release: issue #23 tcpretrans fix + low-latency perf (PR #21)

.github/workflows/ci.yml

@@ -29,3 +29,26 @@ jobs:
 
       - name: Vet
         run: go vet ./...
+
+  bpf-smoke:
+    name: BPF object smoke build
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Go
+        uses: actions/setup-go@v5
+        with:
+          go-version: "1.24"
+
+      - name: Install BPF toolchain
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y clang llvm linux-tools-generic libbpf-dev
+          sudo ln -sf $(ls /usr/lib/linux-tools/*/bpftool | head -n1) /usr/local/bin/bpftool
+
+      - name: Generate embedded BPF objects
+        run: make generate
+
+      - name: Build with embedded objects
+        run: go build -v ./...

.github/workflows/release.yml

@@ -34,6 +34,17 @@ jobs:
               with:
                   go-version: "1.24"
 
+            - name: Install BPF toolchain
+              run: |
+                  sudo apt-get update
+                  sudo apt-get install -y clang llvm linux-tools-generic libbpf-dev
+                  # linux-tools-generic installs bpftool under a kernel-versioned path;
+                  # expose it on PATH for `make generate`.
+                  sudo ln -sf $(ls /usr/lib/linux-tools/*/bpftool | head -n1) /usr/local/bin/bpftool
+
+            - name: Generate embedded BPF objects
+              run: make generate
+
             - name: Run GoReleaser
               uses: goreleaser/goreleaser-action@v6
               with:

.gitignore

@@ -12,6 +12,10 @@ coverage.out
 dist/
 demo.tape
 
+# Generated BPF artifacts (produced by `make generate`)
+internal/ebpf/bpf/*.o
+internal/ebpf/c/vmlinux.h
+
 # IDE
 .idea/
 .vscode/

Makefile

@@ -6,13 +6,28 @@ all: build
 build:
 	go build -o melisai cmd/melisai/main.go
 
-# Compile BPF programs
-# Requires clang and llvm
-# Ensure internal/ebpf/bpf directory exists
+# Compile BPF programs into internal/ebpf/bpf/*.o for embedding via //go:embed.
+# Requires (on the host running this target):
+#   - clang, llvm        (apt: clang llvm)
+#   - bpftool            (apt: linux-tools-generic, then ln -s to /usr/local/bin)
+#   - libbpf headers     (apt: libbpf-dev)
+#   - kernel BTF         (/sys/kernel/btf/vmlinux — kernel built with CONFIG_DEBUG_INFO_BTF=y)
+#
+# The generated vmlinux.h is host-kernel specific but the resulting .o is CO-RE
+# (BPF Type Format-relocatable) and will run on any kernel with BTF support.
+# macOS/dev machines without these tools should skip `make generate` — the
+# build will then rely on the on-disk fallback in loader.go.
 generate:
+	@command -v clang >/dev/null || { echo "ERROR: clang not found (install clang/llvm)"; exit 1; }
+	@command -v bpftool >/dev/null || { echo "ERROR: bpftool not found (install linux-tools-generic)"; exit 1; }
+	@test -f /sys/kernel/btf/vmlinux || { echo "ERROR: /sys/kernel/btf/vmlinux not present (kernel needs CONFIG_DEBUG_INFO_BTF=y)"; exit 1; }
 	mkdir -p internal/ebpf/bpf
-	clang -g -O2 -target bpf -D__TARGET_ARCH_x86 -I/usr/include/x86_64-linux-gnu \
-		-c internal/ebpf/c/tcpretrans.bpf.c -o internal/ebpf/bpf/tcpretrans.o
+	bpftool btf dump file /sys/kernel/btf/vmlinux format c > internal/ebpf/c/vmlinux.h
+	clang -g -O2 -target bpf -D__TARGET_ARCH_x86 \
+		-I/usr/include/x86_64-linux-gnu \
+		-I internal/ebpf/c \
+		-c internal/ebpf/c/tcpretrans.bpf.c \
+		-o internal/ebpf/bpf/tcpretrans.o
 
 test:
 	go test -race -count=1 -timeout 120s ./...
@@ -33,3 +48,4 @@ test-validation:
 clean:
 	rm -f melisai
 	rm -f internal/ebpf/bpf/*.o
+	rm -f internal/ebpf/c/vmlinux.h

cmd/melisai/main.go

@@ -9,6 +9,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"os"
+	"runtime/pprof"
 	"strings"
 	"time"
 
@@ -82,6 +83,7 @@ Schema v1.1.0 with structured sub-objects for network, memory, GPU data.`,
 		collectMaxEvents int
 		collectQuiet     bool
 		collectVerbose   bool
+		collectPprof     string
 	)
 
 	collectCmd := &cobra.Command{
@@ -104,6 +106,19 @@ Examples:
   melisai collect --profile standard --focus network -o net.json
   melisai collect --profile deep --pid 12345 -o app.json`,
 		RunE: func(cmd *cobra.Command, args []string) error {
+			// CPU profiling for performance analysis of melisai itself.
+			if collectPprof != "" {
+				f, err := os.Create(collectPprof)
+				if err != nil {
+					return fmt.Errorf("create pprof file: %w", err)
+				}
+				defer f.Close()
+				if err := pprof.StartCPUProfile(f); err != nil {
+					return fmt.Errorf("start cpu profile: %w", err)
+				}
+				defer pprof.StopCPUProfile()
+			}
+
 			cfg := collector.DefaultConfig()
 			cfg.Profile = collectProfile
 			cfg.Version = version
@@ -169,6 +184,7 @@ Examples:
 	collectCmd.Flags().IntVar(&collectMaxEvents, "max-events", 1000, "Max events per collector")
 	collectCmd.Flags().BoolVarP(&collectQuiet, "quiet", "q", false, "Suppress progress output")
 	collectCmd.Flags().BoolVarP(&collectVerbose, "verbose", "v", false, "Enable debug logging")
+	collectCmd.Flags().StringVar(&collectPprof, "pprof", "", "Write CPU profile to file (e.g. melisai_cpu.prof)")
 
 	// --- install command ---
 	var installDryRun bool