my prefect server setup
prefect-metrics.waow.tech
python
orchestration
fix: use wrangler for CF Pages deploy instead of raw API
The site has Pages Functions (functions/ dir) that wrangler compiles into
a _worker.bundle. The raw Direct Upload API doesn't bundle functions,
so deploying without wrangler causes 500 errors on the live site.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
+37
-85
+37
-85
flows/atlas.py
+37
-85
flows/atlas.py
···
2
2
Rebuild the atlas (2D semantic map) and deploy to Cloudflare Pages.
3
3
4
4
Clones leaflet-search, runs the build-atlas script (UMAP + HDBSCAN),
5
-
then deploys the site to Cloudflare Pages via the Direct Upload API.
5
+
then deploys the site to Cloudflare Pages via wrangler.
6
6
7
7
Requires:
8
8
- Secret block "tpuf-token" (turbopuffer API key)
9
9
- Secret block "cloudflare-api-token" (Pages edit permission)
10
10
"""
11
11
12
-
import hashlib
13
-
import json
14
12
import os
15
13
import subprocess
16
14
import tempfile
17
15
from pathlib import Path
18
16
19
-
import httpx
20
17
from prefect import flow, get_run_logger, task
21
18
from prefect.blocks.system import Secret
22
19
23
20
REPO_URL = "https://github.com/zzstoatzz/leaflet-search.git"
24
-
CF_API = "https://api.cloudflare.com/client/v4"
25
-
CF_ACCOUNT_ID = "8feb33b5fb57ce2bc093bc6f4141f40a"
26
21
CF_PROJECT = "leaflet-search"
27
22
28
23
···
63
58
64
59
@task
65
60
def deploy_to_pages(site_dir: Path, api_token: str) -> str:
66
-
"""Deploy site/ to Cloudflare Pages via Direct Upload API."""
67
-
logger = get_run_logger()
68
-
headers = {"Authorization": f"Bearer {api_token}"}
61
+
"""Deploy site/ to Cloudflare Pages via wrangler.
69
62
70
-
# build manifest: path -> truncated SHA-256
71
-
manifest: dict[str, str] = {}
72
-
content_by_hash: dict[str, bytes] = {}
63
+
Uses wrangler because the site has Pages Functions (functions/ dir)
64
+
that must be compiled into a _worker.bundle. The raw Direct Upload API
65
+
doesn't handle function bundling, and deploying without it causes 500s.
66
+
"""
67
+
logger = get_run_logger()
68
+
env = {**os.environ, "CLOUDFLARE_API_TOKEN": api_token}
73
69
74
-
for path in sorted(site_dir.rglob("*")):
75
-
if path.is_dir() or path.name.startswith("."):
76
-
continue
77
-
content = path.read_bytes()
78
-
h = hashlib.sha256(content).hexdigest()[:32]
79
-
rel = "/" + str(path.relative_to(site_dir))
80
-
manifest[rel] = h
81
-
content_by_hash[h] = content
82
-
83
-
logger.info(f"deploying {len(manifest)} files")
84
-
85
-
# create deployment — CF expects multipart/form-data (-F fields in curl)
86
-
# httpx: files={(name, (None, value))} sends multipart form fields
87
-
resp = httpx.post(
88
-
f"{CF_API}/accounts/{CF_ACCOUNT_ID}/pages/projects/{CF_PROJECT}/deployments",
89
-
headers=headers,
90
-
files={
91
-
"manifest": (None, json.dumps(manifest), "application/json"),
92
-
"branch": (None, "main"),
93
-
},
94
-
timeout=60,
70
+
# install node + wrangler if not already available
71
+
subprocess.run(
72
+
["bash", "-c",
73
+
"command -v npx >/dev/null 2>&1 || "
74
+
"(apt-get update -qq && apt-get install -y -qq nodejs npm >/dev/null 2>&1)"],
75
+
env=env, capture_output=True, timeout=120,
95
76
)
96
-
if not resp.is_success:
97
-
logger.error(f"create deployment failed ({resp.status_code}): {resp.text[:500]}")
98
-
resp.raise_for_status()
99
-
body = resp.json()
100
-
deployment = body["result"]
101
-
logger.info(f"deployment {deployment['id']} created, keys: {list(deployment.keys())}")
102
-
jwt = deployment.get("jwt") or body.get("jwt")
103
-
if not jwt:
104
-
logger.error(f"no jwt in response: {json.dumps(body, indent=2)[:1000]}")
105
-
raise RuntimeError("no jwt in deployment response")
106
-
107
-
# check which files need uploading
108
-
jwt_headers = {"Authorization": f"Bearer {jwt}"}
109
-
resp = httpx.post(
110
-
f"{CF_API}/pages/assets/check-missing",
111
-
headers=jwt_headers,
112
-
json={"hashes": list(content_by_hash.keys())},
113
-
timeout=30,
77
+
subprocess.run(
78
+
["npm", "install", "--global", "wrangler"],
79
+
env=env, capture_output=True, text=True, timeout=120,
114
80
)
115
-
if not resp.is_success:
116
-
logger.error(f"check-missing failed ({resp.status_code}): {resp.text[:500]}")
117
-
resp.raise_for_status()
118
-
missing = set(resp.json())
119
-
logger.info(f"{len(missing)} files to upload ({len(manifest) - len(missing)} cached)")
120
81
121
-
# upload missing files
122
-
if missing:
123
-
batch: list[tuple[str, tuple[str, bytes, str]]] = []
124
-
for h in missing:
125
-
batch.append((h, ("blob", content_by_hash[h], "application/octet-stream")))
126
-
if len(batch) >= 50:
127
-
r = httpx.post(
128
-
f"{CF_API}/pages/assets/upload",
129
-
headers=jwt_headers,
130
-
files=batch,
131
-
timeout=120,
132
-
)
133
-
if not r.is_success:
134
-
logger.error(f"upload failed ({r.status_code}): {r.text[:500]}")
135
-
r.raise_for_status()
136
-
batch = []
137
-
if batch:
138
-
r = httpx.post(
139
-
f"{CF_API}/pages/assets/upload",
140
-
headers=jwt_headers,
141
-
files=batch,
142
-
timeout=120,
143
-
)
144
-
if not r.is_success:
145
-
logger.error(f"upload failed ({r.status_code}): {r.text[:500]}")
146
-
r.raise_for_status()
82
+
result = subprocess.run(
83
+
["wrangler", "pages", "deploy", ".",
84
+
f"--project-name={CF_PROJECT}", "--branch=main", "--commit-dirty=true"],
85
+
cwd=str(site_dir),
86
+
env=env,
87
+
capture_output=True,
88
+
text=True,
89
+
timeout=180,
90
+
)
91
+
for line in result.stdout.strip().splitlines():
92
+
logger.info(line)
93
+
if result.returncode != 0:
94
+
logger.error(result.stderr)
95
+
raise RuntimeError(f"wrangler deploy failed:\n{result.stderr}")
147
96
148
-
url = deployment.get("url", "")
149
-
logger.info(f"deployed: {url}")
150
-
return url
97
+
# extract deployment URL from wrangler output
98
+
for line in reversed(result.stdout.strip().splitlines()):
99
+
if "https://" in line:
100
+
url = line.split("https://", 1)[1].split()[0]
101
+
return f"https://{url}"
102
+
return ""
151
103
152
104
153
105
@flow(name="rebuild-atlas", log_prints=True)