ohos-htrace 0.1.0

OpenHarmony htrace file parser & cold-start analyzer

ohos-htrace

Pure-Python parser & cold-start analyzer for OpenHarmony .htrace trace files.

No external binaries needed — replaces the C++ trace_streamer tool with a native Python implementation that parses .htrace protobuf data directly into an in-memory SQLite database.

---

✨ Features

• 🚀 Pure Python — no streamer.exe dependency, works on any platform
• ⚡ Fast — parses 120MB htrace in ~3 seconds
• 🎯 Cold-start analysis — automatic detection of app launch timing
• 📊 Keyword breakdown — measure time spent in specific functions
• 🔧 CLI & API — use from command line or import in your Python code
• 📦 pip installable — pip install ohos-htrace

📦 Installation

pip install ohos-htrace

Or install from source:

git clone https://github.com/peng-yg/ohos-htrace.git
cd ohos-htrace
pip install -e .

🚀 Quick Start

Command Line

# Basic cold-start analysis
ohos-htrace trace.htrace com.example.app

# With keyword duration breakdown
ohos-htrace trace.htrace aweme -k AppSpawn,LaunchAbility,LoadModule

# JSON output (for scripting)
ohos-htrace trace.htrace aweme --json

# Verbose output with top-10 functions
ohos-htrace trace.htrace aweme -k AppSpawn,LaunchAbility -v

Example output:

Parsing trace.htrace (118.7 MB) ...
Done in 3.6s

  Process:       ss.hm.ugc.aweme (pid=4575)
  Cold Start:    2566.08 ms
  Callstack:     69,933 nodes
  Depth-0:       16,916 blocks
  Max Gap:       27.9 ms
  Frames:        0

  Keyword Durations:
    AppSpawn: 19.12ms (5 hits)
    LaunchAbility: 3175.94ms (2 hits)
    LoadModule: 1164.25ms (3 hits)

Python API

import ohos_htrace

# One-shot analysis
result = ohos_htrace.parse_and_analyze(
    "trace.htrace",
    "aweme",
    keywords=["AppSpawn", "LaunchAbility", "LoadModule"],
)

print(f"Cold start: {result['cold_start_ms']:.2f}ms")
print(f"Max gap: {result['max_gap_ms']:.1f}ms")

for kw, info in result["keyword_durations"]["by_keyword"].items():
    print(f"  {kw}: {info['total_ms']:.2f}ms ({info['count']} hits)")

Low-Level API

import ohos_htrace

# Step 1: Parse htrace → in-memory SQLite
conn = ohos_htrace.parse_to_memory_db("trace.htrace")

# Step 2: Query like a normal database
cur = conn.cursor()
cur.execute("SELECT name, pid FROM process")
for name, pid in cur.fetchall():
    print(f"  {name} (pid={pid})")

# Step 3: Use analysis functions
procs = ohos_htrace.find_matching_processes(conn, "aweme")
for proc in procs:
    tree = ohos_htrace.get_process_callstack(conn, proc["id"])
    cold_ms, gaps = ohos_htrace.compute_cold_start_time(tree)
    print(f"  {proc['name']}: {cold_ms:.2f}ms")

conn.close()

📐 Database Schema

The in-memory database mirrors the C++ trace_streamer output:

Table	Columns	Description
process	id, name, pid	Process list
thread	id, ipid, name, tid	Thread list (ipid → process.id)
callstack	id, callid, parent_id, name, depth, ts, dur	Function call stack (callid → thread.id)
frame_slice	id, vsync, ts, dur, type_desc, flag, ipid	Frame rendering data

🔍 How It Works

.htrace file
    │
    ├─ 1024-byte Header (magic: OHOSPROF)
    │
    └─ Segments: [4-byte length][protobuf payload]
         │
         └─ ProfilerPluginData
              │
              └─ ftrace-plugin → TracePluginResult
                   │
                   ├─ ftrace_cpu_detail[].event[]
                   │    ├─ print_format (B|E|S|F events → callstack)
                   │    ├─ task_rename_format (→ process names)
                   │    └─ sched_switch_format (→ thread names)
                   │
                   └─ comm_dict[] (→ thread name resolution)

Key implementation details:

• tgid extraction from buf — Uses the tgid from B|<tgid>|<name> (not event.tgid), matching C++ streamer behavior. This is critical for correct process attribution after fork().
• Event sorting — Events from different CPU cores are sorted by (tgid, pid, ts) before B/E stack matching, preventing cross-core timing artifacts.
• Async events — Supports S|F (async start/finish) events with cookie-based pairing.

🆚 Comparison with trace_streamer (C++)

Tested on a 118.7MB htrace file:

Metric	trace_streamer	ohos-htrace	Match
Cold Start	2566.08 ms	2566.08 ms	✅
Max Gap	27.9 ms	27.9 ms	✅
AppSpawn	19.12ms (5)	19.12ms (5)	✅
LaunchAbility	3175.94ms (2)	3175.94ms (2)	✅
JsAbilityStage	1575.63ms (5)	1575.63ms (5)	✅
LoadModule	1164.25ms (3)	1164.25ms (3)	✅
Parse Time	4.6s	3.6s	⚡ faster
Callstack	71,114	69,933	~98% (binder events)

Remaining differences: binder transaction events (~1,181 entries) and frame_slice data (from hidump-plugin, not yet parsed). These do not affect cold-start analysis accuracy.

📋 Requirements

• Python ≥ 3.10
• protobuf ≥ 4.21.0

📄 License

Apache License 2.0