Name missing observations at migrate's verify steps instead of crashing (#183)

changelog.d/183-migrate-observations-missing.md

@@ -0,0 +1,3 @@
+### Fixed
+
+- `cloudfall migrate --yes` no longer crashes with `HANDLER_CRASHED` when it reaches its `audit` or `verify-routes` step on a project with no server observations: the step fails with `migrate_observations_missing`, naming the missing directory and `cloudfall observe`, and a rerun after `observe` resumes at that step (#183)

sdk/src/cloudfall/migrate.py

@@ -80,6 +80,7 @@
 _PAUSE_CODES = frozenset({_ERROR_DNS_UNVERIFIED, _ERROR_TTL_HIGH})
 _ERROR_AUDIT_DRIFT = "migrate_audit_drift"
 _ERROR_ROUTES_UNHEALTHY = "migrate_routes_unhealthy"
+_ERROR_OBSERVATIONS_MISSING = "migrate_observations_missing"
 _ERROR_DATABASE_UNKNOWN = "migrate_database_unknown"
 _ERROR_DATA_SCHEMA_WITHOUT_DATA = "migrate_data_schema_without_data"
 
@@ -729,7 +730,23 @@ def _resolve_names(hostname: str) -> set[str]:
     return {str(record[4][0]) for record in records}
 
 
+def _require_observations(config: AgentConfig) -> None:
+    """Refuse a verify step whose server snapshots were never recorded.
+
+    Only the missing directory: a malformed snapshot still fails validation,
+    and a server without a snapshot is drift the audit reports.
+    """
+    if not config.observed_directory.is_dir():
+        detail = (
+            "no server observations at "
+            f"{config.observed_directory}; record them with "
+            "`cloudfall observe`, then rerun the migration"
+        )
+        raise MigrateError(_ERROR_OBSERVATIONS_MISSING, detail)
+
+
 def _verify_audit(config: AgentConfig) -> dict[str, object]:
+    _require_observations(config)
     state = validate_config(config.project_directory, config.schema_directory)
     observations = load_observations(
         config.observed_directory, config.schema_directory
@@ -746,6 +763,8 @@ def _verify_audit(config: AgentConfig) -> dict[str, object]:
 
 
 def _verify_routes(config: AgentConfig) -> dict[str, object]:
+    # Before any domain is probed: the view needs the server snapshots too.
+    _require_observations(config)
     state = validate_config(config.project_directory, config.schema_directory)
     inventory = PlatformInventory.from_state(state)
     inspect_domains(

sdk/tests/test_migrate.py

@@ -325,6 +325,59 @@ def test_s007_12_cli_migrate_refuses_a_bad_data_schema_before_any_playbook(
     assert not (tmp_path / "inventory.json").exists()
 
 
+@pytest.mark.parametrize("step_id", ["audit", "verify-routes"])
+def test_cli_migrate_names_missing_observations_at_the_verify_step(
+    step_id: str,
+    tmp_path: Path,
+    capsys: pytest.CaptureFixture[str],
+) -> None:
+    """A verify step without server snapshots fails typed, not crashed (#183)."""
+    options = _options(tmp_path, execute=True)
+    # Every step before this one completes; the CLI run resumes here.
+    stop = MigrateError("migrate_step_failed", "stopped for the test")
+    execute_migration(
+        _config(tmp_path),
+        options,
+        runners=_runners([], {step_id: stop}),
+    )
+    observed = tmp_path / "observed"
+
+    exit_code = main(
+        [
+            "migrate",
+            "--project",
+            str(EXAMPLES),
+            "--schemas",
+            str(SCHEMAS),
+            "--engine",
+            str(ENGINE),
+            "--plan-file",
+            str(options.plan_file),
+            "--observed",
+            str(observed),
+            "--service-observed",
+            str(tmp_path / "observed-services"),
+            "--build",
+            "crm-backend=main",
+            "--yes",
+        ]
+    )
+
+    payload = json.loads(capsys.readouterr().out)
+    assert payload["error"]["code"] != "HANDLER_CRASHED"
+    assert payload["error"]["code"] == "ENGINE_STEP_FAILED"
+    assert exit_code == 87
+    context = payload["error"]["context"]
+    assert context["step"] == step_id
+    assert context["code"] == "migrate_observations_missing"
+    assert str(observed) in context["message"]
+    assert "cloudfall observe" in context["message"]
+    assert payload["data"]["next"] == step_id
+    plan = json.loads(options.plan_file.read_text(encoding="utf-8"))
+    pending = [step["id"] for step in plan["steps"] if step["status"] == "pending"]
+    assert pending[0] == step_id
+
+
 # dns-verify's proxy addresses, pinned as they were before domains apply.
 
 

sdk/tests/test_owned_postgresql_converge.py

@@ -48,7 +48,6 @@
     SopsSecretProvider,
 )
 from cloudfall.validation import (
-    ConfigValidationError,
     SchemaCatalog,
     validate_config,
 )
@@ -436,12 +435,16 @@ def test_s006_6_migrate_records_the_hash_in_its_services_step(tmp_path: Path) ->
     secrets = _secrets(project, f"BILLING_PASSWORD={PASSWORD}\n")
     recorder = Recorder()
 
-    # Every playbook "passes"; the run stops at the audit, which reads
-    # snapshots no stand-in wrote, after the services step was saved.
-    with pytest.raises(ConfigValidationError, match="observation directory"):
-        execute_migration(
-            _agent_config(project, recorder, secrets), _migrate_options(project)
-        )
+    # Every playbook "passes"; the run stops at the audit, which finds no
+    # snapshots (no stand-in wrote any), after the services step was saved.
+    result = execute_migration(
+        _agent_config(project, recorder, secrets), _migrate_options(project)
+    )
+
+    assert result["status"] == "error"
+    assert result["step"] == "audit"
+    error = cast("dict[str, object]", result["error"])
+    assert error["code"] == "migrate_observations_missing"
 
     plan = (project / "tmp" / "plan.json").read_text(encoding="utf-8")
     services = next(