config-disassembler 0.11.9

Disassemble config files into smaller files and reassemble on demand.
Documentation 
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

"""Summarize a dhat-heap.json written by `replay_workload --features dhat-heap`.

Prints which crate call sites hold the most memory at the global heap peak
(dhat's "gb" = bytes live at t-gmax), so a memory fix can target what is
actually live rather than what allocates the most in total.

Each allocation is attributed to the first crate frame in its stack (the crate
code that asked for the memory). A frame is the crate's when its symbol starts
with `config_disassembler::` (Windows / full symbols) or its source path is one
of the crate's own files (Linux with line-table debug info reports bare names,
e.g. `flush_text_buffer (xml/parsers/parse_xml_cdata.rs:91)`). Allocations with
no crate frame (tokio, std, the replay harness) are grouped under their first
non-allocator frame.

Usage: python3 examples/dhat_summary.py dhat-heap.json [top_n]
"""

import json
import re
import sys
from collections import defaultdict

# Allocator / std / dependency frames: skipped when looking for "who asked".
# Windows reports qualified names; Linux reports bare names with source paths,
# where dependencies carry a version (`indexmap-2.14.0/src/...`) and std lives
# under `alloc/`, `core/`, `std/` or bare `src/<std module>/`.
ALLOCATOR_FRAME = re.compile(
    r"dhat::|alloc::alloc|alloc::raw_vec|__rust_|core::alloc|<alloc::|std::alloc|hashbrown::raw|"
    r"alloc::vec::|alloc::string::|alloc::boxed::|alloc::collections::|indexmap::|serde_json::|"
    r"core::ptr::|core::iter::|core::ops::|"
    r"\(dhat-|\(alloc/|\(core/|\(std/|\(src/(?:raw_vec|vec|string|alloc|collections|boxed|slice|str)/|"
    r"\([A-Za-z_]+-\d+\.\d+"
)

# The crate's own source files as they appear in frame paths (`src/xml/...` or
# `xml/...`, plus the top-level modules).
CRATE_PATH = re.compile(
    r"\((?:.*/)?(?:src/)?"
    r"(?:xml/|disassemble\.rs|reassemble\.rs|format\.rs|meta\.rs|cli\.rs|xml_cmd\.rs|ignore_file\.rs|error\.rs)"
)
DEPENDENCY_PATH = re.compile(r"\([A-Za-z_]+-\d+\.\d+")


def frame_name(raw: str) -> str:
    # "0x7ff6...: some::function (file.rs:12:5)" -> "some::function (file.rs:12)"
    name = raw.split(": ", 1)[-1]
    return re.sub(r":\d+\)$", ")", name)


def is_crate_frame(name: str) -> bool:
    if name.startswith("config_disassembler::"):
        return True
    if DEPENDENCY_PATH.search(name):
        return False
    return bool(CRATE_PATH.search(name))


def main() -> None:
    path = sys.argv[1]
    top_n = int(sys.argv[2]) if len(sys.argv) > 2 else 15
    data = json.load(open(path))
    frames = data["ftbl"]

    total_at_peak = sum(pp["gb"] for pp in data["pps"])
    by_site = defaultdict(lambda: {"bytes": 0, "blocks": 0, "inner": defaultdict(int)})
    for pp in data["pps"]:
        if pp["gb"] == 0:
            continue
        names = [frame_name(frames[i]) for i in pp["fs"]]
        site = next((n for n in names if is_crate_frame(n)), None)
        if site is None:
            site = next((n for n in names if not ALLOCATOR_FRAME.search(n)), names[-1] if names else "?")
        # The frame just inside the crate frame says WHAT was allocated
        # (e.g. serde_json map insert, String clone, Vec growth).
        idx = names.index(site)
        inner = names[idx - 1] if idx > 0 else "(direct)"
        entry = by_site[site]
        entry["bytes"] += pp["gb"]
        entry["blocks"] += pp["gbk"]
        entry["inner"][inner] += pp["gb"]

    mb = lambda b: b / (1024 * 1024)
    print(f"heap at global peak: {mb(total_at_peak):.1f} MB across {len(by_site)} call sites\n")
    print("| MB at peak | share | blocks | crate call site | largest inner allocation |")
    print("|---:|---:|---:|---|---|")
    ranked = sorted(by_site.items(), key=lambda kv: kv[1]["bytes"], reverse=True)[:top_n]
    for site, entry in ranked:
        inner, _ = max(entry["inner"].items(), key=lambda kv: kv[1])
        share = entry["bytes"] / total_at_peak * 100 if total_at_peak else 0
        print(
            f"| {mb(entry['bytes']):.2f} | {share:.1f}% | {entry['blocks']} | "
            f"`{site[:110]}` | `{inner[:90]}` |"
        )


if __name__ == "__main__":
    main()