Datasets:
File size: 6,492 Bytes
e1ced61 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 | """Builders for stage, resource-snapshot, and dataset manifests.
Every stage output carries ``input_manifest_sha256``, ``code_commit``,
``config_sha256``, and ``created_at`` so that an artifact can be traced to the
exact code, config, and upstream inputs that produced it (``docs/02_DATA_PIPELINE.md``
section 1). Image paths are stored relative to the manifest so outputs reproduce
across mounts.
"""
from __future__ import annotations
from collections.abc import Iterable
from dataclasses import dataclass
from pathlib import Path
from typing import Any
from .atomic_io import atomic_write_json, count_jsonl, read_jsonl
from .hashing import join_manifest_sha256, sha256_file, sha256_text
# Reserved provenance keys present on every stage manifest.
STAGE_PROVENANCE_KEYS = (
"input_manifest_sha256",
"code_commit",
"config_sha256",
"created_at",
)
def stage_manifest(
*,
stage: str,
input_manifest_sha256: str | None,
code_commit: str,
config_sha256: str,
created_at: str,
renderer_version: str | None = None,
answer_normalizer_version: str | None = None,
extra: dict[str, Any] | None = None,
) -> dict[str, Any]:
"""Build a stage manifest record with required provenance fields.
``created_at`` must be supplied by the caller (RFC 3339, UTC preferred) so
that the pipeline stays deterministic and never relies on wall-clock time
inside a reproducible stage.
"""
record: dict[str, Any] = {
"stage": stage,
"input_manifest_sha256": input_manifest_sha256,
"code_commit": code_commit,
"config_sha256": config_sha256,
"created_at": created_at,
}
if renderer_version is not None:
record["renderer_version"] = renderer_version
if answer_normalizer_version is not None:
record["answer_normalizer_version"] = answer_normalizer_version
if extra:
for key, value in extra.items():
if key in STAGE_PROVENANCE_KEYS:
raise ValueError(f"extra manifest key collides with reserved key {key!r}")
record[key] = value
return record
def input_manifest_hash_from_paths(paths: Iterable[str | Path]) -> str:
"""Hash an ordered set of input artifact files by their per-file SHA-256."""
return join_manifest_sha256(sha256_file(path) for path in paths)
def input_manifest_hash_from_jsonl(path: str | Path) -> str:
"""Fold every row's ``sha256``/``base_id`` field of a JSONL file into one hash."""
digests: list[str] = []
for row in read_jsonl(path):
digest = row.get("sha256") or row.get("base_id") or row.get("group_id")
if digest is None:
raise ValueError(f"row in {path} has no sha256/base_id/group_id field")
digests.append(str(digest))
return join_manifest_sha256(digests)
@dataclass(frozen=True)
class FileEntry:
relative_path: str
size: int
sha256: str
def snapshot_file_tree(root: str | Path) -> list[FileEntry]:
"""Return ``{relative_path, size, sha256}`` for every regular file under ``root``.
Symlinks are resolved to their target's content. Entries are sorted by
relative path so the snapshot is deterministic.
"""
root_path = Path(root)
entries: list[FileEntry] = []
if not root_path.exists():
return entries
for path in sorted(root_path.rglob("*")):
if not path.is_file():
continue
if path.is_symlink():
target = path.resolve()
if not target.is_file():
continue
relative = path.relative_to(root_path).as_posix()
entries.append(
FileEntry(
relative_path=relative,
size=path.stat().st_size,
sha256=sha256_file(path),
)
)
return entries
def resource_snapshot_row(
*,
logical_name: str,
repo_id: str,
requested_revision: str,
resolved_commit: str,
local_path: str,
downloaded_at: str,
file_count: int,
total_bytes: int,
manifest_sha256: str,
kind: str = "model",
) -> dict[str, Any]:
"""Build one ``resource_snapshot.jsonl`` row (``docs/01_RESOURCE_CATALOG.md`` §7)."""
if resolved_commit != requested_revision:
raise ValueError(
f"resolved commit {resolved_commit} != requested {requested_revision} "
f"for {logical_name}; this is a preflight failure"
)
return {
"logical_name": logical_name,
"kind": kind,
"repo_id": repo_id,
"requested_revision": requested_revision,
"resolved_commit": resolved_commit,
"local_path": local_path,
"downloaded_at": downloaded_at,
"file_count": file_count,
"total_bytes": total_bytes,
"manifest_sha256": manifest_sha256,
}
def file_tree_manifest_sha256(entries: Iterable[FileEntry]) -> str:
"""SHA-256 of the canonical ``relative_path\0size\0sha256`` list (sorted)."""
payload = "".join(
f"{e.relative_path}\0{e.size}\0{e.sha256}\n"
for e in sorted(entries, key=lambda e: e.relative_path)
)
return sha256_text(payload)
def artifact_manifest_sha256(path: str | Path) -> str:
"""Hash one file by bytes or one directory by its deterministic file tree."""
source = Path(path)
if source.is_file():
return sha256_file(source)
if source.is_dir():
return file_tree_manifest_sha256(snapshot_file_tree(source))
raise FileNotFoundError(f"artifact does not exist: {source}")
def build_dataset_manifest(
*,
dataset_name: str,
items_path: str | Path,
extra: dict[str, Any] | None = None,
) -> dict[str, Any]:
"""Build a ``<dataset_name>.manifest.json`` summarizing a normalized JSONL file."""
items_path = Path(items_path)
record: dict[str, Any] = {
"dataset_name": dataset_name,
"items_path": items_path.name,
"row_count": count_jsonl(items_path),
"items_sha256": sha256_file(items_path),
}
if extra:
record.update(extra)
return record
def write_dataset_manifest(
*,
output_path: str | Path,
dataset_name: str,
items_path: str | Path,
extra: dict[str, Any] | None = None,
) -> dict[str, Any]:
"""Build and atomically write a dataset manifest, returning the record."""
record = build_dataset_manifest(dataset_name=dataset_name, items_path=items_path, extra=extra)
atomic_write_json(output_path, record)
return record
|