| 227 | |
| 228 | |
| 229 | def main() -> int: |
| 230 | ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter) |
| 231 | ap.add_argument("dump") |
| 232 | ap.add_argument("--binary", help="the binary that produced the dump, for symbolisation") |
| 233 | ap.add_argument("--top", type=int, default=20) |
| 234 | ap.add_argument("--sort", choices=sorted(SITE_SORTS), default="bytes") |
| 235 | ap.add_argument("--json", action="store_true", help="emit the merged tables as JSON") |
| 236 | args = ap.parse_args() |
| 237 | |
| 238 | procs = load(args.dump) |
| 239 | if not procs: |
| 240 | print("no waste-sanitizer dump found in", args.dump, file=sys.stderr) |
| 241 | return 2 |
| 242 | |
| 243 | names_by_pid = {} |
| 244 | for pid, p in procs.items(): |
| 245 | addrs = list(p["sites"].keys()) + [a for (_, a) in p["work"].keys()] |
| 246 | names_by_pid[pid] = symbolise(args.binary, p["header"].get("image_base", "0x0"), addrs) |
| 247 | |
| 248 | if args.json: |
| 249 | out = {"processes": {}} |
| 250 | for pid, p in procs.items(): |
| 251 | names = names_by_pid[pid] |
| 252 | out["processes"][str(pid)] = { |
| 253 | "header": p["header"], |
| 254 | "phases": [h.get("phase_label") for h in p["headers"]], |
| 255 | "summary": summarise(p, names), |
| 256 | "sites": [dict(r, name=names.get(a, a), |
| 257 | **{(k if k.endswith("_phase") else k + "_max"): v |
| 258 | for k, v in p["snap_max"].get(a, {}).items()}) |
| 259 | for a, r in p["sites"].items()], |
| 260 | "work": [dict(w, name=names.get(a, a)) for (_, a), w in p["work"].items()], |
| 261 | } |
| 262 | json.dump(out, sys.stdout, indent=1) |
| 263 | print() |
| 264 | return 0 |
| 265 | |
| 266 | # Work-heavy processes first: the index worker is the one that matters. |
| 267 | order = sorted(procs.items(), key=lambda kv: -kv[1]["header"].get("usable_bytes", 0)) |
| 268 | for pid, p in order: |
| 269 | h = p["header"] |
| 270 | names = names_by_pid[pid] |
| 271 | s = summarise(p, names) |
| 272 | print("=" * 110) |
| 273 | print(f"process {pid} last dump={h.get('why')} phases={len(p['headers'])} allocs={h['allocs']:,} " |
| 274 | f"cum={h['usable_bytes'] / MB:,.0f} MB live={h['live_bytes'] / MB:,.0f} MB " |
| 275 | f"rss={h.get('rss_bytes', 0) / MB:,.0f} MB sites={h['sites']:,} fill={h.get('fill')} access={h.get('access')} " |
| 276 | f"layer={h.get('layer_vm_bytes', 0) / MB:,.0f} MB/{h.get('layer_threads', 0)} threads") |
| 277 | blind = {k: h.get(k, 0) for k in ("site_table_full", "pointer_table_full", "work_table_full", "untracked_frees")} |
| 278 | if any(blind.values()): |
| 279 | print(" blind spots: " + " ".join(f"{k}={v:,}" for k, v in blind.items())) |
| 280 | if h.get("foreign_frees", 0): |
| 281 | print(f" outside the shim: {h['foreign_frees']:,} frees of blocks another allocator made " |
| 282 | "(C runtime / system DLL; their allocations are invisible to the layer)") |
| 283 | print() |
| 284 | print(" WASTE SUMMARY") |
| 285 | for key, label, unit in WASTE_CLASSES: |
| 286 | v = s.get(key, 0) |