File size: 4,241 Bytes
5a5d5a5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: OpenMDW-1.1

import atexit
import os
import sys
from typing import Any

import torch.distributed as dist
from loguru._logger import Core, Logger

RANK0_ONLY = True
LEVEL = os.environ.get("LOGURU_LEVEL", "INFO")
RANK = int(os.environ.get("RANK", "0"))


def make_new_logger(depth: int = 1) -> Logger:
    return Logger(
        core=Core(),
        exception=None,
        depth=depth,
        record=False,
        lazy=False,
        colors=False,
        raw=False,
        capture=True,
        patchers=[],
        extra={},
    )


logger = make_new_logger(depth=1)
atexit.register(logger.remove)


def _add_relative_path(record: dict[str, Any]) -> None:
    try:
        start = os.getcwd()
        record["extra"]["relative_path"] = os.path.relpath(record["file"].path, start)
    except OSError:
        # CWD may have been removed (e.g. on some ranks in distributed jobs).
        # Fall back to the absolute path so logging still works.
        record["extra"]["relative_path"] = f"<cwd-unavailable>:{record['file'].path}"


*options, _, extra = logger._options  # type: ignore
logger._options = tuple([*options, [_add_relative_path], extra])  # type: ignore


def init_loguru_stdout() -> None:
    logger.remove()
    datetime_format = get_datetime_format()
    machine_format = get_machine_format()
    message_format = get_message_format()
    logger.add(
        sys.stdout,
        level=LEVEL,
        format=f"{datetime_format}{machine_format}{message_format}",
        filter=_rank0_only_filter,
    )


def init_loguru_file(path: str) -> None:
    datetime_format = get_datetime_format()
    machine_format = get_machine_format()
    message_format = get_message_format()
    logger.add(
        path,
        encoding="utf8",
        level=LEVEL,
        format=f"{datetime_format}{machine_format}{message_format}",
        rotation="100 MB",
        filter=lambda result: _rank0_only_filter(result) or not RANK0_ONLY,
        enqueue=True,
    )


def get_datetime_format() -> str:
    return "[<green>{time:MM-DD HH:mm:ss}</green>|"


def get_machine_format() -> str:
    node_id = os.environ.get("NGC_ARRAY_INDEX", "0")
    num_nodes = int(os.environ.get("NGC_ARRAY_SIZE", "1"))
    machine_format = ""
    rank = 0
    if dist.is_available():
        if not RANK0_ONLY and dist.is_initialized():
            rank = dist.get_rank()
            world_size = dist.get_world_size()
            machine_format = (
                f"<red>[Node{node_id:<3}/{num_nodes:<3}][RANK{rank:<5}/{world_size:<5}]" + "[{process.name:<8}]</red>| "
            )
    return machine_format


def get_message_format() -> str:
    message_format = "<level>{level}</level>|<cyan>{extra[relative_path]}:{line}:{function}</cyan>] {message}"
    return message_format


def _rank0_only_filter(record: Any) -> bool:
    is_rank0 = record["extra"].get("rank0_only", True)
    if RANK == 0 and is_rank0:
        return True
    if not is_rank0:
        record["message"] = f"[RANK {RANK}] " + record["message"]
    return not is_rank0


def trace(message: str, rank0_only: bool = True) -> None:
    logger.opt(depth=1).bind(rank0_only=rank0_only).trace(message)


def debug(message: str, rank0_only: bool = True) -> None:
    logger.opt(depth=1).bind(rank0_only=rank0_only).debug(message)


def info(message: str, rank0_only: bool = True) -> None:
    logger.opt(depth=1).bind(rank0_only=rank0_only).info(message)


def success(message: str, rank0_only: bool = True) -> None:
    logger.opt(depth=1).bind(rank0_only=rank0_only).success(message)


def warning(message: str, rank0_only: bool = True) -> None:
    logger.opt(depth=1).bind(rank0_only=rank0_only).warning(message)


def error(message: str, rank0_only: bool = True) -> None:
    logger.opt(depth=1).bind(rank0_only=rank0_only).error(message)


def critical(message: str, rank0_only: bool = True) -> None:
    logger.opt(depth=1).bind(rank0_only=rank0_only).critical(message)


def exception(message: str, rank0_only: bool = True) -> None:
    logger.opt(depth=1).bind(rank0_only=rank0_only).exception(message)


# Execute at import time.
init_loguru_stdout()