fix: make CORS reconciliation recoverable

Authored-By: @architect <architect@fritzlab.net>

README.md

@@ -52,6 +52,7 @@ artifacts:
   - name: distributions
     type: static
     content_dir: dist
+    cors_origins: ['*']
     publish:
       bucket: baseline-dist
       credentials:
@@ -133,12 +134,16 @@ any route's mutable objects change.
 Mutable default and override partitions receive their final cache policy before
 the matching prefix-scoped stale deletion, so publication never exposes a
 provisional cache policy or a pointer to a missing immutable target.
-An append-only `site-publish-history.json` beside each generated site retains every bucket's access
-class and absolute immutable prefixes, including removed routes and rules. When a route move places
-a retired prefix inside the new sync scope, that subtree is excluded; a current-file collision
-fails publication. Protected access remains sticky across artifact renames and legacy mode, so
-publishing the same artifact publicly requires a new bucket. Decommissioning removes the live
-application and manifests while retaining this history because its Garage bucket is not purged.
+Generated Ingress annotations and `site-publish-history.yaml` retain every
+seen bucket's access, prior route, and cumulative bucket-relative immutable key
+prefixes, including while an artifact is absent. Removed or renamed rules stay
+recorded. When a move places a retired prefix inside the new sync scope, its
+immutable subtrees are excluded; a current-file collision fails publication.
+The bucket-keyed history rejects a protected-to-public transition even when the
+artifact is renamed; publishing that artifact publicly requires a new bucket.
+Legacy single-surface is public for this downgrade check. Removing or renaming
+an immutable rule preserves its prior URLs; current mutable content at one of
+those paths is rejected instead of replacing it.
 
 Artifact input directories must be pairwise disjoint after filesystem
 resolution. Publication stops before build or upload if one contains another or
@@ -147,6 +152,12 @@ protected input from entering a public artifact through dereference. Split
 storage endpoints are pinned to Garage, and each website
 authority is derived from its bucket; a site cannot expose an arbitrary backend.
 
+`cors_origins` is reconciled as a bucket policy on every split publication. Values are either `*`
+or HTTPS origins; browser access is limited to `GET` and `HEAD`. Omitting the field removes stale
+CORS from that bucket. Protected artifacts cannot allow wildcard CORS. All immutable objects and
+all bucket CORS policies complete before mutable channels change; if any policy write fails, the
+policies already attempted are restored to their pre-publication values.
+
 Each split route gets a bucket-specific `<bucket>.web.sjc001.fritzlab.net`
 ExternalName Service annotated to disable pass-host-header and a separate Ingress. Route
 Ingresses share the hostname's certificate Secret. The access middleware and

scripts/deploy.py

@@ -8,10 +8,13 @@ import os
 import re
 import shutil
 import subprocess
-from pathlib import Path, PurePosixPath
+import tempfile
+from pathlib import Path
 from urllib.error import HTTPError, URLError
 from urllib.request import Request, urlopen
 
+import yaml
+
 from utils import (
     NAMESPACE,
     clone_apps,
@@ -28,7 +31,6 @@ from utils import (
 GARAGE_ADMIN_ENDPOINT = os.environ.get(
     "GARAGE_ADMIN_ENDPOINT", "http://garage.storage.svc:3903"
 )
-HISTORY_FILE = "site-publish-history.json"
 
 
 def validate_artifact_output(site_dir, artifact):
@@ -168,6 +170,98 @@ def publication_aws_env(artifact, credential_env_names=None):
     return aws_env
 
 
+def configure_cors(bucket, origins, endpoint, aws_env):
+    """Reconcile read-only browser access without exposing publication credentials."""
+    if origins is None:
+        return
+    config = None
+    if origins:
+        config = {
+            "CORSRules": [{
+                "AllowedOrigins": origins,
+                "AllowedMethods": ["GET", "HEAD"],
+                "AllowedHeaders": ["*"],
+                "ExposeHeaders": ["ETag"],
+                "MaxAgeSeconds": 3600,
+            }],
+        }
+    set_cors_configuration(bucket, config, endpoint, aws_env)
+
+
+def set_cors_configuration(bucket, config, endpoint, aws_env):
+    """Apply an exact bucket CORS configuration, or remove it when absent."""
+    if config is None:
+        run([
+            "aws", "--endpoint-url", endpoint, "s3api", "delete-bucket-cors",
+            "--bucket", bucket,
+        ], env=aws_env)
+        return
+    with tempfile.NamedTemporaryFile("w", suffix=".json", encoding="utf-8") as handle:
+        json.dump(config, handle)
+        handle.flush()
+        run([
+            "aws", "--endpoint-url", endpoint, "s3api", "put-bucket-cors",
+            "--bucket", bucket, "--cors-configuration", f"file://{handle.name}",
+        ], env=aws_env)
+
+
+def get_cors_configuration(bucket, endpoint, aws_env):
+    """Read the exact bucket CORS configuration for rollback."""
+    result = _aws_capture([
+        "aws", "--endpoint-url", endpoint, "s3api", "get-bucket-cors",
+        "--bucket", bucket, "--output", "json",
+    ], aws_env)
+    if result.returncode == 0:
+        try:
+            config = json.loads(result.stdout)
+        except json.JSONDecodeError as error:
+            raise RuntimeError(f"get-bucket-cors returned invalid JSON for {bucket}") from error
+        if not isinstance(config, dict) or not isinstance(config.get("CORSRules"), list):
+            raise RuntimeError(f"get-bucket-cors returned an invalid policy for {bucket}")
+        return config
+    error = f"{result.stdout}\n{result.stderr}"
+    if "NoSuchCORSConfiguration" in error:
+        return None
+    raise RuntimeError(f"get-bucket-cors failed for {bucket}: {error.strip()}")
+
+
+def reconcile_artifact_cors(artifacts, credential_env_names=None):
+    """Reconcile all policies, restoring the prior set if any write fails."""
+    snapshots = []
+    for artifact in artifacts:
+        if artifact["cors_origins"] is None:
+            continue
+        aws_env = publication_aws_env(artifact, credential_env_names)
+        snapshots.append((
+            artifact,
+            aws_env,
+            get_cors_configuration(artifact["bucket"], artifact["s3_endpoint"], aws_env),
+        ))
+
+    attempted = []
+    try:
+        for artifact, aws_env, previous in snapshots:
+            attempted.append((artifact, aws_env, previous))
+            configure_cors(
+                artifact["bucket"], artifact["cors_origins"], artifact["s3_endpoint"], aws_env,
+            )
+    except Exception as error:
+        rollback_errors = []
+        for artifact, aws_env, previous in reversed(attempted):
+            try:
+                set_cors_configuration(
+                    artifact["bucket"], previous, artifact["s3_endpoint"], aws_env,
+                )
+            except Exception as rollback_error:
+                rollback_errors.append(f"{artifact['bucket']}: {rollback_error}")
+        if rollback_errors:
+            raise RuntimeError(
+                f"CORS reconciliation failed ({error}); rollback also failed for "
+                f"{'; '.join(rollback_errors)}"
+            ) from error
+        raise
+
+
 def publish_route_immutables(artifact, route, site_dir, credential_env_names=None):
     """Publish one route's immutable partitions during the global preflight."""
     html_dir = site_dir / artifact["build_dir"]
@@ -177,39 +271,47 @@ def publish_route_immutables(artifact, route, site_dir, credential_env_names=Non
             publish_immutable_rule(artifact, route, rule, html_dir, aws_env)
 
 
+def immutable_key_prefixes(artifact, route):
+    """Return immutable partitions as bucket-relative key prefixes."""
+    route_prefix = route["path"].strip("/")
+    return [
+        "/".join(part for part in (route_prefix, rule["path"]) if part)
+        for rule in artifact["cache_rules"] if _is_immutable(rule)
+    ]
+
+
+def retained_immutable_paths(artifact, route, previous_contract):
+    """Carry all bucket history forward so later route moves cannot delete it."""
+    previous_paths = previous_contract["immutable_paths"] if previous_contract else []
+    return sorted(set(previous_paths) | set(immutable_key_prefixes(artifact, route)))
+
+
 def retired_immutable_filters(artifact, route, html_dir, previous_contract):
-    """Protect every historical immutable prefix inside the current sync scope."""
+    """Protect historical immutable keys that fall inside the current sync scope."""
     if not previous_contract:
         return []
     current_prefix = route["path"].strip("/")
     filters = []
-    current_immutable = set(immutable_prefixes(artifact, route))
-    for immutable_prefix in previous_contract["immutable_prefixes"]:
+    current_immutable = set(immutable_key_prefixes(artifact, route))
+    for immutable_path in previous_contract["immutable_paths"]:
+        if immutable_path in current_immutable:
+            continue
         if current_prefix:
             marker = f"{current_prefix}/"
-            if not immutable_prefix.startswith(marker):
+            if not immutable_path.startswith(marker):
                 continue
-            relative_path = immutable_prefix[len(marker):]
+            retired_path = immutable_path[len(marker):]
         else:
-            relative_path = immutable_prefix
-        collision = html_dir / relative_path
-        if (immutable_prefix not in current_immutable and collision.exists()
-                and any(path.is_file() for path in collision.rglob("*"))):
+            retired_path = immutable_path
+        collision = html_dir / retired_path
+        if collision.exists() and any(path.is_file() for path in collision.rglob("*")):
             raise RuntimeError(
-                f"current artifact collides with retired immutable partition: {relative_path}"
+                f"current artifact collides with retired immutable partition: {retired_path}"
             )
-        filters.extend(("--exclude", f"{relative_path}/*"))
+        filters.extend(("--exclude", f"{retired_path}/*"))
     return filters
 
 
-def immutable_prefixes(artifact, route):
-    route_prefix = route["path"].strip("/")
-    return sorted({
-        "/".join(part for part in (route_prefix, rule["path"]) if part)
-        for rule in artifact["cache_rules"] if _is_immutable(rule)
-    })
-
-
 def s3_sync(artifact, route, site_dir, credential_env_names=None, previous_contract=None):
     endpoint = artifact["s3_endpoint"]
     html_dir = site_dir / artifact["build_dir"]
@@ -310,16 +412,32 @@ def ensure_bucket_aliases(site_name, aliases, admin_token):
             raise
 
 
-def render_site_manifests(site_name, action_dir, app_dir, manifests_dir, cfg):
+def render_site_manifests(
+    site_name, action_dir, app_dir, manifests_dir, cfg, previous_contracts=None,
+):
     """Always re-render manifests from current site.yaml. Templates own
     domain + aliases, so changes propagate without manual edits."""
     manifests_dir.mkdir(parents=True, exist_ok=True)
     artifact_by_name = {artifact["name"]: artifact for artifact in cfg["artifacts"]}
     routes = []
+    previous_contracts = previous_contracts or {}
+    next_contracts = {bucket: dict(contract)
+                      for bucket, contract in previous_contracts.items()}
     for route in cfg["routes"]:
         artifact = artifact_by_name[route["artifact"]]
         resource_name = k8s_name(site_name) if cfg["compatibility"] else f"{k8s_name(site_name)}-{route['name']}"
-        routes.append({**route, "resource_name": resource_name, "artifact_config": artifact})
+        previous = previous_contracts.get(artifact["bucket"])
+        immutable_paths = retained_immutable_paths(artifact, route, previous)
+        next_contracts[artifact["bucket"]] = {
+            "path": route["path"],
+            "access": "public" if route["access"] == "legacy" else route["access"],
+            "artifact": route["artifact"],
+            "immutable_paths": immutable_paths,
+        }
+        routes.append({
+            **route, "resource_name": resource_name, "artifact_config": artifact,
+            "immutable_paths_json": json.dumps(immutable_paths, separators=(",", ":")),
+        })
     template_vars = {
         "site": site_name,
         "site_k8s": k8s_name(site_name),
@@ -330,87 +448,77 @@ def render_site_manifests(site_name, action_dir, app_dir, manifests_dir, cfg):
         "routes": routes,
     }
     render_templates(action_dir, template_vars, app_dir, manifests_dir)
+    if not cfg["compatibility"] or previous_contracts:
+        (app_dir / "site-publish-history.yaml").write_text(yaml.safe_dump(
+            {"version": 1, "buckets": next_contracts}, sort_keys=True,
+        ))
+
+
+def _validate_route_contract(bucket, contract, path):
+    expected = {"path", "access", "artifact", "immutable_paths"}
+    if (not isinstance(bucket, str) or not bucket or not isinstance(contract, dict)
+            or set(contract) != expected
+            or contract.get("access") not in {"public", "protected"}
+            or not isinstance(contract.get("path"), str)
+            or not contract["path"].startswith("/")
+            or not isinstance(contract.get("artifact"), str) or not contract["artifact"]
+            or not isinstance(contract.get("immutable_paths"), list)
+            or any(not isinstance(item, str) or not item or item.startswith("/")
+                   or any(part in {"", ".", ".."} for part in item.split("/"))
+                   for item in contract["immutable_paths"])):
+        raise RuntimeError(f"invalid site-publish route history in {path}")
+    return {
+        "path": contract["path"], "access": contract["access"],
+        "artifact": contract["artifact"],
+        "immutable_paths": sorted(set(contract["immutable_paths"])),
+    }
 
 
 def previous_route_contracts(app_dir):
-    """Read the append-only bucket history kept beside generated manifests."""
-    path = app_dir / HISTORY_FILE
-    if not path.exists():
-        return {}
-    try:
-        document = json.loads(path.read_text())
-    except (OSError, json.JSONDecodeError) as exc:
-        raise RuntimeError(f"invalid site-publish route history in {path}") from exc
-    if (not isinstance(document, dict) or set(document) != {"schemaVersion", "buckets"}
-            or document["schemaVersion"] != 1 or not isinstance(document["buckets"], dict)):
-        raise RuntimeError(f"invalid site-publish route history in {path}")
-    for bucket, contract in document["buckets"].items():
-        if (not isinstance(bucket, str) or not isinstance(contract, dict)
-                or set(contract) != {"access", "artifact", "immutablePrefixes", "routePath"}
-                or contract["access"] not in {"legacy", "protected", "public"}
-                or not isinstance(contract["artifact"], str)
-                or not isinstance(contract["routePath"], str)
-                or not contract["routePath"].startswith("/")
-                or not isinstance(contract["immutablePrefixes"], list)
-                or any(not _valid_immutable_prefix(value)
-                       for value in contract["immutablePrefixes"])
-                or len(contract["immutablePrefixes"]) != len(set(contract["immutablePrefixes"]))):
+    """Read bucket-keyed route history from generated Ingresses."""
+    history_path = app_dir / "site-publish-history.yaml"
+    if history_path.exists():
+        document = yaml.safe_load(history_path.read_text())
+        if (not isinstance(document, dict) or set(document) != {"version", "buckets"}
+                or document["version"] != 1 or not isinstance(document["buckets"], dict)):
+            raise RuntimeError(f"invalid site-publish route history in {history_path}")
+        return {
+            bucket: _validate_route_contract(bucket, contract, history_path)
+            for bucket, contract in document["buckets"].items()
+        }
+    contracts = {}
+    manifests = app_dir / "manifests"
+    if not manifests.exists():
+        return contracts
+    for path in sorted(manifests.glob("ingress*.yaml")):
+        document = yaml.safe_load(path.read_text()) or {}
+        annotations = document.get("metadata", {}).get("annotations", {})
+        artifact = annotations.get("site-publish.fritzlab.net/artifact")
+        access = annotations.get("site-publish.fritzlab.net/access")
+        bucket = annotations.get("site-publish.fritzlab.net/bucket")
+        immutable_paths_json = annotations.get("site-publish.fritzlab.net/immutable-paths")
+        route_path = annotations.get("site-publish.fritzlab.net/route-path")
+        values = (artifact, access, bucket, immutable_paths_json, route_path)
+        if all(value is None for value in values):
+            continue
+        if (not all(isinstance(value, str) for value in values)
+                or access not in {"public", "protected"} or not route_path.startswith("/")):
             raise RuntimeError(f"invalid site-publish route history in {path}")
-    return {
-        bucket: {
-            "access": contract["access"],
-            "artifact": contract["artifact"],
-            "immutable_prefixes": sorted(contract["immutablePrefixes"]),
-            "path": contract["routePath"],
-        }
-        for bucket, contract in document["buckets"].items()
-    }
-
-
-def next_route_contracts(cfg, previous_contracts):
-    """Carry protected access and immutable prefixes forward for every known bucket."""
-    contracts = json.loads(json.dumps(previous_contracts))
-    artifacts = {artifact["name"]: artifact for artifact in cfg["artifacts"]}
-    for route in cfg["routes"]:
-        artifact = artifacts[route["artifact"]]
-        bucket = artifact["bucket"]
-        previous = previous_contracts.get(bucket)
-        access = "protected" if (
-            route["access"] == "protected" or previous and previous["access"] == "protected"
-        ) else route["access"]
-        contracts[bucket] = {
-            "access": access,
-            "artifact": route["artifact"],
-            "immutable_prefixes": sorted(set(
-                (previous or {}).get("immutable_prefixes", []) + immutable_prefixes(artifact, route)
-            )),
-            "path": route["path"],
-        }
+        try:
+            immutable_paths = json.loads(immutable_paths_json)
+        except json.JSONDecodeError as exc:
+            raise RuntimeError(f"invalid site-publish route history in {path}") from exc
+        if not isinstance(immutable_paths, list) or any(not isinstance(item, str) for item in immutable_paths):
+            raise RuntimeError(f"invalid site-publish route history in {path}")
+        if bucket in contracts:
+            raise RuntimeError(f"duplicate site-publish route history for bucket {bucket}")
+        contracts[bucket] = _validate_route_contract(bucket, {
+            "path": route_path, "access": access, "artifact": artifact,
+            "immutable_paths": immutable_paths,
+        }, path)
     return contracts
 
 
-def write_route_contracts(app_dir, contracts):
-    app_dir.mkdir(parents=True, exist_ok=True)
-    document = {
-        "schemaVersion": 1,
-        "buckets": {
-            bucket: {
-                "access": contract["access"],
-                "artifact": contract["artifact"],
-                "immutablePrefixes": contract["immutable_prefixes"],
-                "routePath": contract["path"],
-            }
-            for bucket, contract in sorted(contracts.items())
-        },
-    }
-    (app_dir / HISTORY_FILE).write_text(f"{json.dumps(document, indent=2, sort_keys=True)}\n")
-
-
-def _valid_immutable_prefix(value):
-    return (isinstance(value, str) and value and not value.startswith("/")
-            and not value.endswith("/") and ".." not in PurePosixPath(value).parts)
-
-
 def validate_route_migrations(cfg, previous_contracts):
     artifacts = {artifact["name"]: artifact for artifact in cfg["artifacts"]}
     for route in cfg["routes"]:
@@ -445,6 +553,10 @@ def deploy_static(site_name, site_dir, action_dir, token, cfg):
         publish_route_immutables(
             artifact_by_name[route["artifact"]], route, site_dir, credential_env_names,
         )
+    # Reconcile every browser-read policy before publishing mutable content.
+    # A CORS failure therefore cannot leave a new channel pointing at a release
+    # whose cross-origin assets browsers cannot consume.
+    reconcile_artifact_cors(cfg["artifacts"], credential_env_names)
     for route in cfg["routes"]:
         s3_sync(
             artifact_by_name[route["artifact"]], route, site_dir, credential_env_names,
@@ -453,8 +565,9 @@ def deploy_static(site_name, site_dir, action_dir, token, cfg):
     if cfg["compatibility"]:
         ensure_bucket_aliases(site_name, cfg["aliases"], os.environ.get("GARAGE_ADMIN_TOKEN"))
 
-    write_route_contracts(app_dir, next_route_contracts(cfg, previous_contracts))
-    render_site_manifests(site_name, action_dir, app_dir, manifests_dir, cfg)
+    render_site_manifests(
+        site_name, action_dir, app_dir, manifests_dir, cfg, previous_contracts,
+    )
 
     commit_and_push(apps_dir, f"Deploy {site_name}", token)
 
@@ -466,12 +579,7 @@ def decommission(site_name, token, buckets=None):
     if not site_path.exists():
         print(f"No manifests for {site_name} — nothing to remove")
         return
-    history_path = site_path / HISTORY_FILE
-    history = history_path.read_bytes() if history_path.exists() else None
     shutil.rmtree(site_path)
-    if history is not None:
-        site_path.mkdir(parents=True)
-        (site_path / HISTORY_FILE).write_bytes(history)
     commit_and_push(apps_dir, f"Decommission {site_name}", token)
     for bucket in buckets or [site_name]:
         print(f"Bucket {bucket} and its objects are NOT purged automatically.")

scripts/utils.py

@@ -1,5 +1,6 @@
 """Shared utilities for the site-publish action."""
 
+import ipaddress
 import os
 import re
 import shutil
@@ -104,6 +105,55 @@ def _strings(value, label):
     return values
 
 
+def _cors_origins(value, label):
+    origins = _list(value, label)
+    if any(not isinstance(item, str) or not item for item in origins):
+        raise ConfigError(f"{label} must be a list of non-empty strings")
+    canonical = []
+    for origin in origins:
+        if origin == "*":
+            canonical.append(origin)
+            continue
+        parsed = urlparse(origin)
+        try:
+            port = parsed.port
+        except ValueError:
+            raise ConfigError(f"{label} must contain '*' or canonical HTTPS origins") from None
+        if parsed.scheme != "https" or not parsed.hostname or parsed.path or parsed.params or (
+            parsed.query or parsed.fragment or parsed.username or parsed.password
+        ):
+            raise ConfigError(f"{label} must contain '*' or canonical HTTPS origins")
+        try:
+            address = ipaddress.ip_address(parsed.hostname)
+        except ValueError:
+            try:
+                hostname = parsed.hostname.encode("idna").decode("ascii")
+            except UnicodeError:
+                raise ConfigError(
+                    f"{label} must contain '*' or canonical HTTPS origins"
+                ) from None
+            try:
+                _hostname(hostname, f"{label} hostname")
+            except ConfigError:
+                raise ConfigError(
+                    f"{label} must contain '*' or canonical HTTPS origins"
+                ) from None
+        else:
+            if getattr(address, "scope_id", None) is not None:
+                raise ConfigError(
+                    f"{label} must contain '*' or canonical HTTPS origins"
+                )
+            hostname = f"[{address.compressed}]" if address.version == 6 else address.compressed
+        canonical.append(f"https://{hostname}{f':{port}' if port not in (None, 443) else ''}")
+    if len(canonical) != len(set(canonical)):
+        raise ConfigError(f"{label} must not contain duplicate canonical origins")
+    if "*" in canonical and len(canonical) != 1:
+        raise ConfigError(f"{label} wildcard must be the only origin")
+    if origins != canonical:
+        raise ConfigError(f"{label} must contain '*' or canonical HTTPS origins")
+    return origins
+
+
 def _hostname(value, label):
     if not isinstance(value, str) or len(value) > 253 or value.endswith("."):
         raise ConfigError(f"{label} must be a lowercase DNS hostname without a trailing dot")
@@ -215,6 +265,7 @@ def _legacy_config(raw, site_name):
         "website_authority": "garage-s3.storage.svc.k8s.sjc001.fritzlab.net",
         "credentials": {"access_key_env": "AWS_ACCESS_KEY_ID", "secret_key_env": "AWS_SECRET_ACCESS_KEY"},
         "cache_rules": [{"path": "", "cache_control": DEFAULT_CACHE_CONTROL}],
+        "cors_origins": None,
     }
     return {
         "version": 1,
@@ -234,7 +285,11 @@ def _legacy_config(raw, site_name):
 def _artifact(item, index):
     label = f"artifacts[{index}]"
     item = _mapping(item, label)
-    _known_keys(item, {"name", "type", "content_dir", "tidy", "excludes", "publish", "cache"}, label)
+    _known_keys(
+        item,
+        {"name", "type", "content_dir", "tidy", "excludes", "publish", "cache", "cors_origins"},
+        label,
+    )
     name = item.get("name")
     if not isinstance(name, str) or not NAME_RE.fullmatch(name):
         raise ConfigError(f"{label}.name must be a DNS label")
@@ -303,6 +358,10 @@ def _artifact(item, index):
         "website_authority": authority,
         "credentials": normalized_credentials,
         "cache_rules": cache_rules,
+        "cors_origins": (
+            _cors_origins(item["cors_origins"], f"{label}.cors_origins")
+            if "cors_origins" in item else []
+        ),
     }
 
 
@@ -389,6 +448,10 @@ def _validate_multi(cfg):
                 raise ConfigError(f"protected route {route['name']} cannot use shared-cache max-age")
             if route["access"] == "public" and "private" in directives:
                 raise ConfigError(f"public route {route['name']} cannot use private cache policy")
+        if route["access"] == "protected" and "*" in artifact["cors_origins"]:
+            raise ConfigError(
+                f"protected route {route['name']} cannot allow wildcard CORS"
+            )
     root = next(route for route in routes if route["path"] == "/")
     if any(route["access"] == "protected" for route in routes) and root["access"] == "public":
         raise ConfigError("a public '/' catch-all would expose unmatched protected content")

templates/ingress.yaml.j2

@@ -4,6 +4,13 @@ metadata:
   name: {{ route.resource_name }}
   namespace: {{ namespace }}
   annotations:
+{%- if not compatibility %}
+    site-publish.fritzlab.net/artifact: {{ route.artifact | tojson }}
+    site-publish.fritzlab.net/access: {{ route.access | tojson }}
+    site-publish.fritzlab.net/bucket: {{ route.artifact_config.bucket | tojson }}
+    site-publish.fritzlab.net/immutable-paths: {{ route.immutable_paths_json | tojson }}
+    site-publish.fritzlab.net/route-path: {{ route.path | tojson }}
+{%- endif %}
     traefik.ingress.kubernetes.io/router.middlewares: https-redirect@file,retry-upstream@file{% if route.access_middleware %},{{ route.access_middleware }}@file{% endif %}{% for m in route.middlewares %},{{ m }}@file{% endfor %}
 spec:
   ingressClassName: traefik

tests/fixtures/split-site.yaml

@@ -15,6 +15,7 @@ artifacts:
   - name: distributions
     type: static
     content_dir: dist
+    cors_origins: ['*']
     publish:
       bucket: baseline-dist
       credentials:

tests/test_contract.py

@@ -113,6 +113,69 @@ class ConfigContractTests(unittest.TestCase):
         self.assertEqual(["/dist", "/"], [route["path"] for route in cfg["routes"]])
         credentials = {artifact["name"]: artifact["credentials"] for artifact in cfg["artifacts"]}
         self.assertNotEqual(credentials["distributions"], credentials["portal"])
+        distributions = next(
+            artifact for artifact in cfg["artifacts"] if artifact["name"] == "distributions"
+        )
+        self.assertEqual(["*"], distributions["cors_origins"])
+
+    def test_cors_origins_are_https_origins_or_wildcard(self):
+        self.assert_invalid(
+            lambda raw: raw["artifacts"][1].__setitem__(
+                "cors_origins", ["http://consumer.example"]
+            ),
+            "must contain '\\*' or canonical HTTPS origins",
+        )
+        for origin in (
+            "https://consumer example",
+            "https://Consumer.example",
+            "https://consumer.example:443",
+            "https://consumer.example/",
+            "https://consumer.example:invalid",
+            "https://[fe80::1%eth0]",
+            "https://[fe80::1%25eth0]",
+        ):
+            with self.subTest(origin=origin):
+                self.assert_invalid(
+                    lambda raw, origin=origin: raw["artifacts"][1].__setitem__(
+                        "cors_origins", [origin]
+                    ),
+                    "must contain '\\*' or canonical HTTPS origins",
+                )
+        self.assert_invalid(
+            lambda raw: raw["artifacts"][1].__setitem__(
+                "cors_origins", ["https://consumer.example", "https://consumer.example:443"]
+            ),
+            "must not contain duplicate canonical origins",
+        )
+        self.assert_invalid(
+            lambda raw: raw["artifacts"][1].__setitem__(
+                "cors_origins", ["*", "https://consumer.example"]
+            ),
+            "wildcard must be the only origin",
+        )
+        raw = copy.deepcopy(self.raw)
+        raw["artifacts"][1]["cors_origins"] = ["https://[2602:817:3000::1]:8443"]
+        cfg = normalize_site_config(raw, "baseline.fritzlab.net")
+        self.assertEqual(
+            ["https://[2602:817:3000::1]:8443"],
+            next(a for a in cfg["artifacts"] if a["name"] == "distributions")["cors_origins"],
+        )
+
+    def test_declared_cors_origins_must_be_a_list(self):
+        for value in (None, False, 0, "", {}):
+            with self.subTest(value=value):
+                self.assert_invalid(
+                    lambda raw, value=value: raw["artifacts"][1].__setitem__(
+                        "cors_origins", value
+                    ),
+                    "cors_origins must be a list",
+                )
+
+    def test_protected_route_rejects_wildcard_cors(self):
+        self.assert_invalid(
+            lambda raw: raw["artifacts"][0].__setitem__("cors_origins", ["*"]),
+            "protected route portal cannot allow wildcard CORS",
+        )
 
     def test_equivalent_route_paths_are_ambiguous(self):
         self.assert_invalid(lambda raw: raw["routes"].append({
@@ -131,6 +194,7 @@ class ConfigContractTests(unittest.TestCase):
             raw["routes"][1]["access"] = {
                 "mode": "protected", "middleware": "authentik-forwardauth"
             }
+            raw["artifacts"][1]["cors_origins"] = []
             raw["artifacts"][0]["cache"]["rules"][0]["cache_control"] = (
                 "public, max-age=0, must-revalidate"
             )
@@ -263,6 +327,7 @@ class GenerationTests(unittest.TestCase):
             "app.yaml", "manifests/certificate.yaml", "manifests/ingress-distributions.yaml",
             "manifests/ingress-portal.yaml", "manifests/kustomization.yaml",
             "manifests/service-distributions.yaml", "manifests/service-portal.yaml",
+            "site-publish-history.yaml",
         }, set(files))
         for content in files.values():
             self.assertIsNotNone(yaml.safe_load(content))
@@ -273,17 +338,16 @@ class GenerationTests(unittest.TestCase):
         self.assertNotIn("passhostheader", files["manifests/ingress-portal.yaml"])
         self.assertIn("baseline-dist.web.sjc001.fritzlab.net", files["manifests/service-distributions.yaml"])
 
-    def test_history_keeps_yaml_ambiguous_artifact_names_as_strings(self):
+    def test_yaml_ambiguous_artifact_name_stays_a_string_annotation(self):
         raw = fixture("split-site.yaml")
         raw["artifacts"][0]["name"] = "yes"
         raw["routes"][0]["artifact"] = "yes"
-        cfg = normalize_site_config(raw, "baseline.fritzlab.net")
-        contracts = deploy.next_route_contracts(cfg, {})
-        with tempfile.TemporaryDirectory() as tmp:
-            app_dir = Path(tmp)
-            deploy.write_route_contracts(app_dir, contracts)
-            restored = deploy.previous_route_contracts(app_dir)
-        self.assertEqual("yes", restored["baseline-portal"]["artifact"])
+        tmp, _, files = self.render(raw)
+        self.addCleanup(tmp.cleanup)
+        ingress = yaml.safe_load(files["manifests/ingress-portal.yaml"])
+        self.assertEqual(
+            "yes", ingress["metadata"]["annotations"]["site-publish.fritzlab.net/artifact"],
+        )
 
     def test_generation_is_deterministic_when_input_lists_are_reversed(self):
         raw = fixture("split-site.yaml")
@@ -341,6 +405,75 @@ class BuildTests(unittest.TestCase):
 
 
 class PublishingTests(unittest.TestCase):
+    def test_cors_policy_is_reconciled_as_read_only_browser_access(self):
+        captured = {}
+
+        def capture(command, **_kwargs):
+            config_path = command[command.index("--cors-configuration") + 1]
+            captured["command"] = command
+            captured["config"] = json.loads(Path(config_path.removeprefix("file://")).read_text())
+
+        with patch.object(deploy, "run", side_effect=capture):
+            deploy.configure_cors(
+                "baseline-dist", ["*"], "http://garage-s3.storage.svc:3900", {}
+            )
+
+        self.assertIn("put-bucket-cors", captured["command"])
+        self.assertEqual(["GET", "HEAD"], captured["config"]["CORSRules"][0]["AllowedMethods"])
+        self.assertEqual(["*"], captured["config"]["CORSRules"][0]["AllowedOrigins"])
+
+    def test_empty_cors_policy_removes_stale_bucket_cors(self):
+        with patch.object(deploy, "run") as request:
+            deploy.configure_cors(
+                "baseline-catalogue", [], "http://garage-s3.storage.svc:3900", {}
+            )
+        self.assertIn("delete-bucket-cors", request.call_args.args[0])
+
+    def test_cors_reconciliation_restores_prior_policies_on_failure(self):
+        artifacts = [
+            {
+                "bucket": "first", "cors_origins": ["https://new.example"],
+                "s3_endpoint": "http://garage-s3.storage.svc:3900", "credentials": {},
+            },
+            {
+                "bucket": "second", "cors_origins": [],
+                "s3_endpoint": "http://garage-s3.storage.svc:3900", "credentials": {},
+            },
+        ]
+        prior = [
+            {"CORSRules": [{"AllowedOrigins": ["https://old.example"]}]},
+            None,
+        ]
+        writes = []
+
+        def write(bucket, config, *_args):
+            writes.append((bucket, config))
+            if bucket == "second" and len(writes) == 2:
+                raise RuntimeError("write failed")
+
+        with patch.object(deploy, "publication_aws_env", return_value={}), patch.object(
+            deploy, "get_cors_configuration", side_effect=prior,
+        ) as read, patch.object(deploy, "set_cors_configuration", side_effect=write), \
+                self.assertRaisesRegex(RuntimeError, "write failed"):
+            deploy.reconcile_artifact_cors(artifacts)
+
+        self.assertEqual(2, read.call_count)
+        self.assertEqual(
+            [
+                ("first", {"CORSRules": [{
+                    "AllowedOrigins": ["https://new.example"],
+                    "AllowedMethods": ["GET", "HEAD"],
+                    "AllowedHeaders": ["*"],
+                    "ExposeHeaders": ["ETag"],
+                    "MaxAgeSeconds": 3600,
+                }]}),
+                ("second", None),
+                ("second", None),
+                ("first", prior[0]),
+            ],
+            writes,
+        )
+
     def test_apps_clone_never_places_token_in_argv_or_log(self):
         calls = []
         secret = "clone-secret-must-not-appear"
@@ -474,7 +607,7 @@ class PublishingTests(unittest.TestCase):
             ):
                 deploy.s3_sync(artifact, route, root, previous_contract={
                     "path": "/foo", "access": "public", "artifact": "distributions",
-                    "immutable_prefixes": ["foo/releases"],
+                    "immutable_paths": ["foo/releases"],
                 })
             rendered = [" ".join(command) for command in commands]
             self.assertTrue(all("foo/releases/*" in command for command in rendered[:2]))
@@ -485,7 +618,7 @@ class PublishingTests(unittest.TestCase):
         previous = {
             "baseline-dist": {
                 "path": "/dist", "access": "protected", "artifact": "old-name",
-                "immutable_prefixes": ["dist/releases"],
+                "immutable_paths": ["dist/releases"],
             }
         }
         with self.assertRaisesRegex(RuntimeError, "cannot become public while reusing protected"):
@@ -505,7 +638,7 @@ class PublishingTests(unittest.TestCase):
         previous = {
             "baseline.fritzlab.net": {
                 "path": "/portal", "access": "protected", "artifact": "portal",
-                "immutable_prefixes": [],
+                "immutable_paths": [],
             }
         }
         with self.assertRaisesRegex(RuntimeError, "cannot become public while reusing protected"):
@@ -522,59 +655,119 @@ class PublishingTests(unittest.TestCase):
             html = Path(tmp)
             filters = deploy.retired_immutable_filters(artifact, route, html, {
                 "path": "/dist", "access": "public", "artifact": "distributions",
-                "immutable_prefixes": ["dist/releases"],
+                "immutable_paths": ["dist/releases"],
             })
             self.assertEqual(["--exclude", "releases/*"], filters)
-            previous = {
-                "baseline-dist": {
-                    "path": "/dist", "access": "public", "artifact": "distributions",
-                    "immutable_prefixes": ["dist/releases"],
-                }
-            }
-            next_contracts = deploy.next_route_contracts(cfg, previous)
-            self.assertEqual(
-                ["dist/releases"], next_contracts["baseline-dist"]["immutable_prefixes"],
-            )
-            self.assertEqual(
-                ["--exclude", "releases/*"],
-                deploy.retired_immutable_filters(
-                    artifact, route, html, next_contracts["baseline-dist"],
-                ),
-            )
             (html / "releases").mkdir()
             (html / "releases" / "replacement.js").write_text("mutable")
             with self.assertRaisesRegex(RuntimeError, "collides with retired immutable"):
                 deploy.retired_immutable_filters(artifact, route, html, {
                     "path": "/dist", "access": "public", "artifact": "distributions",
-                    "immutable_prefixes": ["dist/releases"],
+                    "immutable_paths": ["dist/releases"],
                 })
 
-    def test_removed_route_history_remains_append_only(self):
+    def test_removed_immutable_rule_remains_in_next_manifest_history(self):
         cfg = normalize_site_config(fixture("split-site.yaml"), "baseline.fritzlab.net")
+        artifact = next(item for item in cfg["artifacts"] if item["name"] == "distributions")
+        artifact["cache_rules"] = [
+            rule for rule in artifact["cache_rules"] if rule["path"] != "releases"
+        ]
         previous = {
-            "retired-bucket": {
-                "path": "/retired", "access": "protected", "artifact": "retired",
-                "immutable_prefixes": ["retired/releases"],
+            "baseline-dist": {
+                "path": "/dist", "access": "public", "artifact": "distributions",
+                "immutable_paths": ["dist/releases"],
             }
         }
-        contracts = deploy.next_route_contracts(cfg, previous)
-        self.assertEqual(previous["retired-bucket"], contracts["retired-bucket"])
-
-    def test_decommission_preserves_history_for_an_unpurged_bucket(self):
         with tempfile.TemporaryDirectory() as tmp:
-            apps = Path(tmp)
-            site = apps / "sjc001/websites/baseline"
-            site.mkdir(parents=True)
-            history = b'{"schemaVersion":1,"buckets":{}}\n'
-            (site / deploy.HISTORY_FILE).write_bytes(history)
-            (site / "app.yaml").write_text("live\n")
-            with patch.object(deploy, "clone_apps", return_value=apps), patch.object(
-                deploy, "commit_and_push"
-            ) as commit:
-                deploy.decommission("baseline", "token", ["baseline-dist"])
-            self.assertEqual(history, (site / deploy.HISTORY_FILE).read_bytes())
-            self.assertFalse((site / "app.yaml").exists())
-            commit.assert_called_once_with(apps, "Decommission baseline", "token")
+            app_dir = Path(tmp) / "app"
+            manifests = app_dir / "manifests"
+            deploy.render_site_manifests(
+                "baseline.fritzlab.net", ROOT, app_dir, manifests, cfg, previous,
+            )
+            first = deploy.previous_route_contracts(app_dir)
+            self.assertEqual(["dist/releases"], first["baseline-dist"]["immutable_paths"])
+            deploy.render_site_manifests(
+                "baseline.fritzlab.net", ROOT, app_dir, manifests, cfg, first,
+            )
+            second = deploy.previous_route_contracts(app_dir)
+            self.assertEqual(first, second)
+
+    def test_protected_split_bucket_survives_absence_and_blocks_legacy(self):
+        raw = fixture("split-site.yaml")
+        raw["artifacts"] = [
+            item for item in raw["artifacts"] if item["name"] == "distributions"
+        ]
+        raw["routes"] = [
+            item for item in raw["routes"] if item["artifact"] == "distributions"
+        ]
+        raw["routes"][0]["path"] = "/"
+        cfg = normalize_site_config(raw, "baseline.fritzlab.net")
+        previous = {
+            "baseline.fritzlab.net": {
+                "path": "/", "access": "protected", "artifact": "portal",
+                "immutable_paths": [],
+            }
+        }
+        with tempfile.TemporaryDirectory() as tmp:
+            app_dir = Path(tmp) / "app"
+            deploy.render_site_manifests(
+                "baseline.fritzlab.net", ROOT, app_dir, app_dir / "manifests", cfg, previous,
+            )
+            retained = deploy.previous_route_contracts(app_dir)
+            self.assertEqual(
+                previous["baseline.fritzlab.net"], retained["baseline.fritzlab.net"],
+            )
+            legacy = normalize_site_config(
+                fixture("legacy-site.yaml"), "baseline.fritzlab.net",
+            )
+            with self.assertRaisesRegex(RuntimeError, "cannot become public while reusing protected"):
+                deploy.validate_route_migrations(legacy, retained)
+
+    def test_malformed_persistent_history_fails_closed(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            app_dir = Path(tmp)
+            (app_dir / "site-publish-history.yaml").write_text(
+                "version: 1\nbuckets:\n  bucket:\n    path: /\n"
+                "    access: public\n    artifact: site\n"
+                "    immutable_paths: [../releases]\n"
+            )
+            with self.assertRaisesRegex(RuntimeError, "invalid site-publish route history"):
+                deploy.previous_route_contracts(app_dir)
+
+    def test_retired_immutable_history_survives_a_route_move(self):
+        cfg = normalize_site_config(fixture("split-site.yaml"), "baseline.fritzlab.net")
+        artifact = next(item for item in cfg["artifacts"] if item["name"] == "distributions")
+        artifact["cache_rules"] = [
+            rule for rule in artifact["cache_rules"] if rule["path"] != "releases"
+        ]
+        route = next(item for item in cfg["routes"] if item["artifact"] == "distributions")
+        route["path"] = "/"
+        previous = {
+            "baseline-dist": {
+                "path": "/foo", "access": "public", "artifact": "distributions",
+                "immutable_paths": ["foo/releases"],
+            }
+        }
+        with tempfile.TemporaryDirectory() as tmp:
+            root = Path(tmp)
+            app_dir = root / "app"
+            manifests = app_dir / "manifests"
+            app_dir.mkdir()
+            deploy.render_site_manifests(
+                "baseline.fritzlab.net", ROOT, app_dir, manifests, cfg, previous,
+            )
+            following = deploy.previous_route_contracts(app_dir)
+            self.assertEqual(["foo/releases"], following["baseline-dist"]["immutable_paths"])
+            html = root / "html"
+            html.mkdir()
+            first_filters = deploy.retired_immutable_filters(
+                artifact, route, html, previous["baseline-dist"],
+            )
+            following_filters = deploy.retired_immutable_filters(
+                artifact, route, html, following["baseline-dist"],
+            )
+            self.assertEqual(["--exclude", "foo/releases/*"], first_filters)
+            self.assertEqual(first_filters, following_filters)
 
     def test_later_route_immutable_failure_stops_all_mutable_publication(self):
         cfg = normalize_site_config(fixture("split-site.yaml"), "baseline.fritzlab.net")
@@ -585,15 +778,42 @@ class PublishingTests(unittest.TestCase):
                     patch.object(deploy, "clone_apps", return_value=root / "apps"), patch.object(
                         deploy, "publish_route_immutables",
                         side_effect=[None, RuntimeError("immutable failed")],
-                    ) as immutable_publish, patch.object(deploy, "s3_sync") as mutable_sync, \
+                    ) as immutable_publish, patch.object(
+                        deploy, "reconcile_artifact_cors"
+                    ) as cors_reconcile, patch.object(deploy, "s3_sync") as mutable_sync, \
                     self.assertRaisesRegex(
                 RuntimeError, "immutable failed"
             ):
                 deploy.deploy_static("baseline", root, root, "token", cfg)
 
             self.assertEqual(2, immutable_publish.call_count)
+            cors_reconcile.assert_not_called()
             mutable_sync.assert_not_called()
 
+    def test_all_cors_policies_complete_before_mutable_publication(self):
+        cfg = normalize_site_config(fixture("split-site.yaml"), "baseline.fritzlab.net")
+        events = []
+        with tempfile.TemporaryDirectory() as tmp:
+            root = Path(tmp)
+            apps = root / "apps"
+            apps.mkdir()
+            with patch.object(deploy, "validate_publication_environment"), patch.object(
+                deploy, "validate_artifact_output"
+            ), patch.object(
+                deploy, "publish_route_immutables", side_effect=lambda *_args: events.append("immutable")
+            ), patch.object(
+                deploy, "reconcile_artifact_cors", side_effect=lambda *_args: events.append("cors")
+            ), patch.object(
+                deploy, "s3_sync", side_effect=lambda *_args: events.append("mutable")
+            ), patch.object(deploy, "clone_apps", return_value=apps), patch.object(
+                deploy, "render_site_manifests"
+            ), patch.object(deploy, "commit_and_push"):
+                deploy.deploy_static("baseline", root, root, "token", cfg)
+
+        self.assertEqual(
+            ["immutable", "immutable", "cors", "mutable", "mutable"], events
+        )
+
 
     def test_absent_artifact_is_detected_before_publish(self):
         cfg = normalize_site_config(fixture("split-site.yaml"), "baseline.fritzlab.net")