Merge pull request #1536 from PolicyEngine/codex/enable-modal-custom-domain

Enable Modal custom domain for household API gateway

Author: anth-volk

.github/scripts/modal-custom-domain-smoke.sh

@@ -0,0 +1,58 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+modal_environment="${MODAL_ENVIRONMENT:-main}"
+modal_get_url_script="${MODAL_GET_URL_SCRIPT:-.github/scripts/modal-get-url.sh}"
+custom_domain_url="${HOUSEHOLD_MODAL_GATEWAY_CUSTOM_DOMAIN_URL:-https://household.api.policyengine.org}"
+
+if [ "${modal_environment}" != "main" ]; then
+  echo "Skipping custom-domain smoke check outside the main Modal environment."
+  exit 0
+fi
+
+gateway_url="$(bash "${modal_get_url_script}")"
+custom_domain_url="${custom_domain_url%/}"
+gateway_url="${gateway_url%/}"
+
+tmpdir="$(mktemp -d)"
+trap 'rm -rf "${tmpdir}"' EXIT
+
+gateway_versions="${tmpdir}/gateway-versions.json"
+custom_versions="${tmpdir}/custom-domain-versions.json"
+
+echo "Checking generated Modal gateway at ${gateway_url}"
+curl -fsS "${gateway_url}/liveness_check" >/dev/null
+curl -fsS "${gateway_url}/versions/us" > "${gateway_versions}"
+
+echo "Checking production custom domain at ${custom_domain_url}"
+curl -fsS "${custom_domain_url}/liveness_check" >/dev/null
+curl -fsS "${custom_domain_url}/versions/us" > "${custom_versions}"
+
+python - "${gateway_versions}" "${custom_versions}" <<'PY'
+import json
+import sys
+from pathlib import Path
+
+
+def load_versions(path: str, label: str) -> dict[str, str]:
+    try:
+        data = json.loads(Path(path).read_text())
+    except json.JSONDecodeError as e:
+        sys.exit(f"{label} /versions/us did not return JSON: {e}")
+
+    if not isinstance(data, dict):
+        sys.exit(f"{label} /versions/us returned a non-object JSON value")
+    return data
+
+
+gateway_versions = load_versions(sys.argv[1], "Generated Modal gateway")
+custom_versions = load_versions(sys.argv[2], "Custom domain")
+
+if gateway_versions != custom_versions:
+    sys.exit(
+        "Custom domain /versions/us does not match the generated Modal "
+        f"gateway. generated={gateway_versions!r} custom={custom_versions!r}"
+    )
+
+print("Custom domain points at the deployed Modal gateway.")
+PY

.github/scripts/test_modal_custom_domain_smoke.py

@@ -0,0 +1,141 @@
+import os
+from pathlib import Path
+import subprocess
+
+
+SCRIPT = ".github/scripts/modal-custom-domain-smoke.sh"
+
+
+def test_modal_custom_domain_smoke_passes_when_versions_match(tmp_path):
+    env = _smoke_env(
+        tmp_path,
+        gateway_versions='{"current":"1.691.1","frontier":"1.691.1"}',
+        custom_versions='{"current":"1.691.1","frontier":"1.691.1"}',
+    )
+
+    result = subprocess.run(
+        ["bash", SCRIPT],
+        capture_output=True,
+        env=env,
+        text=True,
+    )
+
+    assert result.returncode == 0, result.stderr
+    assert (
+        "Custom domain points at the deployed Modal gateway." in result.stdout
+    )
+
+
+def test_modal_custom_domain_smoke_fails_when_custom_domain_is_not_gateway(
+    tmp_path,
+):
+    env = _smoke_env(
+        tmp_path,
+        gateway_versions='{"current":"1.691.1","frontier":"1.691.1"}',
+        custom_versions="OK",
+    )
+
+    result = subprocess.run(
+        ["bash", SCRIPT],
+        capture_output=True,
+        env=env,
+        text=True,
+    )
+
+    assert result.returncode == 1
+    assert "Custom domain /versions/us did not return JSON" in result.stderr
+
+
+def test_modal_custom_domain_smoke_fails_when_versions_differ(tmp_path):
+    env = _smoke_env(
+        tmp_path,
+        gateway_versions='{"current":"1.691.1","frontier":"1.691.1"}',
+        custom_versions='{"current":"1.690.0","frontier":"1.691.1"}',
+    )
+
+    result = subprocess.run(
+        ["bash", SCRIPT],
+        capture_output=True,
+        env=env,
+        text=True,
+    )
+
+    assert result.returncode == 1
+    assert "does not match the generated Modal gateway" in result.stderr
+
+
+def test_modal_custom_domain_smoke_skips_non_main_environments(tmp_path):
+    curl_log = tmp_path / "curl.log"
+    env = _smoke_env(
+        tmp_path,
+        gateway_versions='{"current":"1.691.1"}',
+        custom_versions='{"current":"1.691.1"}',
+    )
+    env["MODAL_ENVIRONMENT"] = "staging"
+    env["CURL_LOG"] = str(curl_log)
+
+    result = subprocess.run(
+        ["bash", SCRIPT],
+        capture_output=True,
+        env=env,
+        text=True,
+    )
+
+    assert result.returncode == 0
+    assert "Skipping custom-domain smoke check" in result.stdout
+    assert not curl_log.exists()
+
+
+def _smoke_env(
+    tmp_path: Path,
+    *,
+    gateway_versions: str,
+    custom_versions: str,
+) -> dict[str, str]:
+    get_url_script = tmp_path / "modal-get-url.sh"
+    get_url_script.write_text(
+        "#!/usr/bin/env bash\n"
+        "set -euo pipefail\n"
+        "echo https://generated-modal.example\n"
+    )
+    get_url_script.chmod(0o755)
+
+    fake_curl = tmp_path / "curl"
+    fake_curl.write_text(
+        """#!/usr/bin/env bash
+set -euo pipefail
+url="${@: -1}"
+if [ -n "${CURL_LOG:-}" ]; then
+  printf '%s\\n' "${url}" >> "${CURL_LOG}"
+fi
+
+case "${url}" in
+  https://generated-modal.example/liveness_check|https://custom-domain.example/liveness_check)
+    echo OK
+    ;;
+  https://generated-modal.example/versions/us)
+    printf '%s\\n' "${GATEWAY_VERSIONS}"
+    ;;
+  https://custom-domain.example/versions/us)
+    printf '%s\\n' "${CUSTOM_VERSIONS}"
+    ;;
+  *)
+    echo "Unexpected URL: ${url}" >&2
+    exit 22
+    ;;
+esac
+"""
+    )
+    fake_curl.chmod(0o755)
+
+    return {
+        **os.environ,
+        "PATH": f"{tmp_path}:{os.environ['PATH']}",
+        "MODAL_ENVIRONMENT": "main",
+        "MODAL_GET_URL_SCRIPT": str(get_url_script),
+        "HOUSEHOLD_MODAL_GATEWAY_CUSTOM_DOMAIN_URL": (
+            "https://custom-domain.example"
+        ),
+        "GATEWAY_VERSIONS": gateway_versions,
+        "CUSTOM_VERSIONS": custom_versions,
+    }

.github/scripts/test_resolve_modal_release_config.py

@@ -62,6 +62,27 @@ def test_resolve_release_uses_workflow_dispatch_inputs():
     assert resolved.config.cleanup_target == "frontier"
 
 
+def test_resolve_release_accepts_both_workflow_dispatch_target():
+    resolved = resolve_release_from_event(
+        {
+            "inputs": {
+                "new_app_target": "both",
+                "promote_existing_frontier": "false",
+                "cleanup_target": "retired",
+            }
+        },
+        fetch_pr_body_for_commit=lambda _repository, _sha: None,
+        event_name="workflow_dispatch",
+    )
+
+    assert resolved.should_deploy is True
+    assert resolved.deploy_mode == "release"
+    assert resolved.config is not None
+    assert resolved.config.new_app_target == "both"
+    assert resolved.config.promote_existing_frontier is False
+    assert resolved.config.cleanup_target == "retired"
+
+
 def test_resolve_release_uses_weekly_default_for_empty_workflow_dispatch():
     resolved = resolve_release_from_event(
         {"inputs": {}},

.github/workflows/deploy-staged.yml

@@ -14,6 +14,7 @@ on:
         options:
           - frontier
           - current
+          - both
           - none
       promote_existing_frontier:
         description: "Move the existing frontier worker to current before deploying a new frontier"
@@ -265,6 +266,9 @@ jobs:
             '${{ needs.resolve-modal-release-config.outputs.config_json }}' \
             '${{ needs.resolve-modal-release-config.outputs.deploy_mode }}'
 
+      - name: Verify production custom domain
+        run: bash .github/scripts/modal-custom-domain-smoke.sh
+
   publish:
     name: Publish to PyPI
     runs-on: ubuntu-latest

changelog.d/1533.changed.md

@@ -0,0 +1 @@
+Allow Modal releases to launch the same newly built worker on both `current` and `frontier`.

changelog.d/1535.added.md

@@ -0,0 +1 @@
+Enable the production Modal gateway custom domain and verify it during deployment.

docs/engineering/skills/modal-release-prs.md

@@ -24,7 +24,7 @@ modal_release:
 
 Allowed values:
 
-- `new_app_target`: `frontier`, `current`, or `none`
+- `new_app_target`: `frontier`, `current`, `both`, or `none`
 - `promote_existing_frontier`: `true` or `false`
 - `cleanup_target`: `none`, `retired`, `frontier`, or `current`
 
@@ -50,6 +50,20 @@ apps in the retired history are stopped after the manifest is updated. Use
 `cleanup_target: none` only when the user explicitly asks to preserve retired
 worker apps after release.
 
+Use this release shape when the newly built worker must become both `current`
+and `frontier` in a single release:
+
+```yaml
+modal_release:
+  new_app_target: both
+  promote_existing_frontier: false
+  cleanup_target: retired
+```
+
+This deploys one new worker app, writes the same app reference into both
+`current` and `frontier`, and moves the previous active workers into the
+manifest's `retired` history.
+
 Do not use PR labels, branch names, model-specific tags, or title prefixes to
 control Modal release behavior. The PR body YAML block is the source of truth.
 The PR-body YAML block is the only automatic signal that a deployment should

policyengine_household_api/modal_release/gateway_app.py

@@ -19,10 +19,45 @@
     "HOUSEHOLD_MODAL_GATEWAY_WEB_ENDPOINT_LABEL",
     "household-api-gateway",
 )
+GATEWAY_CUSTOM_DOMAIN = "household.api.policyengine.org"
+GATEWAY_CUSTOM_DOMAINS_ENV = "HOUSEHOLD_MODAL_GATEWAY_CUSTOM_DOMAINS"
 
 
-def gateway_wsgi_app_options() -> dict[str, object]:
-    return {"label": GATEWAY_WEB_ENDPOINT_LABEL}
+def gateway_custom_domains(
+    *,
+    modal_environment: str | None = None,
+    custom_domains: str | None = None,
+) -> tuple[str, ...]:
+    if custom_domains is None:
+        custom_domains = os.getenv(GATEWAY_CUSTOM_DOMAINS_ENV)
+
+    if custom_domains is not None:
+        return tuple(
+            domain.strip()
+            for domain in custom_domains.split(",")
+            if domain.strip()
+        )
+
+    environment = modal_environment or os.getenv("MODAL_ENVIRONMENT", "main")
+    if environment == "main":
+        return (GATEWAY_CUSTOM_DOMAIN,)
+
+    return ()
+
+
+def gateway_wsgi_app_options(
+    *,
+    modal_environment: str | None = None,
+    custom_domains: str | None = None,
+) -> dict[str, object]:
+    options: dict[str, object] = {"label": GATEWAY_WEB_ENDPOINT_LABEL}
+    domains = gateway_custom_domains(
+        modal_environment=modal_environment,
+        custom_domains=custom_domains,
+    )
+    if domains:
+        options["custom_domains"] = domains
+    return options
 
 
 app = modal.App(GATEWAY_APP_NAME)

policyengine_household_api/modal_release/manifest.py

@@ -207,6 +207,22 @@ def apply_release_config(
         next_manifest["current"] = deepcopy(dict(new_app or {}))
         next_manifest["frontier"] = None
 
+    elif config.new_app_target == NewAppTarget.BOTH:
+        retired = _retire_entry(
+            retired,
+            next_manifest.get("current"),
+            retired_at=retired_at,
+            reason="replaced-current",
+        )
+        retired = _retire_entry(
+            retired,
+            next_manifest.get("frontier"),
+            retired_at=retired_at,
+            reason="replaced-frontier",
+        )
+        next_manifest["current"] = deepcopy(dict(new_app or {}))
+        next_manifest["frontier"] = deepcopy(dict(new_app or {}))
+
     next_manifest["retired"] = retired
     return validate_manifest(next_manifest)
 

policyengine_household_api/modal_release/release_config.py

@@ -38,6 +38,7 @@ class ModalReleaseConfigError(ValueError):
 class NewAppTarget(StrEnum):
     FRONTIER = "frontier"
     CURRENT = "current"
+    BOTH = "both"
     NONE = "none"