xce009 commited on
Commit
39f50c2
·
1 Parent(s): 9d1b0c3
Files changed (2) hide show
  1. .gitignore +3 -1
  2. deploy.py +0 -990
.gitignore CHANGED
@@ -126,4 +126,6 @@ persistence/
126
 
127
  local_deploy.py
128
  test_vector_store_async.py
129
- deploy_sdk.py
 
 
 
126
 
127
  local_deploy.py
128
  test_vector_store_async.py
129
+ deploy_sdk.py
130
+
131
+ deploy_hf.py
deploy.py DELETED
@@ -1,990 +0,0 @@
1
- #!/usr/bin/env python3
2
- """
3
- deploy.py — Persistent data deployment for Hugging Face Spaces.
4
-
5
- Usage:
6
- python deploy.py
7
-
8
- Flow:
9
- 1. Back up existing data from the deployed Space -> persistence/
10
- 2. Commit and push local changes to the Space repository
11
- 3. Wait for the Space to redeploy (poll /health)
12
- 4. Restore the backed-up data to the new deployment
13
- 5. Generate a comprehensive audit report
14
- """
15
-
16
- import argparse
17
- import io
18
- import json
19
- import os
20
- import re
21
- import subprocess
22
- import sys
23
- import tarfile
24
- import time
25
- import urllib.error
26
- import urllib.request
27
- from datetime import datetime, timezone
28
- from pathlib import Path
29
- from typing import Any, Optional
30
-
31
- # ─────────────────────────────────────────────────────────────
32
- # Console encoding detection
33
- # ─────────────────────────────────────────────────────────────
34
-
35
- _USE_UNICODE = True
36
- try:
37
- "\u2713".encode(sys.stdout.encoding or "utf-8")
38
- except (UnicodeEncodeError, UnicodeDecodeError, LookupError):
39
- _USE_UNICODE = False
40
-
41
- # ─────────────────────────────────────────────────────────────
42
- # ANSI colours & helpers
43
- # ─────────────────────────────────────────────────────────────
44
-
45
- GREEN = "\033[92m"
46
- RED = "\033[91m"
47
- YELLOW = "\033[93m"
48
- CYAN = "\033[96m"
49
- BOLD = "\033[1m"
50
- RESET = "\033[0m"
51
-
52
-
53
- def _c(colour: str, text: str) -> str:
54
- return f"{colour}{text}{RESET}"
55
-
56
-
57
- def _ok(text: str) -> str:
58
- return _c(GREEN, text)
59
-
60
-
61
- def _fail(text: str) -> str:
62
- return _c(RED, text)
63
-
64
-
65
- def _warn(text: str) -> str:
66
- return _c(YELLOW, text)
67
-
68
-
69
- def _info(text: str) -> str:
70
- return _c(CYAN, text)
71
-
72
-
73
- def _bold(text: str) -> str:
74
- return _c(BOLD, text)
75
-
76
-
77
- # Icon constants (must be defined outside f-string expressions for Python < 3.12)
78
- # Unicode vs ASCII fallbacks for Windows console compatibility
79
- if _USE_UNICODE:
80
- _OK_ICON = _ok("\u2713")
81
- _FAIL_ICON = _fail("\u2717")
82
- _WARN_ICON = _warn("\u26A0")
83
- _INFO_ICON = _info("\u25B8")
84
- _BULLET = "\u2022"
85
- _DASH = "\u2014"
86
- _TABLE_TL = "\u2554"
87
- _TABLE_TR = "\u2557"
88
- _TABLE_BL = "\u255A"
89
- _TABLE_BR = "\u255D"
90
- _TABLE_H = "\u2550"
91
- _TABLE_V = "\u2551"
92
- _TABLE_TM = "\u2560"
93
- _TABLE_BM = "\u255A"
94
- _TABLE_ML = "\u251C"
95
- _TABLE_MR = "\u2524"
96
- _TABLE_MM = "\u253C"
97
- _TABLE_MH = "\u2500"
98
- _TABLE_MV = "\u2502"
99
- _TABLE_TML = "\u255F"
100
- _TABLE_TMR = "\u257E"
101
- else:
102
- _OK_ICON = _ok("v")
103
- _FAIL_ICON = _fail("x")
104
- _WARN_ICON = _warn("!")
105
- _INFO_ICON = _info(">")
106
- _BULLET = "*"
107
- _DASH = "-"
108
- _TABLE_TL = "+"
109
- _TABLE_TR = "+"
110
- _TABLE_BL = "+"
111
- _TABLE_BR = "+"
112
- _TABLE_H = "="
113
- _TABLE_V = "|"
114
- _TABLE_TM = "+"
115
- _TABLE_BM = "+"
116
- _TABLE_ML = "+"
117
- _TABLE_MR = "+"
118
- _TABLE_MM = "+"
119
- _TABLE_MH = "-"
120
- _TABLE_MV = "|"
121
- _TABLE_TML = "+"
122
- _TABLE_TMR = "-"
123
-
124
-
125
- def _indicator(status: str) -> str:
126
- mapping = {
127
- "passed": _OK_ICON,
128
- "failed": _FAIL_ICON,
129
- "skipped": _WARN_ICON,
130
- "running": _INFO_ICON,
131
- }
132
- return mapping.get(status, _DASH)
133
-
134
-
135
- def _timestamp() -> str:
136
- return datetime.now(timezone.utc).strftime("%H:%M:%S")
137
-
138
-
139
- def _human_size(n_bytes: int) -> str:
140
- for unit in ("B", "KB", "MB", "GB"):
141
- if n_bytes < 1024:
142
- return f"{n_bytes:.2f} {unit}"
143
- n_bytes /= 1024
144
- return f"{n_bytes:.2f} TB"
145
-
146
-
147
- def _human_duration(seconds: float) -> str:
148
- if seconds < 60:
149
- return f"{seconds:.1f}s"
150
- mins = int(seconds // 60)
151
- secs = int(seconds % 60)
152
- return f"{mins}m {secs}s"
153
-
154
-
155
- # ─────────────────────────────────────────────────────────────
156
- # Environment loading
157
- # ─────────────────────────────────────────────────────────────
158
-
159
-
160
- def load_env(env_path: str = ".env") -> dict[str, str]:
161
- env: dict[str, str] = {}
162
- p = Path(env_path)
163
- if not p.is_file():
164
- return env
165
- for line in p.read_text(encoding="utf-8").splitlines():
166
- line = line.strip()
167
- if not line or line.startswith("#"):
168
- continue
169
- m = re.match(r"^([A-Za-z_][A-Za-z_0-9]*)\s*=\s*(.*?)\s*$", line)
170
- if not m:
171
- continue
172
- key = m.group(1)
173
- val = m.group(2)
174
- if val.startswith('"') and val.endswith('"'):
175
- val = val[1:-1]
176
- elif val.startswith("'") and val.endswith("'"):
177
- val = val[1:-1]
178
- env[key] = val
179
- return env
180
-
181
-
182
- # ─────────────────────────────────────────────────────────────
183
- # HTTP helpers
184
- # ─────────────────────────────────────────────────────────────
185
-
186
- BACKUP_ARCHIVE = "backup.tar.gz"
187
- RESTORE_ENDPOINT = "/api/v1/backup/restore"
188
- BACKUP_ENDPOINT = "/api/v1/backup"
189
- HEALTH_ENDPOINT = "/health"
190
- HF_API_SPACES = "https://huggingface.co/api/spaces"
191
-
192
-
193
- def _build_url(base: str, path: str) -> str:
194
- base = base.rstrip("/")
195
- path = path.lstrip("/")
196
- return f"{base}/{path}"
197
-
198
-
199
- def http_get(url: str, headers: dict[str, str], timeout: int = 120) -> tuple[int, bytes]:
200
- req = urllib.request.Request(url, headers=headers, method="GET")
201
- with urllib.request.urlopen(req, timeout=timeout) as resp:
202
- return resp.status, resp.read()
203
-
204
-
205
- def http_get_stream(
206
- url: str, headers: dict[str, str], dest: Path, timeout: int = 300
207
- ) -> tuple[int, int]:
208
- req = urllib.request.Request(url, headers=headers, method="GET")
209
- with urllib.request.urlopen(req, timeout=timeout) as resp:
210
- total = 0
211
- with open(dest, "wb") as f:
212
- while True:
213
- chunk = resp.read(65536)
214
- if not chunk:
215
- break
216
- f.write(chunk)
217
- total += len(chunk)
218
- return resp.status, total
219
-
220
-
221
- def http_post_multipart(
222
- url: str,
223
- file_path: Path,
224
- field_name: str,
225
- headers: dict[str, str],
226
- timeout: int = 300,
227
- ) -> tuple[int, bytes]:
228
- boundary = "----DeployBoundary" + hex(int(time.time() * 1e6))[2:]
229
- data = io.BytesIO()
230
-
231
- data.write(f"--{boundary}\r\n".encode())
232
- data.write(
233
- f'Content-Disposition: form-data; name="{field_name}"; filename="{file_path.name}"\r\n'.encode()
234
- )
235
- data.write(b"Content-Type: application/gzip\r\n\r\n")
236
- data.write(file_path.read_bytes())
237
- data.write(f"\r\n--{boundary}--\r\n".encode())
238
-
239
- body = data.getvalue()
240
- content_type = f"multipart/form-data; boundary={boundary}"
241
-
242
- req_headers = {**headers, "Content-Type": content_type}
243
- req = urllib.request.Request(url, data=body, headers=req_headers, method="POST")
244
- with urllib.request.urlopen(req, timeout=timeout) as resp:
245
- return resp.status, resp.read()
246
-
247
-
248
- def http_get_json(
249
- url: str, headers: dict[str, str], timeout: int = 30
250
- ) -> Optional[dict[str, Any]]:
251
- try:
252
- req = urllib.request.Request(url, headers=headers, method="GET")
253
- with urllib.request.urlopen(req, timeout=timeout) as resp:
254
- return json.loads(resp.read().decode("utf-8"))
255
- except Exception:
256
- return None
257
-
258
-
259
- def http_post_json(
260
- url: str, headers: dict[str, str], data: Optional[bytes] = None, timeout: int = 30
261
- ) -> Optional[int]:
262
- try:
263
- req = urllib.request.Request(url, data=data, headers=headers, method="POST")
264
- with urllib.request.urlopen(req, timeout=timeout) as resp:
265
- return resp.status
266
- except urllib.error.HTTPError as e:
267
- return e.code
268
- except Exception:
269
- return None
270
-
271
-
272
- def http_head(url: str, timeout: int = 30) -> Optional[int]:
273
- try:
274
- req = urllib.request.Request(url, method="HEAD")
275
- with urllib.request.urlopen(req, timeout=timeout) as resp:
276
- return resp.status
277
- except urllib.error.URLError:
278
- return None
279
-
280
-
281
- # ─────────────────────────────────────────────────────────────
282
- # Git helpers
283
- # ─────────────────────────────────────────────────────────────
284
-
285
-
286
- def _run_git(args: list[str], cwd: str | None = None) -> tuple[int, str]:
287
- cmd = ["git"] + args
288
- result = subprocess.run(cmd, capture_output=True, text=True, cwd=cwd)
289
- return result.returncode, result.stdout.strip()
290
-
291
-
292
- def git_status(cwd: str) -> list[str]:
293
- rc, out = _run_git(["status", "--porcelain"], cwd=cwd)
294
- if rc != 0:
295
- return []
296
- lines = [line for line in out.split("\n") if line.strip()]
297
- return lines
298
-
299
-
300
- def git_add_all(cwd: str) -> bool:
301
- rc, _ = _run_git(["add", "-A", "."], cwd=cwd)
302
- return rc == 0
303
-
304
-
305
- def git_commit(cwd: str, message: str) -> tuple[bool, str]:
306
- rc, out = _run_git(["commit", "-m", message], cwd=cwd)
307
- if rc == 0:
308
- m = re.search(r"\[[^\]]+ ([a-f0-9]+)\]", out)
309
- sha = m.group(1) if m else "unknown"
310
- return True, sha
311
- if "nothing to commit" in out.lower() or "no changes" in out.lower():
312
- return True, "no-change"
313
- return False, out
314
-
315
-
316
- def git_push(cwd: str, remote: str = "origin", branch: str = "main") -> tuple[bool, str]:
317
- rc, out = _run_git(["push", remote, branch], cwd=cwd)
318
- return rc == 0, out
319
-
320
-
321
- def _parse_hf_space_from_url(url: str) -> tuple[str, str]:
322
- m = re.search(r"huggingface\.co/spaces/([^/]+)/([^/\s]+)", url)
323
- if m:
324
- return m.group(1), m.group(2).rstrip("/")
325
- return "aetherbase", "llm-ready-data"
326
-
327
-
328
- def git_remote_url(cwd: str) -> str:
329
- rc, out = _run_git(["remote", "get-url", "origin"], cwd=cwd)
330
- if rc != 0:
331
- return ""
332
- return out
333
-
334
-
335
- def git_log(cwd: str, n: int = 3) -> str:
336
- _, out = _run_git(["log", "--oneline", f"-{n}"], cwd=cwd)
337
- return out
338
-
339
-
340
- # ─────────────────────────────────────────────────────────────
341
- # Core deployment logic
342
- # ─────────────────────────────────────────────────────────────
343
-
344
-
345
- class AuditReport:
346
- def __init__(self):
347
- self.start_time = time.time()
348
- self.fields: dict[str, Any] = {
349
- "backup_started": "",
350
- "backup_completed": "",
351
- "backup_files": 0,
352
- "backup_size": 0,
353
- "backup_status": "pending",
354
- "commit_status": "pending",
355
- "commit_hash": "",
356
- "push_status": "pending",
357
- "deploy_status": "pending",
358
- "deploy_duration": "",
359
- "restore_status": "pending",
360
- "restore_files": 0,
361
- "verification_status": "pending",
362
- "errors": [],
363
- "warnings": [],
364
- }
365
-
366
- def end(self):
367
- self.fields["total_duration"] = _human_duration(time.time() - self.start_time)
368
-
369
- def set(self, key: str, value: Any):
370
- self.fields[key] = value
371
-
372
- def error(self, msg: str):
373
- self.fields["errors"].append(msg)
374
-
375
- def warn(self, msg: str):
376
- self.fields["warnings"].append(msg)
377
-
378
- def _row(self, label: str, status: str, detail: str) -> str:
379
- icon = _indicator(status)
380
- return f"{_TABLE_MV} {label:<28} {icon:<2} {detail:<30} {_TABLE_MV}"
381
-
382
- def _sep(self) -> str:
383
- return f"{_TABLE_ML}{_TABLE_MH * 28}{_TABLE_MM}{_TABLE_MH * 4}{_TABLE_MM}{_TABLE_MH * 32}{_TABLE_MR}"
384
-
385
- def print(self):
386
- self.end()
387
- f = self.fields
388
- err_count = len(f["errors"])
389
- warn_count = len(f["warnings"])
390
-
391
- start_dt = datetime.fromtimestamp(self.start_time).strftime("%c")
392
-
393
- lines: list[str] = []
394
- lines.append("")
395
- lines.append(
396
- f"{_TABLE_TL}{_TABLE_H * 16}{_TABLE_H * 16}{_TABLE_H * 18}{_TABLE_H * 20}"
397
- f"{_TABLE_TR}"
398
- )
399
-
400
- title = "DEPLOYMENT AUDIT REPORT"
401
- padding = (70 - len(title)) // 2
402
- lines.append(
403
- f"{_TABLE_V} {' ' * padding}{_bold(title)}{' ' * (70 - padding - len(title))} {_TABLE_V}"
404
- )
405
- lines.append(
406
- f"{_TABLE_V} {' ' * (70 - len(start_dt))}{start_dt} {_TABLE_V}"
407
- )
408
- lines.append(
409
- f"{_TABLE_TM}{_TABLE_H * 70}{_TABLE_H * 0}"
410
- )
411
-
412
- lines.append(self._row("Category", "", "Status | Detail"))
413
- lines.append(
414
- f"{_TABLE_TML}{_TABLE_MH * 70}"
415
- )
416
-
417
- lines.append(self._row("", "", ""))
418
-
419
- # ── Backup section ──
420
- lines.append(
421
- self._row(
422
- "Backup Started",
423
- "passed" if f["backup_started"] else "failed",
424
- f["backup_started"] or "\u2014",
425
- )
426
- )
427
- lines.append(
428
- self._row(
429
- "Backup Completed",
430
- "passed" if f["backup_completed"] else "failed",
431
- f["backup_completed"] or "\u2014",
432
- )
433
- )
434
- lines.append(
435
- self._row(
436
- "Files Backed Up",
437
- "passed" if f["backup_files"] > 0 else "failed",
438
- str(f["backup_files"]),
439
- )
440
- )
441
- lines.append(
442
- self._row(
443
- "Total Backup Size",
444
- "passed" if f["backup_size"] > 0 else "failed",
445
- _human_size(f["backup_size"]),
446
- )
447
- )
448
-
449
- lines.append(self._sep())
450
-
451
- # ── Deployment section ──
452
- lines.append(
453
- self._row(
454
- "Commit Status",
455
- f["commit_status"],
456
- f["commit_hash"] or "\u2014",
457
- )
458
- )
459
- lines.append(
460
- self._row(
461
- "Push Status",
462
- f["push_status"],
463
- _ok("Pushed") if f["push_status"] == "passed" else _fail("Failed"),
464
- )
465
- )
466
- dep_icon = "passed" if f["deploy_status"] == "passed" else "failed"
467
- lines.append(
468
- self._row(
469
- "HF Space Deployment",
470
- dep_icon,
471
- f["deploy_duration"] or "\u2014",
472
- )
473
- )
474
-
475
- lines.append(self._sep())
476
-
477
- # ── Restore section ──
478
- lines.append(
479
- self._row(
480
- "Data Restoration",
481
- f["restore_status"],
482
- _ok("Restored") if f["restore_status"] == "passed" else _fail("Failed"),
483
- )
484
- )
485
- lines.append(
486
- self._row(
487
- "Files Restored",
488
- "passed" if f["restore_files"] > 0 else "failed",
489
- str(f["restore_files"]),
490
- )
491
- )
492
- lines.append(
493
- self._row(
494
- "Verification Status",
495
- f["verification_status"],
496
- _ok("Verified") if f["verification_status"] == "passed" else _fail("Failed"),
497
- )
498
- )
499
-
500
- lines.append(self._sep())
501
-
502
- # ── Summary section ──
503
- lines.append(
504
- self._row(
505
- "Total Execution Time",
506
- "passed",
507
- f.get("total_duration", ""),
508
- )
509
- )
510
- lines.append(
511
- self._row(
512
- "Errors",
513
- "failed" if err_count > 0 else "passed",
514
- str(err_count),
515
- )
516
- )
517
- lines.append(
518
- self._row(
519
- "Warnings",
520
- "passed" if warn_count == 0 else "skipped",
521
- str(warn_count),
522
- )
523
- )
524
-
525
- lines.append(
526
- f"{_TABLE_BL}{_TABLE_H * 16}{_TABLE_H * 16}{_TABLE_H * 18}{_TABLE_H * 20}"
527
- f"{_TABLE_BR}"
528
- )
529
-
530
- for line in lines:
531
- print(line)
532
-
533
- if err_count > 0:
534
- print(f"\n{_fail('Errors:')}")
535
- for e in f["errors"]:
536
- print(f" {_fail(_BULLET)} {e}")
537
-
538
- if warn_count > 0:
539
- print(f"\n{_warn('Warnings:')}")
540
- for w in f["warnings"]:
541
- print(f" {_warn(_BULLET)} {w}")
542
-
543
-
544
- def run_deployment(args: argparse.Namespace, env: dict[str, str]) -> int:
545
- audit = AuditReport()
546
- cwd = os.getcwd()
547
- persistence_dir = Path(cwd) / "persistence"
548
- persistence_dir.mkdir(parents=True, exist_ok=True)
549
-
550
- space_url = args.space_url or env.get("SPACE_URL", "https://aetherbase-llm-ready-data.hf.space")
551
- api_key = args.api_key or env.get("api_key", env.get("API_KEY", ""))
552
- hf_token = args.hf_token or env.get("hf_token", env.get("HF_TOKEN", ""))
553
- skip_backup = args.skip_backup
554
-
555
- if not api_key:
556
- print(f" {_fail('ERROR:')} No API key found. Set API_KEY in .env or pass --api-key.")
557
- return 1
558
-
559
- if not hf_token:
560
- print(f" {_warn('WARNING:')} No HF token. Can't verify Space build status. Set HF_TOKEN in .env or pass --hf-token.")
561
- hf_token = None
562
-
563
- auth_headers = {
564
- "Authorization": f"Bearer {api_key}",
565
- "User-Agent": "deploy.py/1.0",
566
- }
567
-
568
- remote_url = git_remote_url(cwd)
569
- owner, space_name = _parse_hf_space_from_url(remote_url) if remote_url else ("aetherbase", "llm-ready-data")
570
- hf_api_headers = None
571
- if hf_token:
572
- hf_api_headers = {
573
- "Authorization": f"Bearer {hf_token}",
574
- "User-Agent": "deploy.py/1.0",
575
- }
576
-
577
- # ──────────────────────────────────────────────
578
- # PHASE 1: LOCK WRITES + BACKUP
579
- # ──────────────────────────────────────────────
580
- print(f"\n{_bold('PHASE 1/4: Lock Writes & Backup')}")
581
- print(f" {_INFO_ICON} Backing up data from: {space_url}")
582
-
583
- if skip_backup:
584
- print(f" {_WARN_ICON} Backup skipped (--skip-backup)")
585
- audit.set("backup_status", "skipped")
586
- else:
587
- # Enable maintenance mode to freeze data state
588
- maint_url = _build_url(space_url, "/api/v1/maintenance/enable")
589
- maint_enabled = False
590
- try:
591
- status = http_post_json(maint_url, auth_headers, timeout=15)
592
- if status == 200:
593
- maint_enabled = True
594
- print(f" {_OK_ICON} Maintenance enabled — writes blocked")
595
- except Exception:
596
- print(f" {_WARN_ICON} Could not enable maintenance (endpoint may not exist yet)")
597
- print(f" {_WARN_ICON} Data may be lost if writes occur during deployment")
598
-
599
- backup_file = persistence_dir / BACKUP_ARCHIVE
600
- backup_url = _build_url(space_url, BACKUP_ENDPOINT)
601
- audit.set("backup_started", _timestamp())
602
- backed_up = False
603
-
604
- # Try remote backup endpoint first
605
- try:
606
- status, total_bytes = http_get_stream(backup_url, auth_headers, backup_file, timeout=120)
607
- if status == 200:
608
- backed_up = True
609
- print(f" {_OK_ICON} Remote backup successful")
610
- except Exception as exc:
611
- print(f" {_WARN_ICON} Remote backup unavailable: {exc}")
612
-
613
- # Fall back to local backup
614
- if not backed_up:
615
- data_dir = Path(cwd) / "data"
616
- if data_dir.is_dir():
617
- print(f" {_INFO_ICON} Falling back to local backup: {data_dir}")
618
- try:
619
- buf = io.BytesIO()
620
- with tarfile.open(fileobj=buf, mode="w:gz") as tar:
621
- for path in sorted(data_dir.rglob("*")):
622
- if path.is_file():
623
- arcname = path.relative_to(data_dir.parent)
624
- tar.add(str(path), arcname=str(arcname))
625
- buf.seek(0)
626
- backup_file.write_bytes(buf.read())
627
-
628
- total_bytes = backup_file.stat().st_size
629
- file_count = 0
630
- with tarfile.open(backup_file, "r:gz") as tar:
631
- file_count = sum(1 for m in tar.getmembers() if m.isfile())
632
-
633
- backed_up = True
634
- print(f" {_OK_ICON} Local backup saved: {backup_file}")
635
- except Exception as exc:
636
- msg = f"Local backup failed: {exc}"
637
- audit.error(msg)
638
- print(f" {_FAIL_ICON} {msg}")
639
- audit.print()
640
- return 1
641
- else:
642
- msg = f"No data found locally ({data_dir}) and remote backup unavailable"
643
- audit.warn(msg)
644
- print(f" {_WARN_ICON} {msg}")
645
- audit.set("backup_files", 0)
646
- audit.set("backup_size", 0)
647
- audit.set("backup_completed", _timestamp())
648
- audit.set("backup_status", "skipped")
649
-
650
- if backed_up:
651
- file_count = 0
652
- try:
653
- with tarfile.open(backup_file, "r:gz") as tar:
654
- file_count = sum(1 for m in tar.getmembers() if m.isfile())
655
- except Exception:
656
- audit.warn("Could not count files in backup archive")
657
- file_count = backup_file.stat().st_size
658
-
659
- audit.set("backup_completed", _timestamp())
660
- audit.set("backup_files", file_count)
661
- audit.set("backup_size", total_bytes)
662
- audit.set("backup_status", "passed")
663
- print(f" {_OK_ICON} Files: {file_count} | Size: {_human_size(total_bytes)}")
664
- if maint_enabled:
665
- print(f" {_OK_ICON} Writes frozen — backup is consistent")
666
-
667
- # ──────────────────────────────────────────────
668
- # PHASE 2: GIT COMMIT & PUSH
669
- # ──────────────────────────────────────────────
670
- print(f"\n{_bold('PHASE 2/4: Git Commit & Push')}")
671
-
672
- if args.skip_git:
673
- print(f" {_WARN_ICON} Git phase skipped (--skip-git)")
674
- audit.set("commit_status", "skipped")
675
- audit.set("commit_hash", "skipped")
676
- audit.set("push_status", "skipped")
677
- else:
678
- changed = git_status(cwd)
679
- if not changed:
680
- print(f" {_WARN_ICON} No changes to commit (working tree clean)")
681
- print(f" {_INFO_ICON} Nothing to deploy — exiting")
682
- audit.set("commit_status", "passed")
683
- audit.set("commit_hash", "no-change")
684
- audit.set("push_status", "passed")
685
- audit.print()
686
- return 0
687
-
688
- print(f" {_INFO_ICON} Staging {len(changed)} file(s)...")
689
- if not git_add_all(cwd):
690
- msg = "Git add failed"
691
- audit.error(msg)
692
- audit.set("commit_status", "failed")
693
- print(f" {_FAIL_ICON} {msg}")
694
- audit.print()
695
- return 1
696
-
697
- commit_msg = args.message or f"deploy: auto-deploy {_timestamp()}"
698
- print(f" {_INFO_ICON} Committing: {commit_msg}")
699
- ok, sha = git_commit(cwd, commit_msg)
700
- if not ok:
701
- msg = f"Git commit failed: {sha}"
702
- audit.error(msg)
703
- audit.set("commit_status", "failed")
704
- print(f" {_FAIL_ICON} {msg}")
705
- audit.print()
706
- return 1
707
-
708
- audit.set("commit_status", "passed")
709
- audit.set("commit_hash", sha)
710
- print(f" {_OK_ICON} Committed: {sha}")
711
-
712
- print(f" {_INFO_ICON} Pushing to origin/main...")
713
- ok, push_out = git_push(cwd)
714
- if not ok:
715
- msg = f"Git push failed"
716
- audit.error(msg)
717
- audit.set("push_status", "failed")
718
- print(f" {_FAIL_ICON} {msg}")
719
- print(f" {push_out}")
720
- audit.print()
721
- return 1
722
-
723
- audit.set("push_status", "passed")
724
- print(f" {_OK_ICON} Push successful")
725
-
726
- # ──────────────────────────────────────────────
727
- # PHASE 3: WAIT FOR REDEPLOYMENT (via HF API)
728
- # ──────────────────────────────────────────────
729
- print(f"\n{_bold('PHASE 3/4: Waiting for HF Space Redeployment')}")
730
-
731
- api_url = f"{HF_API_SPACES}/{owner}/{space_name}"
732
- deploy_start = time.time()
733
- max_wait = args.timeout
734
- poll_interval = 10
735
- waited = 0
736
- deployed = False
737
- last_error = ""
738
- seen_building = False
739
-
740
- print(f" {_INFO_ICON} Tracking {owner}/{space_name} via HF API...")
741
-
742
- if hf_api_headers:
743
- initial = http_get_json(api_url, hf_api_headers, timeout=15) or {}
744
- prev_stage = initial.get("runtime", {}).get("stage", "")
745
-
746
- while waited < max_wait:
747
- data = http_get_json(api_url, hf_api_headers, timeout=15)
748
- stage = (data or {}).get("runtime", {}).get("stage", "")
749
-
750
- if stage == "BUILDING":
751
- if not seen_building:
752
- print(f" {_INFO_ICON} Build started (stage: BUILDING)")
753
- seen_building = True
754
-
755
- elif stage == "RUNNING":
756
- if seen_building:
757
- print(f" {_OK_ICON} Build complete (stage: RUNNING)")
758
- deployed = True
759
- break
760
- if waited > 30 and prev_stage == "RUNNING":
761
- print(f" {_OK_ICON} Space is running (may have skipped BUILDING stage)")
762
- deployed = True
763
- break
764
-
765
- elif stage in ("PAUSED", "STOPPED", "NO_APP"):
766
- last_error = f"Space in unexpected state: {stage}"
767
- deployed = False
768
- break
769
-
770
- prev_stage = stage
771
- time.sleep(poll_interval)
772
- waited += poll_interval
773
- if waited % 30 == 0 and not seen_building:
774
- print(f" {_INFO_ICON} Still waiting for build to start... ({waited}s)")
775
- elif waited % 30 == 0 and seen_building:
776
- print(f" {_INFO_ICON} Still building... ({waited}s)")
777
-
778
- deploy_duration = time.time() - deploy_start
779
-
780
- if not hf_api_headers:
781
- print(f" {_WARN_ICON} No HF token — assuming deployment is underway")
782
- print(f" {_INFO_ICON} Waiting {max_wait}s for build + startup...")
783
- time.sleep(max_wait)
784
- deployed = True
785
-
786
- if deployed:
787
- health_url = _build_url(space_url, HEALTH_ENDPOINT)
788
- try:
789
- req = urllib.request.Request(health_url, method="GET")
790
- with urllib.request.urlopen(req, timeout=15) as resp:
791
- if resp.status == 200:
792
- print(f" {_OK_ICON} App health check passed (new instance serving)")
793
- except Exception:
794
- print(f" {_WARN_ICON} App health check unavailable (may still be starting)")
795
-
796
- audit.set("deploy_status", "passed")
797
- audit.set("deploy_duration", _human_duration(deploy_duration))
798
- print(f" {_OK_ICON} Space redeployed in {_human_duration(deploy_duration)}")
799
- else:
800
- audit.set("deploy_status", "failed")
801
- audit.set("deploy_duration", _human_duration(deploy_duration))
802
- msg = f"Space did not redeploy within {max_wait}s (last stage: {stage}, error: {last_error})"
803
- audit.error(msg)
804
- print(f" {_FAIL_ICON} {msg}")
805
- audit.print()
806
- return 1
807
-
808
- # ──────────────────────────────────────────────
809
- # PHASE 4: DATA RESTORE
810
- # ──────────────────────────────────────────────
811
- print(f"\n{_bold('PHASE 4/4: Data Restoration')}")
812
-
813
- if skip_backup:
814
- print(f" {_WARN_ICON} Restore skipped (no backup)")
815
- audit.set("restore_status", "skipped")
816
- audit.set("verification_status", "skipped")
817
- audit.set("restore_files", 0)
818
- else:
819
- backup_file = persistence_dir / BACKUP_ARCHIVE
820
- if not backup_file.is_file():
821
- msg = f"Backup file not found: {backup_file}"
822
- audit.error(msg)
823
- audit.set("restore_status", "failed")
824
- print(f" {_FAIL_ICON} {msg}")
825
- audit.print()
826
- return 1
827
-
828
- restore_url = _build_url(space_url, RESTORE_ENDPOINT)
829
- print(f" {_INFO_ICON} Restoring data to: {space_url}")
830
-
831
- try:
832
- status, body = http_post_multipart(
833
- restore_url, backup_file, "file", auth_headers, timeout=300
834
- )
835
-
836
- if status == 200:
837
- try:
838
- result = json.loads(body)
839
- restored = result.get("files_restored", 0)
840
- except Exception:
841
- restored = 0
842
-
843
- audit.set("restore_files", max(restored, 1))
844
- audit.set("restore_status", "passed")
845
- print(f" {_OK_ICON} Data restored: {restored} files")
846
-
847
- # Verification
848
- print(f" {_INFO_ICON} Verifying data restoration...")
849
- try:
850
- verify_backup = persistence_dir / "verify_check.tar.gz"
851
- status_v, _ = http_get_stream(
852
- _build_url(space_url, BACKUP_ENDPOINT),
853
- auth_headers,
854
- verify_backup,
855
- timeout=120,
856
- )
857
-
858
- if status_v == 200:
859
- with tarfile.open(verify_backup, "r:gz") as tar:
860
- verify_files = sum(1 for m in tar.getmembers() if m.isfile())
861
- verify_backup.unlink(missing_ok=True)
862
-
863
- if verify_files >= audit.fields.get("backup_files", 0) * 0.9:
864
- audit.set("verification_status", "passed")
865
- print(f" {_OK_ICON} Verification passed: {verify_files} files found")
866
- else:
867
- audit.set("verification_status", "failed")
868
- audit.warn(
869
- f"Verification file count ({verify_files}) mismatch with backup ({audit.fields.get('backup_files', 0)})"
870
- )
871
- print(f" {_WARN_ICON} Verification: file count mismatch ({verify_files} vs {audit.fields.get('backup_files', 0)})")
872
- else:
873
- audit.set("verification_status", "failed")
874
- audit.warn(f"Verification request returned HTTP {status_v}")
875
- print(f" {_WARN_ICON} Verification failed (HTTP {status_v})")
876
- except Exception as exc:
877
- audit.set("verification_status", "failed")
878
- audit.warn(f"Verification error: {exc}")
879
- print(f" {_WARN_ICON} Verification error: {exc}")
880
- else:
881
- audit.set("restore_status", "failed")
882
- msg = f"Restore returned HTTP {status}: {body[:200]}"
883
- audit.error(msg)
884
- print(f" {_FAIL_ICON} {msg}")
885
- audit.print()
886
- return 1
887
-
888
- except Exception as exc:
889
- audit.set("restore_status", "failed")
890
- msg = f"Restore failed: {exc}"
891
- audit.error(msg)
892
- print(f" {_FAIL_ICON} {msg}")
893
- audit.print()
894
- return 1
895
-
896
- # ──────────────────────────────────────────────
897
- # SUMMARY
898
- # ──────────────────────────────────────────────
899
- print(f"\n{_bold('=' * 70)}")
900
- audit.print()
901
-
902
- has_errors = len(audit.fields["errors"]) > 0
903
- if has_errors:
904
- print(f"\n{_fail('Deployment completed with errors.')}")
905
- return 1
906
- else:
907
- print(f"\n{_ok('Deployment completed successfully.')}")
908
- return 0
909
-
910
-
911
- # ─────────────────────────────────────────────────────────────
912
- # CLI entrypoint
913
- # ─────────────────────────────────────────────────────────────
914
-
915
-
916
- def main():
917
- parser = argparse.ArgumentParser(
918
- description="Deploy to Hugging Face Spaces with persistent data handling.",
919
- formatter_class=argparse.RawDescriptionHelpFormatter,
920
- epilog=(
921
- "Environment variables (from .env or process env):\n"
922
- " API_KEY API key for the Hugging Face Space\n"
923
- " SPACE_URL Base URL of the deployed Space\n"
924
- " HF_TOKEN Hugging Face API token (for build tracking)\n"
925
- ),
926
- )
927
- parser.add_argument(
928
- "--space-url",
929
- default="",
930
- help="Base URL of the Hugging Face Space (e.g. https://aetherbase-llm-ready-data.hf.space)",
931
- )
932
- parser.add_argument(
933
- "--api-key",
934
- default="",
935
- help="API key for authenticating with the Space",
936
- )
937
- parser.add_argument(
938
- "--message", "-m",
939
- default="",
940
- help="Git commit message",
941
- )
942
- parser.add_argument(
943
- "--timeout",
944
- type=int,
945
- default=300,
946
- help="Maximum wait time (seconds) for Space redeployment (default: 300)",
947
- )
948
- parser.add_argument(
949
- "--hf-token",
950
- default="",
951
- help="Hugging Face API token (for tracking Space build status)",
952
- )
953
- parser.add_argument(
954
- "--skip-backup",
955
- action="store_true",
956
- help="Skip the backup and restore phases",
957
- )
958
- parser.add_argument(
959
- "--skip-git",
960
- action="store_true",
961
- help="Skip the git commit and push phase",
962
- )
963
- parser.add_argument(
964
- "--env-file",
965
- default=".env",
966
- help="Path to .env file (default: .env)",
967
- )
968
-
969
- args = parser.parse_args()
970
- env = load_env(args.env_file)
971
-
972
- # Merge with process environment (process env takes precedence)
973
- for key in ("API_KEY", "SPACE_URL", "HF_TOKEN"):
974
- if os.environ.get(key):
975
- env[key] = os.environ[key]
976
-
977
- try:
978
- rc = run_deployment(args, env)
979
- except KeyboardInterrupt:
980
- print(f"\n{_warn('Deployment interrupted by user.')}")
981
- rc = 130
982
- except Exception as exc:
983
- print(f"\n{_fail('Unexpected error: ' + str(exc))}")
984
- rc = 1
985
-
986
- sys.exit(rc)
987
-
988
-
989
- if __name__ == "__main__":
990
- main()