootout-matrix x86_64 Compilation -0.062111 benchmarks/shootout-matrix x86_64 Execution 0.011016 benchmarks/shootout-matrix x86_64 Instantiation 0.012293 benchmarks/shootout-memmove x86_64 Compilation -0.045281 benchmarks/shootout-memmove x86_64 Execution 0.001623 benchmarks/shootout-memmove x86_64 Instantiation 0.027841 benchmarks/shootout-minicsv x86_64 Compilation -0.029276 benchmarks/shootout-minicsv x86_64 Execution 0.000702 benchmarks/shootout-minicsv x86_64 Instantiation 0.020804 benchmarks/shootout-nestedloop x86_64 Compilation 0.016071 benchmarks/shootout-nestedloop x86_64 Execution 0.063317 benchmarks/shootout-nestedloop x86_64 Instantiation 0.055417 benchmarks/shootout-random x86_64 Compilation 0.027010 benchmarks/shootout-random x86_64 Execution 0.002454 benchmarks/shootout-random x86_64 Instantiation 0.036361 benchmarks/shootout-ratelimit x86_64 Compilation 0.039516 benchmarks/shootout-ratelimit x86_64 Execution 0.006016 benchmarks/shootout-ratelimit x86_64 Instantiation 0.001117 benchmarks/shootout-seqhash x86_64 Compilation 0.009882 benchmarks/shootout-seqhash x86_64 Execution 0.000350 benchmarks/shootout-seqhash x86_64 Instantiation -0.009995 benchmarks/shootout-sieve x86_64 Compilation -0.005880 benchmarks/shootout-sieve x86_64 Execution 0.001220 benchmarks/shootout-sieve x86_64 Instantiation 0.033825 benchmarks/shootout-switch x86_64 Compilation -0.067979 benchmarks/shootout-switch x86_64 Execution 0.002203 benchmarks/shootout-switch x86_64 Instantiation -0.064441 benchmarks/shootout-xblabla20 x86_64 Compilation -0.049738 benchmarks/shootout-xblabla20 x86_64 Execution 0.000380 benchmarks/shootout-xblabla20 x86_64 Instantiation -0.002152 benchmarks/shootout-xchacha20 x86_64 Compilation -0.005208 benchmarks/shootout-xchacha20 x86_64 Execution 0.005069 benchmarks/shootout-xchacha20 x86_64 Instantiation -0.004924 benchmarks/spidermonkey x86_64 Compilation -0.021401 benchmarks/spidermonkey x86_64 Execution 0.002470 benchmarks/spidermonkey x86_64 Instantiation -0.004928
Wasmtime GitHub notifications bot (Oct 04 2022 at 20:18):cfallin commented on issue #4916:
@jlb6740 a question for you: does
pct_changeindicate a percent, as the column name implies, so e.g.0.002470means 0.00247% (i.e., 24.7 * 10^-6)? Or is it a raw ratio, so 0.247%?A few other comments on report formatting:
- It would be nice to sort overall by phase, so we have all the instantiation numbers together, and compilation together, and execution together. Generally if I'm testing a PR that e.g. updates the way instantiation is done, I want to look just at those numbers; I don't want to have to visually skip to every third row.
- Could we have a mean (geomean probably, rather than arithmetic) for each category?
These would make the report a little easier to grok at a glance than the sea of numbers we currently have!
Wasmtime GitHub notifications bot (Oct 04 2022 at 20:42):jlb6740 commented on issue #4916:
@jlb6740 a question for you: does
pct_changeindicate a percent, as the column name implies, so e.g.0.002470means 0.00247% (i.e., 24.7 * 10^-6)? Or is it a raw ratio, so 0.247%?
Pct_change is being calculated by pandas: https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.pct_change.html
See the examples there. Pct can be confusing because you don't know if you are supposed to factor in the 100x or not but here pandas is printing out what is really a factor. Maybe we should call it factor change to avoid confusion even though they call it pct_change.
- Note, I cut down on the number of runs so there might be more variability here that we can reduce at the expense of time to execute. Currently we have:
--processes 1 \ --iterations-per-process 2 \while the defaults are 10 process and 10 iterations per process.
A few other comments on report formatting:
- It would be nice to sort overall by phase, so we have all the instantiation numbers together, and compilation together, and execution together. Generally if I'm testing a PR that e.g. updates the way instantiation is done, I want to look just at those numbers; I don't want to have to visually skip to every third row.
Yeah, I agree. .. will be easy to do.
- Could we have a mean (geomean probably, rather than arithmetic) for each category?
These would make the report a little easier to grok at a glance than the sea of numbers we currently have!
Yes .. that is a good idea. I have some concepts in an internal benchmark that we have called wasmbench that is based on sightglass. It does things like calculate geomean as you describe and a ratio to native performance to produce a final score. Eventually I'd like to donate these to sightglass potentially a separate benchmark to sightglass but I am digressing .. yes I agree we should apply geomeans here!
Wasmtime GitHub notifications bot (Oct 04 2022 at 20:48):cfallin commented on issue #4916:
Maybe we should call it factor change to avoid confusion even though they call it pct_change.
Yes, I'd favor that; or multiply all the numbers by 100 so it truly is a percentage. I would argue that calling this value as-is
pct_changeis a bug: percent literally means "per hundred" and so if it's missing that factor of a hundred, it is incorrect.Given that, I'm now seeing these numbers in a new light. I see a bunch of variation on the order of 3-5%... for a PR with no changes. That's a very concerning level of variability: if our PRs sometimes make optimizations with effects on the order of 1% or less (but stable, measurable effects), this variability will simply swamp any such change and so will make the measurements unusable.
In order to trust these numbers to evaluate changes, I'd want to see a much lower noise floor -- something like +/- 0.1% at most. Or else we should filter any results out with a delta less than our known variability (i.e., not statistically significant). Could we try turning up the iteration count and see how that affects things?
Wasmtime GitHub notifications bot (Oct 04 2022 at 20:59):jlb6740 commented on issue #4916:
Yes, agreed we should do both .. crank up processes and iterations and print geo means.
Wasmtime GitHub notifications bot (Oct 17 2022 at 05:12):jlb6740 commented on issue #4916:
|Test|Test|
|1|1|
Wasmtime GitHub notifications bot (Oct 17 2022 at 05:13):jlb6740 commented on issue #4916:
Test Test 1 1
Wasmtime GitHub notifications bot (Oct 17 2022 at 05:14):jlb6740 commented on issue #4916:
Test Test 1 1
Test2 Test2 1 1
Wasmtime GitHub notifications bot (Oct 17 2022 at 18:10):jlb6740 commented on issue #4916:
:up::down:
Last updated: Nov 22 2024 at 16:03 UTC