File size: 21,475 Bytes
551c68e |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 |
import json
import os
import platform
import sqlite3
import time
from datetime import datetime
from pathlib import Path
from threading import Lock
try:
import fcntl
except ImportError: # fcntl is not available on Windows
fcntl = None
import huggingface_hub as hf
import pandas as pd
try: # absolute imports when installed from PyPI
from trackio.commit_scheduler import CommitScheduler
from trackio.dummy_commit_scheduler import DummyCommitScheduler
from trackio.utils import (
TRACKIO_DIR,
deserialize_values,
serialize_values,
)
except ImportError: # relative imports when installed from source on Spaces
from commit_scheduler import CommitScheduler
from dummy_commit_scheduler import DummyCommitScheduler
from utils import TRACKIO_DIR, deserialize_values, serialize_values
class ProcessLock:
"""A file-based lock that works across processes. Is a no-op on Windows."""
def __init__(self, lockfile_path: Path):
self.lockfile_path = lockfile_path
self.lockfile = None
self.is_windows = platform.system() == "Windows"
def __enter__(self):
"""Acquire the lock with retry logic."""
if self.is_windows:
return self
self.lockfile_path.parent.mkdir(parents=True, exist_ok=True)
self.lockfile = open(self.lockfile_path, "w")
max_retries = 100
for attempt in range(max_retries):
try:
fcntl.flock(self.lockfile.fileno(), fcntl.LOCK_EX | fcntl.LOCK_NB)
return self
except IOError:
if attempt < max_retries - 1:
time.sleep(0.1)
else:
raise IOError("Could not acquire database lock after 10 seconds")
def __exit__(self, exc_type, exc_val, exc_tb):
"""Release the lock."""
if self.is_windows:
return
if self.lockfile:
fcntl.flock(self.lockfile.fileno(), fcntl.LOCK_UN)
self.lockfile.close()
class SQLiteStorage:
_dataset_import_attempted = False
_current_scheduler: CommitScheduler | DummyCommitScheduler | None = None
_scheduler_lock = Lock()
@staticmethod
def _get_connection(db_path: Path) -> sqlite3.Connection:
conn = sqlite3.connect(str(db_path), timeout=30.0)
conn.execute("PRAGMA journal_mode = WAL")
conn.row_factory = sqlite3.Row
return conn
@staticmethod
def _get_process_lock(project: str) -> ProcessLock:
lockfile_path = TRACKIO_DIR / f"{project}.lock"
return ProcessLock(lockfile_path)
@staticmethod
def get_project_db_filename(project: str) -> Path:
"""Get the database filename for a specific project."""
safe_project_name = "".join(
c for c in project if c.isalnum() or c in ("-", "_")
).rstrip()
if not safe_project_name:
safe_project_name = "default"
return f"{safe_project_name}.db"
@staticmethod
def get_project_db_path(project: str) -> Path:
"""Get the database path for a specific project."""
filename = SQLiteStorage.get_project_db_filename(project)
return TRACKIO_DIR / filename
@staticmethod
def init_db(project: str) -> Path:
"""
Initialize the SQLite database with required tables.
If there is a dataset ID provided, copies from that dataset instead.
Returns the database path.
"""
db_path = SQLiteStorage.get_project_db_path(project)
db_path.parent.mkdir(parents=True, exist_ok=True)
with SQLiteStorage._get_process_lock(project):
with sqlite3.connect(db_path, timeout=30.0) as conn:
conn.execute("PRAGMA journal_mode = WAL")
cursor = conn.cursor()
cursor.execute("""
CREATE TABLE IF NOT EXISTS metrics (
id INTEGER PRIMARY KEY AUTOINCREMENT,
timestamp TEXT NOT NULL,
run_name TEXT NOT NULL,
step INTEGER NOT NULL,
metrics TEXT NOT NULL
)
""")
cursor.execute("""
CREATE TABLE IF NOT EXISTS configs (
id INTEGER PRIMARY KEY AUTOINCREMENT,
run_name TEXT NOT NULL,
config TEXT NOT NULL,
created_at TEXT NOT NULL,
UNIQUE(run_name)
)
""")
cursor.execute(
"""
CREATE INDEX IF NOT EXISTS idx_metrics_run_step
ON metrics(run_name, step)
"""
)
cursor.execute(
"""
CREATE INDEX IF NOT EXISTS idx_configs_run_name
ON configs(run_name)
"""
)
conn.commit()
return db_path
@staticmethod
def export_to_parquet():
"""
Exports all projects' DB files as Parquet under the same path but with extension ".parquet".
"""
# don't attempt to export (potentially wrong/blank) data before importing for the first time
if not SQLiteStorage._dataset_import_attempted:
return
all_paths = os.listdir(TRACKIO_DIR)
db_paths = [f for f in all_paths if f.endswith(".db")]
for db_path in db_paths:
db_path = TRACKIO_DIR / db_path
parquet_path = db_path.with_suffix(".parquet")
if (not parquet_path.exists()) or (
db_path.stat().st_mtime > parquet_path.stat().st_mtime
):
with sqlite3.connect(db_path) as conn:
df = pd.read_sql("SELECT * from metrics", conn)
# break out the single JSON metrics column into individual columns
metrics = df["metrics"].copy()
metrics = pd.DataFrame(
metrics.apply(
lambda x: deserialize_values(json.loads(x))
).values.tolist(),
index=df.index,
)
del df["metrics"]
for col in metrics.columns:
df[col] = metrics[col]
df.to_parquet(parquet_path)
@staticmethod
def import_from_parquet():
"""
Imports to all DB files that have matching files under the same path but with extension ".parquet".
"""
all_paths = os.listdir(TRACKIO_DIR)
parquet_paths = [f for f in all_paths if f.endswith(".parquet")]
for parquet_path in parquet_paths:
parquet_path = TRACKIO_DIR / parquet_path
db_path = parquet_path.with_suffix(".db")
df = pd.read_parquet(parquet_path)
with sqlite3.connect(db_path) as conn:
# fix up df to have a single JSON metrics column
if "metrics" not in df.columns:
# separate other columns from metrics
metrics = df.copy()
other_cols = ["id", "timestamp", "run_name", "step"]
df = df[other_cols]
for col in other_cols:
del metrics[col]
# combine them all into a single metrics col
metrics = json.loads(metrics.to_json(orient="records"))
df["metrics"] = [
json.dumps(serialize_values(row)) for row in metrics
]
df.to_sql("metrics", conn, if_exists="replace", index=False)
@staticmethod
def get_scheduler():
"""
Get the scheduler for the database based on the environment variables.
This applies to both local and Spaces.
"""
with SQLiteStorage._scheduler_lock:
if SQLiteStorage._current_scheduler is not None:
return SQLiteStorage._current_scheduler
hf_token = os.environ.get("HF_TOKEN")
dataset_id = os.environ.get("TRACKIO_DATASET_ID")
space_repo_name = os.environ.get("SPACE_REPO_NAME")
if dataset_id is None or space_repo_name is None:
scheduler = DummyCommitScheduler()
else:
scheduler = CommitScheduler(
repo_id=dataset_id,
repo_type="dataset",
folder_path=TRACKIO_DIR,
private=True,
allow_patterns=["*.parquet", "media/**/*"],
squash_history=True,
token=hf_token,
on_before_commit=SQLiteStorage.export_to_parquet,
)
SQLiteStorage._current_scheduler = scheduler
return scheduler
@staticmethod
def log(project: str, run: str, metrics: dict, step: int | None = None):
"""
Safely log metrics to the database. Before logging, this method will ensure the database exists
and is set up with the correct tables. It also uses a cross-process lock to prevent
database locking errors when multiple processes access the same database.
This method is not used in the latest versions of Trackio (replaced by bulk_log) but
is kept for backwards compatibility for users who are connecting to a newer version of
a Trackio Spaces dashboard with an older version of Trackio installed locally.
"""
db_path = SQLiteStorage.init_db(project)
with SQLiteStorage._get_process_lock(project):
with SQLiteStorage._get_connection(db_path) as conn:
cursor = conn.cursor()
cursor.execute(
"""
SELECT MAX(step)
FROM metrics
WHERE run_name = ?
""",
(run,),
)
last_step = cursor.fetchone()[0]
if step is None:
current_step = 0 if last_step is None else last_step + 1
else:
current_step = step
current_timestamp = datetime.now().isoformat()
cursor.execute(
"""
INSERT INTO metrics
(timestamp, run_name, step, metrics)
VALUES (?, ?, ?, ?)
""",
(
current_timestamp,
run,
current_step,
json.dumps(serialize_values(metrics)),
),
)
conn.commit()
@staticmethod
def bulk_log(
project: str,
run: str,
metrics_list: list[dict],
steps: list[int] | None = None,
timestamps: list[str] | None = None,
config: dict | None = None,
):
"""
Safely log bulk metrics to the database. Before logging, this method will ensure the database exists
and is set up with the correct tables. It also uses a cross-process lock to prevent
database locking errors when multiple processes access the same database.
"""
if not metrics_list:
return
if timestamps is None:
timestamps = [datetime.now().isoformat()] * len(metrics_list)
db_path = SQLiteStorage.init_db(project)
with SQLiteStorage._get_process_lock(project):
with SQLiteStorage._get_connection(db_path) as conn:
cursor = conn.cursor()
if steps is None:
steps = list(range(len(metrics_list)))
elif any(s is None for s in steps):
cursor.execute(
"SELECT MAX(step) FROM metrics WHERE run_name = ?", (run,)
)
last_step = cursor.fetchone()[0]
current_step = 0 if last_step is None else last_step + 1
processed_steps = []
for step in steps:
if step is None:
processed_steps.append(current_step)
current_step += 1
else:
processed_steps.append(step)
steps = processed_steps
if len(metrics_list) != len(steps) or len(metrics_list) != len(
timestamps
):
raise ValueError(
"metrics_list, steps, and timestamps must have the same length"
)
data = []
for i, metrics in enumerate(metrics_list):
data.append(
(
timestamps[i],
run,
steps[i],
json.dumps(serialize_values(metrics)),
)
)
cursor.executemany(
"""
INSERT INTO metrics
(timestamp, run_name, step, metrics)
VALUES (?, ?, ?, ?)
""",
data,
)
if config:
current_timestamp = datetime.now().isoformat()
cursor.execute(
"""
INSERT OR REPLACE INTO configs
(run_name, config, created_at)
VALUES (?, ?, ?)
""",
(run, json.dumps(serialize_values(config)), current_timestamp),
)
conn.commit()
@staticmethod
def get_logs(project: str, run: str) -> list[dict]:
"""Retrieve logs for a specific run. Logs include the step count (int) and the timestamp (datetime object)."""
db_path = SQLiteStorage.get_project_db_path(project)
if not db_path.exists():
return []
with SQLiteStorage._get_connection(db_path) as conn:
cursor = conn.cursor()
cursor.execute(
"""
SELECT timestamp, step, metrics
FROM metrics
WHERE run_name = ?
ORDER BY timestamp
""",
(run,),
)
rows = cursor.fetchall()
results = []
for row in rows:
metrics = json.loads(row["metrics"])
metrics = deserialize_values(metrics)
metrics["timestamp"] = row["timestamp"]
metrics["step"] = row["step"]
results.append(metrics)
return results
@staticmethod
def load_from_dataset():
dataset_id = os.environ.get("TRACKIO_DATASET_ID")
space_repo_name = os.environ.get("SPACE_REPO_NAME")
if dataset_id is not None and space_repo_name is not None:
hfapi = hf.HfApi()
updated = False
if not TRACKIO_DIR.exists():
TRACKIO_DIR.mkdir(parents=True, exist_ok=True)
with SQLiteStorage.get_scheduler().lock:
try:
files = hfapi.list_repo_files(dataset_id, repo_type="dataset")
for file in files:
# Download parquet and media assets
if not (file.endswith(".parquet") or file.startswith("media/")):
continue
if (TRACKIO_DIR / file).exists():
continue
hf.hf_hub_download(
dataset_id, file, repo_type="dataset", local_dir=TRACKIO_DIR
)
updated = True
except hf.errors.EntryNotFoundError:
pass
except hf.errors.RepositoryNotFoundError:
pass
if updated:
SQLiteStorage.import_from_parquet()
SQLiteStorage._dataset_import_attempted = True
@staticmethod
def get_projects() -> list[str]:
"""
Get list of all projects by scanning the database files in the trackio directory.
"""
if not SQLiteStorage._dataset_import_attempted:
SQLiteStorage.load_from_dataset()
projects: set[str] = set()
if not TRACKIO_DIR.exists():
return []
for db_file in TRACKIO_DIR.glob("*.db"):
project_name = db_file.stem
projects.add(project_name)
return sorted(projects)
@staticmethod
def get_runs(project: str) -> list[str]:
"""Get list of all runs for a project."""
db_path = SQLiteStorage.get_project_db_path(project)
if not db_path.exists():
return []
with SQLiteStorage._get_connection(db_path) as conn:
cursor = conn.cursor()
cursor.execute(
"SELECT DISTINCT run_name FROM metrics",
)
return [row[0] for row in cursor.fetchall()]
@staticmethod
def get_max_steps_for_runs(project: str) -> dict[str, int]:
"""Get the maximum step for each run in a project."""
db_path = SQLiteStorage.get_project_db_path(project)
if not db_path.exists():
return {}
with SQLiteStorage._get_connection(db_path) as conn:
cursor = conn.cursor()
cursor.execute(
"""
SELECT run_name, MAX(step) as max_step
FROM metrics
GROUP BY run_name
"""
)
results = {}
for row in cursor.fetchall():
results[row["run_name"]] = row["max_step"]
return results
@staticmethod
def store_config(project: str, run: str, config: dict) -> None:
"""Store configuration for a run."""
db_path = SQLiteStorage.init_db(project)
with SQLiteStorage._get_process_lock(project):
with SQLiteStorage._get_connection(db_path) as conn:
cursor = conn.cursor()
current_timestamp = datetime.now().isoformat()
cursor.execute(
"""
INSERT OR REPLACE INTO configs
(run_name, config, created_at)
VALUES (?, ?, ?)
""",
(run, json.dumps(serialize_values(config)), current_timestamp),
)
conn.commit()
@staticmethod
def get_run_config(project: str, run: str) -> dict | None:
"""Get configuration for a specific run."""
db_path = SQLiteStorage.get_project_db_path(project)
if not db_path.exists():
return None
with SQLiteStorage._get_connection(db_path) as conn:
cursor = conn.cursor()
try:
cursor.execute(
"""
SELECT config FROM configs WHERE run_name = ?
""",
(run,),
)
row = cursor.fetchone()
if row:
config = json.loads(row["config"])
return deserialize_values(config)
return None
except sqlite3.OperationalError as e:
if "no such table: configs" in str(e):
return None
raise
@staticmethod
def delete_run(project: str, run: str) -> bool:
"""Delete a run from the database (both metrics and config)."""
db_path = SQLiteStorage.get_project_db_path(project)
if not db_path.exists():
return False
with SQLiteStorage._get_process_lock(project):
with SQLiteStorage._get_connection(db_path) as conn:
cursor = conn.cursor()
try:
cursor.execute("DELETE FROM metrics WHERE run_name = ?", (run,))
cursor.execute("DELETE FROM configs WHERE run_name = ?", (run,))
conn.commit()
return True
except sqlite3.Error:
return False
@staticmethod
def get_all_run_configs(project: str) -> dict[str, dict]:
"""Get configurations for all runs in a project."""
db_path = SQLiteStorage.get_project_db_path(project)
if not db_path.exists():
return {}
with SQLiteStorage._get_connection(db_path) as conn:
cursor = conn.cursor()
try:
cursor.execute(
"""
SELECT run_name, config FROM configs
"""
)
results = {}
for row in cursor.fetchall():
config = json.loads(row["config"])
results[row["run_name"]] = deserialize_values(config)
return results
except sqlite3.OperationalError as e:
if "no such table: configs" in str(e):
return {}
raise
def finish(self):
"""Cleanup when run is finished."""
pass
|