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#!/usr/bin/env python3
"""Batch-submit PDFs to a running Equalify Reflow API and collect results.
Two input modes:
--manifest PATH Read PDF paths from a manifest file (one path per line,
`#` introduces a comment). Paths are resolved relative
to the manifest file's parent directory. Use this for
reproducible benchmark corpora.
--pdf-dir PATH Glob `*.pdf` from a directory. Use this for ad-hoc runs
over a local fixture set.
Outputs a timestamped directory under `batch-results/` (or `--output`) with
per-document `metadata.json`, `result.md`, and `figures/`, plus an aggregate
`summary.json`.
"""
from __future__ import annotations
import argparse
import asyncio
import json
import os
import statistics
import sys
import time
from datetime import UTC, datetime
from pathlib import Path
import httpx
# ---------------------------------------------------------------------------
# Configuration
# ---------------------------------------------------------------------------
DEFAULT_API_URL = os.environ.get("BATCH_API_URL", "http://localhost:8080")
DEFAULT_API_KEY = os.environ.get("BATCH_API_KEY", "")
DEFAULT_CONCURRENCY = int(os.environ.get("BATCH_CONCURRENCY", "2"))
POLL_INTERVAL = 10 # seconds
SUBMIT_TIMEOUT = 120 # seconds for upload
POLL_TIMEOUT = 900 # max wait per job (15min accommodates GPU cold-start)
AUTO_APPROVE_PII = True
SUBMIT_DELAY = 5 # seconds between submissions to avoid rate limiting
MAX_RETRIES = 5
RETRY_BASE_DELAY = 10
REPO_ROOT = Path(__file__).resolve().parent.parent
DEFAULT_RESULTS_BASE = REPO_ROOT / "batch-results"
# ---------------------------------------------------------------------------
# Manifest + discovery
# ---------------------------------------------------------------------------
def read_manifest(manifest: Path) -> list[Path]:
"""Parse a manifest file into a list of absolute PDF paths.
Paths in the manifest are resolved relative to the manifest's parent dir.
Lines starting with `#` and blank lines are ignored. An inline ` # ...`
comment on a path line is also stripped.
"""
if not manifest.is_file():
raise FileNotFoundError(f"Manifest not found: {manifest}")
base = manifest.parent
paths: list[Path] = []
for raw in manifest.read_text().splitlines():
line = raw.split("#", 1)[0].strip()
if not line:
continue
p = (base / line).resolve() if not Path(line).is_absolute() else Path(line)
paths.append(p)
return paths
def discover_pdfs(pdf_dir: Path) -> list[Path]:
if not pdf_dir.is_dir():
raise NotADirectoryError(f"PDF directory not found: {pdf_dir}")
return sorted(pdf_dir.glob("*.pdf"))
# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------
def short_name(path: Path) -> str:
stem = path.stem
return stem[:30] if len(stem) > 30 else stem
def mb(path: Path) -> str:
return f"{path.stat().st_size / 1_048_576:.1f} MB"
# Rate-limit gate: only one submission at a time with delay between
_submit_lock = asyncio.Lock()
async def throttled_submit(
client: httpx.AsyncClient,
pdf: Path,
label: str,
api_url: str,
api_key: str,
) -> httpx.Response:
"""Submit with retry + backoff for 429/5xx, and inter-submission delay."""
last_error: str | None = None
resp: httpx.Response | None = None
for attempt in range(MAX_RETRIES + 1):
async with _submit_lock:
if attempt > 0:
delay = RETRY_BASE_DELAY * (2 ** (attempt - 1))
print(f" [{label}] Retry {attempt}/{MAX_RETRIES} after {delay}s...")
await asyncio.sleep(delay)
else:
await asyncio.sleep(SUBMIT_DELAY)
with open(pdf, "rb") as f:
resp = await client.post(
f"{api_url}/api/v1/documents/submit",
headers={"X-API-Key": api_key},
files={"file": (pdf.name, f, "application/pdf")},
timeout=SUBMIT_TIMEOUT,
)
if resp.status_code == 429:
retry_after = int(resp.headers.get("Retry-After", RETRY_BASE_DELAY))
print(f" [{label}] Rate limited (429), waiting {retry_after}s...")
await asyncio.sleep(retry_after)
last_error = "429 Too Many Requests"
continue
if resp.status_code >= 500:
last_error = f"{resp.status_code} Server Error"
continue
if resp.status_code == 413:
resp.raise_for_status()
resp.raise_for_status()
return resp
assert resp is not None
raise httpx.HTTPStatusError(
f"Failed after {MAX_RETRIES} retries: {last_error}",
request=resp.request,
response=resp,
)
# ---------------------------------------------------------------------------
# Core
# ---------------------------------------------------------------------------
async def submit_and_process(
client: httpx.AsyncClient,
pdf: Path,
idx: int,
total: int,
sem: asyncio.Semaphore,
results_dir: Path,
api_url: str,
api_key: str,
) -> dict:
label = short_name(pdf)
async with sem:
print(f"[{idx}/{total}] Submitting: {label} ({mb(pdf)})", flush=True)
t0 = time.monotonic()
try:
resp = await throttled_submit(client, pdf, label, api_url, api_key)
data = resp.json()
job_id = data["job_id"]
print(f" [{label}] Job ID: {job_id}", flush=True)
except Exception as e:
elapsed = time.monotonic() - t0
print(f" [{label}] SUBMIT ERROR: {e}", flush=True)
return {"file": pdf.name, "status": "submit_error", "error": str(e), "elapsed": elapsed}
deadline = time.monotonic() + POLL_TIMEOUT
consecutive_errors = 0
last_phase: str | None = None
while time.monotonic() < deadline:
await asyncio.sleep(POLL_INTERVAL)
try:
resp = await client.get(
f"{api_url}/api/v1/documents/{job_id}",
headers={"X-API-Key": api_key},
timeout=30,
)
resp.raise_for_status()
consecutive_errors = 0
data = resp.json()
status = data.get("status", "unknown")
if status == "processing":
phase = data.get("processing_phase", "")
jobs_done = data.get("jobs_complete", 0)
jobs_total = data.get("jobs_total", 0)
progress = f"{phase} ({jobs_done}/{jobs_total})" if jobs_total else phase
if progress != last_phase:
elapsed_so_far = time.monotonic() - t0
print(f" [{label}] {elapsed_so_far:.0f}s — {progress}", flush=True)
last_phase = progress
elif status != last_phase:
print(f" [{label}] status: {status}", flush=True)
last_phase = status
if status == "awaiting_approval" and AUTO_APPROVE_PII:
token = data.get("approval_token", "")
if token:
try:
approve_resp = await client.post(
f"{api_url}/api/v1/approval/{token}/decision",
headers={
"X-API-Key": api_key,
"Content-Type": "application/json",
},
json={"decision": "approved", "reviewed_by": "batch_run"},
timeout=30,
)
approve_resp.raise_for_status()
pii_count = len(data.get("pii_findings", []))
print(f" [{label}] Auto-approved {pii_count} PII findings", flush=True)
except Exception as e:
print(f" [{label}] PII approve error: {e}", flush=True)
continue
if status == "completed":
elapsed = time.monotonic() - t0
pages = data.get("total_pages", "?")
cost = data.get("llm_cost", {}).get("estimated_cost_dollars", 0)
print(f" [{label}] OK in {elapsed / 60:.1f}m — {pages}p, ${cost:.4f}", flush=True)
await save_result(client, data, results_dir / label)
return {
"file": pdf.name,
"status": "completed",
"job_id": job_id,
"pages": pages,
"cost": cost,
"tokens": data.get("llm_cost", {}).get("total_tokens", 0),
"elapsed": elapsed,
"edits": data.get("total_edits", 0),
}
if status == "failed":
elapsed = time.monotonic() - t0
error = data.get("error", "unknown")
print(f" [{label}] FAIL in {elapsed / 60:.1f}m", flush=True)
return {
"file": pdf.name,
"status": "failed",
"job_id": job_id,
"error": error,
"elapsed": elapsed,
}
except Exception as e:
consecutive_errors += 1
if consecutive_errors >= 5:
print(f" [{label}] Too many poll errors, giving up: {e}", flush=True)
elapsed = time.monotonic() - t0
return {"file": pdf.name, "status": "error", "job_id": job_id, "error": str(e), "elapsed": elapsed}
print(f" [{label}] Poll error ({consecutive_errors}/5): {e}", flush=True)
elapsed = time.monotonic() - t0
print(f" [{label}] TIMEOUT after {elapsed / 60:.1f}m", flush=True)
return {"file": pdf.name, "status": "timeout", "job_id": job_id, "elapsed": elapsed}
async def save_result(client: httpx.AsyncClient, data: dict, out_dir: Path) -> None:
out_dir.mkdir(parents=True, exist_ok=True)
(out_dir / "metadata.json").write_text(json.dumps(data, indent=2, default=str))
md_url = data.get("markdown_url", "")
if md_url:
try:
resp = await client.get(md_url, timeout=30)
resp.raise_for_status()
(out_dir / "result.md").write_bytes(resp.content)
except Exception:
pass
figures = data.get("figures", [])
if figures:
fig_dir = out_dir / "figures"
fig_dir.mkdir(exist_ok=True)
for fig in figures:
url = fig.get("url", "")
fid = fig.get("figure_id", "figure")
if url:
try:
resp = await client.get(url, timeout=30)
resp.raise_for_status()
(fig_dir / f"{fid}.png").write_bytes(resp.content)
except Exception:
pass
def parse_args() -> argparse.Namespace:
p = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
src = p.add_mutually_exclusive_group(required=True)
src.add_argument("--manifest", type=Path, help="Path to a manifest file listing PDFs to process.")
src.add_argument("--pdf-dir", type=Path, help="Directory to glob *.pdf from.")
p.add_argument("--output", type=Path, default=None, help="Output directory (default: batch-results/<timestamp>).")
p.add_argument("--api-url", default=DEFAULT_API_URL, help=f"API base URL (default: {DEFAULT_API_URL}).")
p.add_argument(
"--concurrency",
type=int,
default=DEFAULT_CONCURRENCY,
help=f"Concurrent jobs (default: {DEFAULT_CONCURRENCY}).",
)
return p.parse_args()
async def main() -> None:
args = parse_args()
api_key = DEFAULT_API_KEY
if not api_key:
print("Error: Set BATCH_API_KEY environment variable", file=sys.stderr)
sys.exit(1)
if args.manifest:
pdfs = read_manifest(args.manifest)
else:
pdfs = discover_pdfs(args.pdf_dir)
missing = [p for p in pdfs if not p.is_file()]
if missing:
print(f"WARNING: {len(missing)} PDFs not found:", file=sys.stderr)
for p in missing:
print(f" - {p}", file=sys.stderr)
pdfs = [p for p in pdfs if p.is_file()]
if not pdfs:
print("No PDFs to process.", file=sys.stderr)
sys.exit(1)
# Sort by size ascending to avoid overloading docling with a huge doc first
pdfs.sort(key=lambda p: p.stat().st_size)
stamp = datetime.now(UTC).strftime("%Y-%m-%dT%H-%M-%S")
results_dir = args.output or (DEFAULT_RESULTS_BASE / stamp)
results_dir.mkdir(parents=True, exist_ok=True)
total = len(pdfs)
print(f"Batch run: {total} PDFs, concurrency={args.concurrency}")
print(f"API: {args.api_url}")
print(f"Output: {results_dir}\n", flush=True)
sem = asyncio.Semaphore(args.concurrency)
async with httpx.AsyncClient() as client:
tasks = [
submit_and_process(client, pdf, i + 1, total, sem, results_dir, args.api_url, api_key)
for i, pdf in enumerate(pdfs)
]
results = await asyncio.gather(*tasks)
completed = [r for r in results if r["status"] == "completed"]
failed = [r for r in results if r["status"] != "completed"]
total_cost = sum(r.get("cost", 0) for r in completed)
total_tokens = sum(r.get("tokens", 0) for r in completed)
elapsed_list = [r["elapsed"] for r in results if "elapsed" in r]
print()
print("=" * 70)
print("BATCH PROCESSING SUMMARY")
print("=" * 70)
print(f" Documents: {total}")
print(f" Completed: {len(completed)} | Failed: {len(failed)}")
print(f" Success rate: {len(completed) / total * 100:.1f}%")
print(f" Total cost: ${total_cost:.4f}")
print(f" Total tokens: {total_tokens:,}")
if elapsed_list:
print(
f" Processing time — mean: {statistics.mean(elapsed_list):.0f}s, "
f"median: {statistics.median(elapsed_list):.0f}s, "
f"p95: {sorted(elapsed_list)[int(len(elapsed_list) * 0.95)]:.0f}s"
)
if completed:
costs_per_page = [
r["cost"] / r["pages"] for r in completed if isinstance(r.get("pages"), int) and r["pages"] > 0
]
if costs_per_page:
print(
f" Cost/page — mean: ${statistics.mean(costs_per_page):.4f}, "
f"median: ${statistics.median(costs_per_page):.4f}"
)
if failed:
print("\n Failures:")
for r in failed:
print(f" - {r['file']}: {r.get('error', '')}")
print(f"\n Results saved to: {results_dir}")
summary = {
"timestamp": stamp,
"api_url": args.api_url,
"concurrency": args.concurrency,
"total": total,
"completed": len(completed),
"failed": len(failed),
"total_cost": total_cost,
"total_tokens": total_tokens,
"results": results,
}
(results_dir / "summary.json").write_text(json.dumps(summary, indent=2, default=str))
if __name__ == "__main__":
asyncio.run(main())