Repository navigation
398 lines (335 loc) · 19 KB
/
Copy pathcomment-profiling-changes.yaml
File metadata and controls
398 lines (335 loc) · 19 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
name: "Profiling Changes"
on:
pull_request:
# Everything the benchmarks build from or load, which also makes up the baseline's cache key below
paths: &benchmark-inputs
- "node-graph/**"
- "document/**"
- "libraries/**"
- "demo-artwork/**"
- "Cargo.toml"
- "Cargo.lock"
- ".cargo/config.toml"
- ".github/workflows/comment-profiling-changes.yaml"
# Measures each master commit's baseline once, so PR runs can restore it rather than measuring their base commit themselves
push:
branches:
- master
paths: *benchmark-inputs
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}
cancel-in-progress: true
env:
CARGO_TERM_COLOR: always
# The four benchmarks, built and run by one `cargo bench` call so their slow thin-LTO links happen in parallel
BENCHMARKS: --bench compile_demo_art_gungraun --bench update_executor_gungraun --bench run_once_gungraun --bench run_cached_gungraun
# The brush benchmark, which needs the GPU nodes and so gets its own call rather than changing what the others build
BRUSH_BENCHMARK: -p interpreted-executor --features gpu --bench brush_stroke_gungraun
jobs:
profile:
# Fork PRs always run on GitHub's own runners, while the `LINUX_RUNNER` repo variable can point the rest at another provider
runs-on: ${{ github.event.pull_request.head.repo.fork && 'ubuntu-latest' || vars.LINUX_RUNNER || 'ubuntu-latest' }}
continue-on-error: true
permissions:
contents: read
pull-requests: write
steps:
- uses: actions/checkout@d23441a48e516b6c34aea4fa41551a30e30af803 # v6.1.0
with:
# A PR's merge commit plus its first parent, the master commit it's measured against
fetch-depth: 2
- name: Install Rust
uses: dtolnay/rust-toolchain@b725eda525362e7cf34af9a7243cac1e3f01a7ec # stable
- name: Cache Rust dependencies
uses: Swatinem/rust-cache@d1527c8f67ef9af357290475d4af4b0fab8f12e0 # v2.9.3
with:
# Cache on Cargo.lock file
cache-on-failure: true
- name: Install gungraun-runner and valgrind
uses: gungraun/setup-gungraun@62591572561bb2f4ed2ae300b8a8bea99ea8ddaf # v1.1.3
- name: Install lavapipe
run: |
# The runners have no GPU, so the brush benchmark renders on Mesa's software Vulkan device
sudo apt-get update
sudo apt-get install -y mesa-vulkan-drivers
# Mesa releases name the ICD with or without the architecture (`lvp_icd.x86_64.json` or `lvp_icd.json`), so ask the package where it put it
icd=$(dpkg -L mesa-vulkan-drivers | grep '/icd\.d/lvp_icd.*\.json$')
echo "VK_DRIVER_FILES=$icd" >> "$GITHUB_ENV"
- name: Checkout baseline commit
if: github.event_name == 'pull_request'
run: git checkout --detach HEAD^1
- name: Get toolchain info
id: toolchain-info
run: |
# Instruction counts match across x64 CPU models, but new releases of Rust, Valgrind, or the C library change them
echo "hash=$( (rustc -vV; valgrind --version; ldd --version | sed -n 1p) | sha256sum | cut -c1-8)" >> $GITHUB_OUTPUT
- name: Restore benchmark baseline
id: restore-baseline
uses: actions/cache/restore@caa296126883cff596d87d8935842f9db880ef25 # v5.1.0
with:
path: target/gungraun
# Keyed by the baseline commit's benchmark inputs, so master commits that don't change them share one baseline
key: ${{ runner.os }}-${{ runner.arch }}-${{ steps.toolchain-info.outputs.hash }}-gungraun-baseline-${{ hashFiles('node-graph/**', 'document/**', 'libraries/**', 'demo-artwork/**', 'Cargo.toml', 'Cargo.lock', '.cargo/config.toml', '.github/workflows/comment-profiling-changes.yaml') }}
- name: Run baseline benchmarks
if: steps.restore-baseline.outputs.cache-hit != 'true'
run: |
cargo bench $BENCHMARKS -- --save-baseline=master
# Until the brush benchmark reaches master, a baseline commit may predate it
if [ -f node-graph/interpreted-executor/benches/brush_stroke_gungraun.rs ]; then
cargo bench $BRUSH_BENCHMARK -- --save-baseline=master
fi
# Saved before the PR benchmarks run, since they write their own results into the same folder
- name: Save benchmark baseline
if: steps.restore-baseline.outputs.cache-hit != 'true'
uses: actions/cache/save@caa296126883cff596d87d8935842f9db880ef25 # v5.1.0
with:
path: target/gungraun
key: ${{ steps.restore-baseline.outputs.cache-primary-key }}
- name: Checkout PR merge commit
if: github.event_name == 'pull_request'
run: git checkout --detach ${{ github.sha }}
- name: Run PR benchmarks
if: github.event_name == 'pull_request'
# Unlike the default, this sets `pipefail`, so a failed benchmark fails the step rather than leaving its results empty
shell: bash
run: |
# The brush benchmark has nothing to compare against until the baseline commit has it
git cat-file -e HEAD^1:node-graph/interpreted-executor/benches/brush_stroke_gungraun.rs 2>/dev/null && brush=true || brush=false
{
cargo bench $BENCHMARKS -- --baseline=master --output-format=json
if $brush; then cargo bench $BRUSH_BENCHMARK -- --baseline=master --output-format=json; fi
} | jq -sc | sed 's/\\"//g' > /tmp/all_output.json
# Split the results into one file per benchmark for the analysis below
split_results() { jq -c --arg prefix "$1::" '[.[] | select(.module_path | startswith($prefix))]' /tmp/all_output.json > "/tmp/$2_output.json"; }
split_results compile_demo_art_gungraun compile
split_results update_executor_gungraun update
split_results run_once_gungraun run_once
split_results run_cached_gungraun run_cached
if $brush; then
split_results brush_stroke_gungraun brush_stroke
fi
- name: Make old comments collapsed by default
# Only run if we have write permissions (not a fork)
if: github.event.pull_request.head.repo.full_name == github.repository
uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
with:
github-token: ${{secrets.GITHUB_TOKEN}}
script: |
const { data: comments } = await github.rest.issues.listComments({
issue_number: context.issue.number,
owner: context.repo.owner,
repo: context.repo.repo,
});
const botComments = comments.filter((comment) =>
comment.user.type === 'Bot' && comment.body.includes('Performance Benchmark Results') && comment.body.includes('<details open>')
);
for (const comment of botComments) {
// Edit the comment to remove the "open" attribute from the <details> tag
await github.rest.issues.updateComment({
owner: context.repo.owner,
repo: context.repo.repo,
comment_id: comment.id,
body: comment.body.replace('<details open>', '<details>')
});
}
- name: Analyze profiling changes
id: analyze
if: github.event_name == 'pull_request'
uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
with:
script: |
const fs = require('fs');
function isSignificantChange(diffPct, absoluteChange, benchmarkType) {
const meetsPercentageThreshold = Math.abs(diffPct) > 5;
const meetsAbsoluteThreshold = absoluteChange > 200000;
const isCachedExecution = benchmarkType === 'run_cached' ||
benchmarkType.includes('Cached Execution');
return isCachedExecution
? (meetsPercentageThreshold && meetsAbsoluteThreshold)
: meetsPercentageThreshold;
}
// Each Callgrind metric holds its counts in `values.old` and `values.new`, plus `change.diff_pct` (a string) when compared against the baseline
function callgrindMetrics(benchmark) {
return benchmark.profiles?.find((profile) => profile.tool === 'Callgrind')?.data?.total?.metrics;
}
// The brush benchmark only has results once the baseline commit has it too
const brush = fs.existsSync('/tmp/brush_stroke_output.json');
const allOutputs = [
JSON.parse(fs.readFileSync('/tmp/compile_output.json', 'utf8')),
JSON.parse(fs.readFileSync('/tmp/update_output.json', 'utf8')),
JSON.parse(fs.readFileSync('/tmp/run_once_output.json', 'utf8')),
JSON.parse(fs.readFileSync('/tmp/run_cached_output.json', 'utf8')),
...(brush ? [JSON.parse(fs.readFileSync('/tmp/brush_stroke_output.json', 'utf8'))] : [])
];
const outputNames = ['compile', 'update', 'run_once', 'run_cached', ...(brush ? ['brush_stroke'] : [])];
const sectionTitles = ['Compilation', 'Update', 'Run Once', 'Cached Execution', ...(brush ? ['Brush Stroke'] : [])];
// Fail instead of reporting no changes when a benchmark has no baseline comparison, such as after gungraun changes its output format
const missingComparisons = [];
allOutputs.forEach((benchmarkOutput, i) => {
if (benchmarkOutput.length === 0) missingComparisons.push(`${outputNames[i]} (no results)`);
benchmarkOutput
.filter((benchmark) => !callgrindMetrics(benchmark)?.Ir?.change)
.forEach((benchmark) => missingComparisons.push(`${benchmark.module_path} ${benchmark.id}`));
});
if (missingComparisons.length > 0) {
core.setFailed(`Benchmark results without a baseline comparison: ${missingComparisons.join(', ')}`);
return;
}
let hasSignificantChanges = false;
let hasRegressions = false;
let regressionDetails = [];
allOutputs.forEach((benchmarkOutput, i) => {
const outputName = outputNames[i];
const sectionTitle = sectionTitles[i];
benchmarkOutput.forEach((benchmark) => {
const instructions = callgrindMetrics(benchmark).Ir;
const diffPct = parseFloat(instructions.change.diff_pct);
const absoluteChange = Math.abs(instructions.values.new - instructions.values.old);
if (isSignificantChange(diffPct, absoluteChange, outputName)) {
hasSignificantChanges = true;
// Only an increase in instruction count is a regression; improvements must not fail CI.
if (diffPct > 0) {
hasRegressions = true;
regressionDetails.push({
module_path: benchmark.module_path,
id: benchmark.id,
diffPct,
absoluteChange,
sectionTitle
});
}
}
});
});
core.setOutput('has-significant-changes', hasSignificantChanges);
core.setOutput('has-regressions', hasRegressions);
core.setOutput('regression-details', JSON.stringify(regressionDetails));
- name: Comment PR
if: github.event.pull_request.head.repo.full_name == github.repository
uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
with:
github-token: ${{secrets.GITHUB_TOKEN}}
script: |
const fs = require('fs');
const compileOutput = JSON.parse(fs.readFileSync('/tmp/compile_output.json', 'utf8'));
const updateOutput = JSON.parse(fs.readFileSync('/tmp/update_output.json', 'utf8'));
const runOnceOutput = JSON.parse(fs.readFileSync('/tmp/run_once_output.json', 'utf8'));
const runCachedOutput = JSON.parse(fs.readFileSync('/tmp/run_cached_output.json', 'utf8'));
// The brush benchmark only has results once the baseline commit has it too, and a section without results is left out
const readOptional = (path) => (fs.existsSync(path) ? JSON.parse(fs.readFileSync(path, 'utf8')) : []);
const brushStrokeOutput = readOptional('/tmp/brush_stroke_output.json');
const hasSignificantChanges = '${{ steps.analyze.outputs.has-significant-changes }}' === 'true';
let commentBody = "";
function formatNumber(num) {
return String(num).replace(/\B(?=(\d{3})+(?!\d))/g, ",");
}
function formatPercentage(pct) {
const sign = pct >= 0 ? '+' : '';
return `${sign}${pct.toFixed(2)}%`;
}
function padRight(str, len) {
return str.padEnd(len);
}
function padLeft(str, len) {
return str.padStart(len);
}
function callgrindMetrics(benchmark) {
return benchmark.profiles?.find((profile) => profile.tool === 'Callgrind')?.data?.total?.metrics;
}
function processBenchmarkOutput(benchmarkOutput, sectionTitle, isLast = false) {
let sectionBody = "";
let hasResults = false;
let hasSignificantChanges = false;
function isSignificantChange(diffPct, absoluteChange, benchmarkType) {
const meetsPercentageThreshold = Math.abs(diffPct) > 5;
const meetsAbsoluteThreshold = absoluteChange > 200000;
const isCachedExecution = benchmarkType === 'run_cached' ||
benchmarkType.includes('Cached Execution');
return isCachedExecution
? (meetsPercentageThreshold && meetsAbsoluteThreshold)
: meetsPercentageThreshold;
}
benchmarkOutput.forEach((benchmark) => {
const metrics = callgrindMetrics(benchmark);
const instructions = metrics?.Ir;
if (!instructions?.change) return;
hasResults = true;
const diffPct = parseFloat(instructions.change.diff_pct);
const changePercentage = formatPercentage(diffPct);
const color = diffPct > 0 ? "red" : "lime";
const name = [benchmark.id, benchmark.description].filter(Boolean).join(':');
sectionBody += `**${benchmark.module_path} ${name}**\n`;
sectionBody += `Instructions: \`${formatNumber(instructions.values.old)}\` (master) → \`${formatNumber(instructions.values.new)}\` (HEAD) : `;
sectionBody += `$$\\color{${color}}${changePercentage.replace("%", "\\\\%")}$$\n\n`;
sectionBody += "<details>\n<summary>Detailed metrics</summary>\n\n```\n";
sectionBody += `Baselines: master| HEAD\n`;
Object.entries(metrics).forEach(([metricName, metricData]) => {
if (!metricData.change) return;
const changePercentage = formatPercentage(parseFloat(metricData.change.diff_pct));
const line = `${padRight(metricName, 20)} ${padLeft(formatNumber(Math.round(metricData.values.old)), 11)}|${padLeft(formatNumber(Math.round(metricData.values.new)), 11)} ${padLeft(changePercentage, 15)}`;
sectionBody += `${line}\n`;
});
sectionBody += "```\n</details>\n\n";
if (isSignificantChange(diffPct, Math.abs(instructions.values.new - instructions.values.old), sectionTitle)) {
hasSignificantChanges = true;
}
});
if (hasResults) {
// Wrap section in collapsible details, open only if there are significant changes
const openAttribute = hasSignificantChanges ? " open" : "";
const ruler = isLast ? "" : "\n\n---";
return `<details${openAttribute}>\n<summary><h2>${sectionTitle}</h2></summary>\n\n${sectionBody}${ruler}\n</details>`;
}
return "";
}
// Process each benchmark category
const sections = [
{ output: compileOutput, title: "🔧 Graph Compilation" },
{ output: updateOutput, title: "🔄 Executor Update" },
{ output: runOnceOutput, title: "🚀 Render: Cold Execution" },
{ output: runCachedOutput, title: "⚡ Render: Cached Execution" },
{ output: brushStrokeOutput, title: "🖌️ Render: Brush Strokes" }
];
// Generate sections and determine which ones have results
const generatedSections = sections.map(({ output, title }) =>
processBenchmarkOutput(output, title, true) // temporarily mark all as last
).filter(section => section.length > 0);
// Re-generate with correct isLast flags
let sectionIndex = 0;
const finalSections = sections.map(({ output, title }) => {
const section = processBenchmarkOutput(output, title, true); // check if it has results
if (section.length > 0) {
const isLast = sectionIndex === generatedSections.length - 1;
sectionIndex++;
return processBenchmarkOutput(output, title, isLast);
}
return "";
}).filter(section => section.length > 0);
// Combine all sections
commentBody = finalSections.join("\n\n");
if (commentBody.length > 0) {
const output = `<details open>\n<summary>Performance Benchmark Results</summary>\n\n${commentBody}\n</details>`;
if (hasSignificantChanges) {
github.rest.issues.createComment({
issue_number: context.issue.number,
owner: context.repo.owner,
repo: context.repo.repo,
body: output
});
} else {
console.log("No significant performance changes detected. Skipping comment.");
console.log(output);
}
} else {
console.log("No benchmark results to display.");
}
- name: Fail on significant regressions
if: steps.analyze.outputs.has-regressions == 'true'
uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
env:
REGRESSION_DETAILS: ${{ steps.analyze.outputs.regression-details }}
with:
script: |
const regressionDetails = JSON.parse(process.env.REGRESSION_DETAILS);
const firstRegression = regressionDetails[0];
core.setFailed(`Significant performance regression detected: ${firstRegression.module_path} ${firstRegression.id} increased by ${firstRegression.absoluteChange.toLocaleString()} instructions (${firstRegression.diffPct.toFixed(2)}%)`);