Files
NODEDC_PLATFORM/infra/deploy-runner/test_module_foundry_runtime_recovery.py
T

473 lines
17 KiB
Python

#!/usr/bin/env python3
import hashlib
import importlib.machinery
import importlib.util
import json
import subprocess
import tarfile
import tempfile
import unittest
from contextlib import ExitStack
from pathlib import Path
from unittest import mock
SCRIPT_DIR = Path(__file__).resolve().parent
RUNNER_PATH = SCRIPT_DIR / "nodedc-deploy"
def load_runner():
loader = importlib.machinery.SourceFileLoader(
"nodedc_module_foundry_runtime_recovery",
str(RUNNER_PATH),
)
spec = importlib.util.spec_from_loader(loader.name, loader)
module = importlib.util.module_from_spec(spec)
loader.exec_module(module)
return module
RUNNER = load_runner()
def runtime(container_char="a", image_char="b"):
return {
"schemaVersion": "nodedc.module-foundry.runtime-inventory.v1",
"composeProject": "nodedc-module-foundry",
"service": RUNNER.MODULE_FOUNDRY_SERVICE,
"containerId": container_char * 64,
"imageId": f"sha256:{image_char * 64}",
"imageRef": "nodedc-module-foundry-nodedc-module-foundry:latest",
"health": "healthy",
}
class ModuleFoundryRuntimeRecoveryTest(unittest.TestCase):
def test_build_retries_before_any_runtime_recreate(self):
first_failure = subprocess.CalledProcessError(17, ["docker", "compose"])
completed = subprocess.CompletedProcess([], 0)
with (
mock.patch.object(
RUNNER,
"component_compose_root",
return_value=Path("/compose"),
),
mock.patch.object(
RUNNER,
"compose_base_cmd",
return_value=["docker", "compose"],
),
mock.patch.object(
RUNNER.subprocess,
"run",
side_effect=[first_failure, completed],
) as run,
mock.patch.object(RUNNER.time, "sleep") as sleep,
mock.patch.object(RUNNER, "run_module_foundry_compose_up") as up,
):
RUNNER.run_module_foundry_compose(
(RUNNER.MODULE_FOUNDRY_SERVICE,)
)
expected = [
"docker",
"compose",
"build",
RUNNER.MODULE_FOUNDRY_SERVICE,
]
self.assertEqual(run.call_count, 2)
self.assertEqual(run.call_args_list[0].args[0], expected)
self.assertEqual(run.call_args_list[1].args[0], expected)
sleep.assert_called_once_with(2)
up.assert_called_once_with((RUNNER.MODULE_FOUNDRY_SERVICE,))
def test_runtime_recreate_uses_built_image_without_build_or_pull(self):
completed = subprocess.CompletedProcess([], 0)
with (
mock.patch.object(
RUNNER,
"component_compose_root",
return_value=Path("/compose"),
),
mock.patch.object(
RUNNER,
"compose_base_cmd",
return_value=["docker", "compose"],
),
mock.patch.object(
RUNNER.subprocess,
"run",
return_value=completed,
) as run,
):
RUNNER.run_module_foundry_compose_up(
(RUNNER.MODULE_FOUNDRY_SERVICE,)
)
up = run.call_args_list[0].args[0]
self.assertEqual(
up,
[
"docker",
"compose",
"up",
"-d",
"--force-recreate",
"--pull",
"never",
"--no-deps",
RUNNER.MODULE_FOUNDRY_SERVICE,
],
)
self.assertNotIn("--build", up)
self.assertEqual(run.call_args_list[1].args[0], ["docker", "compose", "ps"])
def test_rollback_restores_source_and_keeps_unchanged_runtime(self):
with tempfile.TemporaryDirectory(
prefix="module-foundry-rollback-unchanged-"
) as directory:
work = Path(directory)
root = work / "source"
backup = work / "backup"
tmp = work / "tmp"
root.mkdir()
backup.mkdir()
tmp.mkdir()
entries = ("old.txt", "new.txt")
(root / "old.txt").write_text("old\n", encoding="utf-8")
RUNNER.create_backup(root, backup, entries, False)
(root / "old.txt").write_text("candidate\n", encoding="utf-8")
(root / "new.txt").write_text("candidate\n", encoding="utf-8")
before = runtime()
(backup / RUNNER.MODULE_FOUNDRY_RUNTIME_BEFORE_FILE).write_text(
json.dumps(before),
encoding="utf-8",
)
with (
mock.patch.object(RUNNER, "TMP_DIR", tmp),
mock.patch.object(
RUNNER,
"module_foundry_runtime_inventory",
return_value=before,
),
mock.patch.object(RUNNER, "run_healthchecks") as health,
mock.patch.object(RUNNER.subprocess, "run") as process,
):
result = RUNNER.rollback_module_foundry_apply(
root,
backup,
entries,
"20260809-000000",
(RUNNER.MODULE_FOUNDRY_SERVICE,),
)
self.assertEqual(
result,
"source-restored-runtime-unchanged:2",
)
self.assertEqual(
(root / "old.txt").read_text(encoding="utf-8"),
"old\n",
)
self.assertFalse((root / "new.txt").exists())
process.assert_not_called()
health.assert_called_once()
def test_rollback_retags_and_recreates_changed_runtime(self):
with tempfile.TemporaryDirectory(
prefix="module-foundry-rollback-changed-"
) as directory:
work = Path(directory)
root = work / "source"
backup = work / "backup"
tmp = work / "tmp"
root.mkdir()
backup.mkdir()
tmp.mkdir()
entries = ("old.txt",)
(root / "old.txt").write_text("old\n", encoding="utf-8")
RUNNER.create_backup(root, backup, entries, False)
(root / "old.txt").write_text("candidate\n", encoding="utf-8")
before = runtime()
candidate = runtime("c", "d")
(backup / RUNNER.MODULE_FOUNDRY_RUNTIME_BEFORE_FILE).write_text(
json.dumps(before),
encoding="utf-8",
)
with (
mock.patch.object(RUNNER, "TMP_DIR", tmp),
mock.patch.object(
RUNNER,
"module_foundry_runtime_inventory",
side_effect=[candidate, before],
) as inventory,
mock.patch.object(
RUNNER,
"inspect_local_image",
return_value=before["imageId"],
),
mock.patch.object(RUNNER, "run_module_foundry_compose_up") as up,
mock.patch.object(RUNNER, "run_healthchecks") as health,
mock.patch.object(RUNNER.subprocess, "run") as process,
):
result = RUNNER.rollback_module_foundry_apply(
root,
backup,
entries,
"20260809-000000",
(RUNNER.MODULE_FOUNDRY_SERVICE,),
)
self.assertEqual(result, "source+runtime-restored:1")
self.assertEqual(
inventory.call_args_list[0].kwargs,
{"required": False, "require_healthy": False},
)
process.assert_called_once_with(
[
str(RUNNER.DOCKER),
"image",
"tag",
before["imageId"],
before["imageRef"],
],
check=True,
)
up.assert_called_once_with((RUNNER.MODULE_FOUNDRY_SERVICE,))
health.assert_called_once()
def test_candidate_must_replace_both_container_and_image(self):
before = runtime()
for candidate in (
runtime("a", "c"),
runtime("c", "b"),
):
with mock.patch.object(
RUNNER,
"module_foundry_runtime_inventory",
return_value=candidate,
):
with self.assertRaises(RUNNER.DeployError):
RUNNER.validate_module_foundry_candidate_runtime(before)
def test_preapply_runtime_rejects_digest_only_image_reference(self):
with tempfile.TemporaryDirectory(
prefix="module-foundry-runtime-reference-"
) as directory:
backup = Path(directory)
before = runtime()
before["imageRef"] = before["imageId"]
(backup / RUNNER.MODULE_FOUNDRY_RUNTIME_BEFORE_FILE).write_text(
json.dumps(before),
encoding="utf-8",
)
with self.assertRaises(RUNNER.DeployError):
RUNNER.read_module_foundry_runtime_before(backup)
def test_recovery_accepts_only_exact_failed_partial_predecessor(self):
with tempfile.TemporaryDirectory(
prefix="module-foundry-map-evidence-"
) as directory:
work = Path(directory)
backups = work / "backups"
failed = work / "failed"
state = work / "state"
tmp = work / "tmp"
installed = work / "installed"
candidate = work / "candidate"
for path in (backups, failed, state, tmp, installed, candidate):
path.mkdir()
recovery_entries = (
"apps/catalog/src",
"scripts",
"server/map-grid-persistence.test.mjs",
)
missing_entries = ("server/map-grid-persistence.test.mjs",)
failed_entries = (
"apps/catalog/src/MapFixturePreview.tsx",
"scripts/map-sector-grid.test.mjs",
)
candidate_files = {
"apps/catalog/src/MapFixturePreview.tsx": "sector-current\n",
"apps/catalog/src/CesiumMapRenderer.tsx": "renderer-next\n",
"scripts/map-sector-grid.test.mjs": "sector-test-current\n",
"scripts/map-grid-lod.test.mjs": "grid-test-next\n",
"server/map-grid-persistence.test.mjs": "server-test-next\n",
}
installed_files = {
"apps/catalog/src/MapFixturePreview.tsx": "sector-current\n",
"apps/catalog/src/CesiumMapRenderer.tsx": "renderer-old\n",
"scripts/map-sector-grid.test.mjs": "sector-test-current\n",
"scripts/map-grid-lod.test.mjs": "grid-test-old\n",
}
for root, values in (
(candidate, candidate_files),
(installed, installed_files),
):
for rel, value in values.items():
path = root / rel
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text(value, encoding="utf-8")
failed_patch_id = "module-foundry-map-failed-test"
failed_artifact_name = "module-foundry-map-failed-test.tgz.1"
failed_backup_id = "module-foundry-map-failed-test-backup"
recovery_patch_id = "module-foundry-map-recovery-test"
failed_message = "test build failed"
backup = backups / failed_backup_id
backup.mkdir()
backup_names = {
"existing-files.txt",
"files.txt",
"manifest.env",
"missing-files.txt",
"source-before.tgz",
}
for name in backup_names:
(backup / name).write_text(f"{name}\n", encoding="utf-8")
backup_hashes = {
name: hashlib.sha256((backup / name).read_bytes()).hexdigest()
for name in backup_names
}
archive_source = work / "archive-source"
(archive_source / "payload").mkdir(parents=True)
(archive_source / "manifest.env").write_text(
"id="
f"{failed_patch_id}\n"
"component=module-foundry\n"
"type=app-overlay\n",
encoding="utf-8",
)
(archive_source / "files.txt").write_text(
"\n".join(failed_entries) + "\n",
encoding="utf-8",
)
for rel in failed_entries:
source = candidate / rel
target = archive_source / "payload" / rel
target.parent.mkdir(parents=True, exist_ok=True)
target.write_bytes(source.read_bytes())
failed_artifact = failed / failed_artifact_name
with tarfile.open(failed_artifact, "w:gz") as archive:
archive.add(
archive_source / "manifest.env",
arcname="manifest.env",
)
archive.add(
archive_source / "files.txt",
arcname="files.txt",
)
archive.add(archive_source / "payload", arcname="payload")
failed_sha256 = hashlib.sha256(
failed_artifact.read_bytes()
).hexdigest()
journal = state / "failed.jsonl"
journal.write_text(
json.dumps(
{
"artifact": failed_artifact_name,
"backup_id": failed_backup_id,
"component": "module-foundry",
"id": failed_patch_id,
"message": failed_message,
"rollback_status": "not-required",
"sha256": failed_sha256,
"started_apply": True,
"status": "failed",
}
)
+ "\n",
encoding="utf-8",
)
manifest = {
"id": recovery_patch_id,
"component": "module-foundry",
"type": "app-overlay",
}
candidate_tree = RUNNER.exact_file_map_sha256(
RUNNER.collect_exact_files(
candidate,
recovery_entries,
"candidate",
)
)
installed_tree = RUNNER.exact_file_map_sha256(
RUNNER.collect_exact_files(
installed,
recovery_entries[:-1],
"installed",
)
)
patches = {
"BACKUPS_DIR": backups,
"FAILED_DIR": failed,
"FAILED_STATE_FILE": journal,
"TMP_DIR": tmp,
"MODULE_FOUNDRY_MAP_RECOVERY_PATCH_ID": recovery_patch_id,
"MODULE_FOUNDRY_MAP_RECOVERY_ENTRIES": recovery_entries,
"MODULE_FOUNDRY_MAP_RECOVERY_MISSING_PREDECESSOR_ENTRIES": (
missing_entries
),
"MODULE_FOUNDRY_MAP_RECOVERY_INSTALLED_TREE_SHA256": (
installed_tree
),
"MODULE_FOUNDRY_MAP_RECOVERY_CANDIDATE_TREE_SHA256": (
candidate_tree
),
"MODULE_FOUNDRY_SECTOR_FAILED_ENTRIES": failed_entries,
"MODULE_FOUNDRY_MAP_FAILED_PATCH_ID": failed_patch_id,
"MODULE_FOUNDRY_MAP_FAILED_ARTIFACT": failed_artifact_name,
"MODULE_FOUNDRY_MAP_FAILED_ARTIFACT_SHA256": failed_sha256,
"MODULE_FOUNDRY_MAP_FAILED_BACKUP_ID": failed_backup_id,
"MODULE_FOUNDRY_MAP_FAILED_MESSAGE": failed_message,
"MODULE_FOUNDRY_MAP_FAILED_BACKUP_SHA256": backup_hashes,
}
with ExitStack() as stack:
for name, value in patches.items():
stack.enter_context(mock.patch.object(RUNNER, name, value))
stack.enter_context(
mock.patch.object(
RUNNER,
"component_root",
return_value=installed,
)
)
stack.enter_context(
mock.patch.object(
RUNNER,
"module_foundry_runtime_inventory",
return_value=runtime(),
)
)
result = RUNNER.validate_module_foundry_map_recovery_evidence(
manifest,
recovery_entries,
candidate,
)
self.assertEqual(
result["mode"],
"failed-map-overlay-source+runtime-reconciliation",
)
(installed / failed_entries[0]).write_text(
"drift\n",
encoding="utf-8",
)
with self.assertRaises(RUNNER.DeployError):
RUNNER.validate_module_foundry_map_recovery_evidence(
manifest,
recovery_entries,
candidate,
)
if __name__ == "__main__":
unittest.main(verbosity=2)