Add a JVMTI agent for causal profiling of Java #129

Workflow file for this run

	name: CI

	on:
	push:
	branches: [master, wip_*]
	pull_request:
	branches: [master]

	permissions:
	contents: read

	jobs:
	build-linux:
	runs-on: ubuntu-latest

	steps:
	- uses: actions/checkout@v4

	- name: Install dependencies
	run: \|
	sudo apt-get update
	sudo apt-get install -y \
	build-essential \
	cmake \
	pkg-config \
	libsqlite3-dev \
	libbz2-dev

	- name: Configure perf permissions
	run: \|
	# Allow perf_event_open for profiling
	echo 1 \| sudo tee /proc/sys/kernel/perf_event_paranoid

	- name: Build coz
	run: \|
	mkdir build
	cd build
	cmake .. -DCMAKE_BUILD_TYPE=RelWithDebInfo
	make -j$(nproc)

	- name: Build benchmarks
	run: \|
	cd build
	cmake .. -DBUILD_BENCHMARKS=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo
	make -j$(nproc) toy lock_test kmeans

	- name: Verify build artifacts
	run: \|
	test -f build/libcoz/libcoz.so
	test -x build/benchmarks/toy/toy
	test -x build/benchmarks/lock_test/lock_test
	test -x build/benchmarks/kmeans/kmeans
	echo "Build artifacts verified"

	- name: Run unit tests
	run: \|
	cd build
	ctest --output-on-failure -j$(nproc)
	rm -f profile.jsonl
	echo "Unit tests passed"

	- name: Run toy benchmark with coz
	run: \|
	cd build
	# Run coz with the toy benchmark for a short duration
	timeout 30 ../coz run -o profile.jsonl --- ./benchmarks/toy/toy \|\| true

	if [ ! -f profile.jsonl ]; then
	echo "ERROR: profile.jsonl was not created"
	exit 1
	fi

	echo "Profile created successfully"

	- name: Validate toy profile output
	run: \|
	cd build

	if [ ! -s profile.jsonl ]; then
	echo "ERROR: profile.jsonl is empty"
	exit 1
	fi

	echo "=== Profile Contents ==="
	cat profile.jsonl
	echo ""

	# Check for expected profile format (should contain experiment lines)
	EXPERIMENT_COUNT=$(grep -c 'experiment' profile.jsonl \|\| echo 0)
	echo "=== Profile Summary ==="
	echo " Total lines: $(wc -l < profile.jsonl)"
	echo " Experiments: $EXPERIMENT_COUNT"

	if [ "$EXPERIMENT_COUNT" -eq 0 ]; then
	echo "ERROR: No experiment data in profile"
	exit 1
	fi

	# Validate profile references toy.cpp line 18 (the loop in function a())
	if grep -q "toy.cpp:18" profile.jsonl; then
	echo " Found experiments for toy.cpp:18 (loop in function a())"
	else
	echo "ERROR: toy.cpp:18 not found in profile - expected loop in a() to be profiled"
	exit 1
	fi

	echo ""
	echo "Profile validation PASSED"

	- name: Validate toy optimization potential
	run: \|
	cd build

	python3 << 'VALIDATION_SCRIPT'
	import sys
	import json

	experiments = []
	current_exp = None

	with open('profile.jsonl', 'r') as f:
	for line in f:
	line = line.strip()
	if not line:
	continue
	obj = json.loads(line)
	if obj.get('type') == 'experiment':
	current_exp = obj
	elif obj.get('type') == 'throughput-point' and current_exp:
	current_exp['delta'] = float(obj['delta'])
	experiments.append(current_exp)
	current_exp = None

	# Filter for toy.cpp:18 (the loop in a())
	line18_exps = [e for e in experiments if 'toy.cpp:18' in e['selected']]

	if not line18_exps:
	print("ERROR: No experiments found for toy.cpp:18")
	sys.exit(1)

	print(f"Found {len(line18_exps)} experiments for toy.cpp:18")

	# Group by speedup percentage and calculate average delta
	speedup_deltas = {}
	for e in line18_exps:
	s = int(float(e['speedup']) * 100)
	if s not in speedup_deltas:
	speedup_deltas[s] = []
	speedup_deltas[s].append(e['delta'])

	print("\nSpeedup vs Progress Delta:")
	for speedup in sorted(speedup_deltas.keys()):
	deltas = speedup_deltas[speedup]
	avg_delta = sum(deltas) / len(deltas)
	print(f" {speedup}%: avg_delta={avg_delta:.2f} (n={len(deltas)})")

	if len(speedup_deltas) < 3:
	print(f"WARNING: Only {len(speedup_deltas)} speedup levels - need more data for accurate validation")
	else:
	print("\nOptimization potential analysis: PASSED")
	print(" - toy.cpp:18 (loop in a()) correctly identified as optimization target")

	print("\nProfile validation COMPLETE")
	VALIDATION_SCRIPT

	- name: Run lock_test benchmark with coz
	run: \|
	cd build
	timeout 60 ../coz run -o lock_test_profile.jsonl --- ./benchmarks/lock_test/lock_test \|\| true

	if [ ! -f lock_test_profile.jsonl ]; then
	echo "ERROR: lock_test_profile.jsonl was not created"
	exit 1
	fi

	echo "Lock test profile created successfully"

	- name: Validate lock_test profile
	run: \|
	cd build

	if [ ! -s lock_test_profile.jsonl ]; then
	echo "ERROR: lock_test_profile.jsonl is empty"
	exit 1
	fi

	echo "=== Lock Test Profile Contents ==="
	cat lock_test_profile.jsonl
	echo ""

	python3 << 'VALIDATION_SCRIPT'
	import sys
	import json

	experiments = []
	current_exp = None

	with open('lock_test_profile.jsonl', 'r') as f:
	for line in f:
	line = line.strip()
	if not line:
	continue
	obj = json.loads(line)
	if obj.get('type') == 'experiment':
	current_exp = obj
	elif obj.get('type') == 'throughput-point' and current_exp:
	current_exp['delta'] = float(obj['delta'])
	experiments.append(current_exp)
	current_exp = None

	# Check for experiments on critical_work (line 12) and local_work (line 20)
	line12_exps = [e for e in experiments if 'lock_test.cpp:12' in e['selected']]
	line20_exps = [e for e in experiments if 'lock_test.cpp:20' in e['selected']]

	print(f"Experiments for line 12 (critical_work): {len(line12_exps)}")
	print(f"Experiments for line 20 (local_work): {len(line20_exps)}")

	if not line12_exps:
	print("ERROR: No experiments found for lock_test.cpp:12 (critical section)")
	sys.exit(1)

	# Check samples to verify critical_work dominates
	with open('lock_test_profile.jsonl', 'r') as f:
	for line in f:
	obj = json.loads(line.strip())
	if obj.get('type') == 'samples':
	loc = obj['location']
	count = obj['count']
	print(f" Samples: {loc} = {count}")

	print("\nLock contention benchmark validation: PASSED")
	print(" - lock_test.cpp:12 (critical section) identified as profiling target")
	VALIDATION_SCRIPT

	- name: Upload profile artifacts
	uses: actions/upload-artifact@v4
	with:
	name: coz-profiles
	path: \|
	build/profile.jsonl
	build/lock_test_profile.jsonl
	retention-days: 7

	build-macos:
	runs-on: macos-latest

	steps:
	- uses: actions/checkout@v4

	- name: Install dependencies
	run: \|
	# cmake is preinstalled on the runner; make sure pkg-config and sqlite
	# (needed to configure the benchmark tree) are available. coreutils
	# provides gtimeout, since macOS has no GNU `timeout`.
	brew install pkg-config sqlite coreutils \|\| true

	- name: Build coz
	run: \|
	mkdir build
	cd build
	cmake .. -DCMAKE_BUILD_TYPE=RelWithDebInfo
	make -j$(sysctl -n hw.ncpu)

	- name: Build benchmarks
	run: \|
	cd build
	cmake .. -DBUILD_BENCHMARKS=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo
	make -j$(sysctl -n hw.ncpu) toy lock_test kmeans

	- name: Verify build artifacts
	run: \|
	# On macOS the injectable library keeps the .so suffix (see
	# libcoz/CMakeLists.txt) so it can be used with DYLD_INSERT_LIBRARIES.
	test -f build/libcoz/libcoz.so
	test -x build/benchmarks/toy/toy
	test -x build/benchmarks/lock_test/lock_test
	test -x build/benchmarks/kmeans/kmeans
	echo "Build artifacts verified"

	- name: Run unit tests
	run: \|
	cd build
	# Includes test_run_kmeans, which profiles kmeans live via kperf and so
	# also validates that macOS sampling actually collects samples.
	ctest --output-on-failure -j$(sysctl -n hw.ncpu)
	echo "Unit tests passed"

	- name: Run toy benchmark with coz
	run: \|
	cd build
	gtimeout 30 ../coz run -o profile.jsonl --- ./benchmarks/toy/toy \|\| true
	if [ ! -s profile.jsonl ]; then
	echo "ERROR: profile.jsonl was not created"
	exit 1
	fi
	echo "Profile created:"
	cat profile.jsonl

	lang-rust:
	name: Rust bindings (${{ matrix.os }})
	runs-on: ${{ matrix.os }}
	strategy:
	fail-fast: false
	matrix:
	os: [ubuntu-latest, macos-latest]

	steps:
	- uses: actions/checkout@v4

	- name: Install dependencies (Linux)
	if: runner.os == 'Linux'
	run: \|
	sudo apt-get update
	sudo apt-get install -y build-essential cmake pkg-config
	echo 1 \| sudo tee /proc/sys/kernel/perf_event_paranoid

	- name: Install dependencies (macOS)
	if: runner.os == 'macOS'
	run: brew install pkg-config coreutils \|\| true

	- name: Build coz
	run: \|
	cmake -S . -B build -DCMAKE_BUILD_TYPE=RelWithDebInfo
	cmake --build build -j3
	test -f build/libcoz/libcoz.so \|\| test -f build/libcoz/libcoz.dylib

	- name: Check the crate on both platforms
	working-directory: rust
	run: \|
	cargo build --release --examples
	cargo test --release

	- name: Profile the Rust toy under coz
	run: \|
	# `timeout` is GNU-only; coreutils provides gtimeout on macOS.
	TIMEOUT=timeout
	command -v gtimeout >/dev/null && TIMEOUT=gtimeout
	# rust/examples/toy runs ~9e6 work units of 10k iterations each across
	# two threads (~1-2 minutes on a 2-core runner); the timeout is a
	# safety net since experiments are written to the profile as they run.
	$TIMEOUT 300 ./coz run -o profile.jsonl --- ./rust/target/release/examples/toy \|\| true
	test -s profile.jsonl

	- name: Validate the Rust profile
	run: python3 .github/scripts/check_profile.py profile.jsonl toy.rs

	lang-go:
	name: Go bindings (${{ matrix.os }})
	runs-on: ${{ matrix.os }}
	strategy:
	fail-fast: false
	matrix:
	os: [ubuntu-latest, macos-latest]

	steps:
	- uses: actions/checkout@v4

	- uses: actions/setup-go@v5
	with:
	go-version: '1.26'

	- name: Install dependencies (Linux)
	if: runner.os == 'Linux'
	run: \|
	sudo apt-get update
	sudo apt-get install -y build-essential cmake pkg-config
	echo 1 \| sudo tee /proc/sys/kernel/perf_event_paranoid

	- name: Install dependencies (macOS)
	if: runner.os == 'macOS'
	run: brew install pkg-config coreutils \|\| true

	- name: Build coz
	run: \|
	cmake -S . -B build -DCMAKE_BUILD_TYPE=RelWithDebInfo
	cmake --build build -j3
	test -f build/libcoz/libcoz.so \|\| test -f build/libcoz/libcoz.dylib

	- name: Vet, format and test
	working-directory: go
	run: \|
	test -z "$(gofmt -l .)" \|\| { echo "gofmt found unformatted files:"; gofmt -l .; exit 1; }
	go vet ./...
	go test -race ./...
	# The package must also build with cgo off, degrading to no-ops.
	CGO_ENABLED=0 go build ./...

	- name: Build the Go toy
	working-directory: go
	run: \|
	# coz cannot read Go's compressed DWARF (.zdebug_* / __zdebug_*).
	go build -ldflags=-compressdwarf=false -o "$RUNNER_TEMP/gotoy" ./example

	- name: Profile the Go toy under coz
	run: \|
	TIMEOUT=timeout
	command -v gtimeout >/dev/null && TIMEOUT=gtimeout
	$TIMEOUT 180 ./coz run -o profile.jsonl --- "$RUNNER_TEMP/gotoy" \|\| true
	test -s profile.jsonl

	- name: Validate the Go profile
	run: python3 .github/scripts/check_profile.py profile.jsonl toy.go

	lang-java:
	name: Java agent (${{ matrix.os }})
	runs-on: ${{ matrix.os }}
	strategy:
	fail-fast: false
	matrix:
	os: [ubuntu-latest, macos-latest]

	steps:
	- uses: actions/checkout@v4

	# Temurin, not GraalVM: the agent samples at safepoints, and the Graal JIT
	# leaves hot counted loops without a safepoint poll, pushing the mean
	# time-to-safepoint from ~100us to ~100ms. See java/README.md.
	- uses: actions/setup-java@v4
	with:
	distribution: temurin
	java-version: '21'

	- name: Install dependencies (Linux)
	if: runner.os == 'Linux'
	run: \|
	sudo apt-get update
	sudo apt-get install -y build-essential cmake pkg-config

	- name: Install dependencies (macOS)
	if: runner.os == 'macOS'
	run: brew install pkg-config coreutils \|\| true

	- name: Build the agent and coz.jar
	run: \|
	cmake -S . -B build -DCMAKE_BUILD_TYPE=RelWithDebInfo -DBUILD_JAVA_AGENT=ON
	cmake --build build --target cozjava coz-jar -j3
	test -f build/java/libcozjava.so
	test -f build/java/coz.jar

	- name: Build the Java toy
	working-directory: java
	run: \|
	mkdir -p "$RUNNER_TEMP/classes"
	# -g keeps the line tables the agent reads through GetLineNumberTable.
	javac -g -cp ../build/java/coz.jar -d "$RUNNER_TEMP/classes" example/Toy.java

	- name: Profile the Java toy under the agent
	run: \|
	TIMEOUT=timeout
	command -v gtimeout >/dev/null && TIMEOUT=gtimeout
	$TIMEOUT 240 java \
	-agentpath:"$PWD/build/java/libcozjava.so=output=$PWD/profile.jsonl,scope=example,verbose=1" \
	-cp "$RUNNER_TEMP/classes:$PWD/build/java/coz.jar" example.Toy \|\| true
	test -s profile.jsonl

	- name: Validate the Java profile
	run: python3 .github/scripts/check_profile.py profile.jsonl Toy.java

Navigation Menu

Uh oh!

Uh oh!

Add a JVMTI agent for causal profiling of Java #129

Workflow file

Add a JVMTI agent for causal profiling of Java #129

Uh oh!

Workflow file for this run