fix(deploy): add replay checkpoint recovery
This commit is contained in:
@@ -0,0 +1,36 @@
|
||||
{
|
||||
"schemaVersion": "nodedc.device-plane.device-control-core-migration-replay-checkpoint-recovery.v2",
|
||||
"mode": "terminal-044-replay-checkpoint-forward-repair",
|
||||
"failedIncidentAudit": "device-control-core-incident-audit-20260822-043",
|
||||
"failedRecovery": {
|
||||
"patchId": "device-control-core-migration-replay-recovery-20260822-044",
|
||||
"artifactSha256": "b893d8c90f98943797d32f486d6477d58a3be69eb1291e28c4a4bbd2e96774b7",
|
||||
"failure": "preflight-replay-checkpoint-race",
|
||||
"startedApply": false
|
||||
},
|
||||
"sourcePredecessor": {
|
||||
"path": "services/device-control-core/migrations/014_device_registry_profile_commands.sql",
|
||||
"sha256": "751accf346b34d2774cc7b9572640d2c25fdb0b1db793ac32183b56f48e26508"
|
||||
},
|
||||
"sourceTarget": {
|
||||
"path": "services/device-control-core/migrations/014_device_registry_profile_commands.sql",
|
||||
"sha256": "38bd86b42828d44c7101d5433ddc36018e92eedeee37b9de296432ad676edd46"
|
||||
},
|
||||
"rootCause": "restarting-core-cycles-exact-committed-migration-checkpoints",
|
||||
"repair": "migration-014-add-constraint-not-valid",
|
||||
"databasePreflight": "exact-replay-checkpoint-005-007-009-011-and-final-compatible-rows",
|
||||
"databaseRowMutation": "none",
|
||||
"databaseSchemaOutcome": "exact-final-migration-016-validated-command-kind-check",
|
||||
"runtimeAction": "build+recreate-device-control-core-only",
|
||||
"runtimePredecessor": "proven-degraded-restarting-exact-preapply-image",
|
||||
"preservedServices": [
|
||||
"device-manager",
|
||||
"device-gateway",
|
||||
"device-postgres",
|
||||
"device-backhaul-target"
|
||||
],
|
||||
"databaseVolume": "nodedc-device-plane-postgres-data",
|
||||
"publicIngress": "disabled",
|
||||
"edgeChannel": "core-initiated-pinned-mtls-registered-edges-only",
|
||||
"rollback": "source+exact-degraded-predecessor-image-runtime"
|
||||
}
|
||||
+106
@@ -0,0 +1,106 @@
|
||||
#!/usr/bin/env node
|
||||
import { createHash } from "node:crypto";
|
||||
import { spawnSync } from "node:child_process";
|
||||
import { cp, mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
|
||||
import { tmpdir } from "node:os";
|
||||
import { dirname, join, resolve } from "node:path";
|
||||
import { fileURLToPath } from "node:url";
|
||||
|
||||
const scriptDir = dirname(fileURLToPath(import.meta.url));
|
||||
const sourceRoot = resolve(scriptDir, "../..");
|
||||
const artifactDir = resolve(
|
||||
process.env.NODEDC_DEPLOY_ARTIFACT_DIR
|
||||
|| resolve(scriptDir, "../deploy-artifacts"),
|
||||
);
|
||||
const [
|
||||
patchId = "device-control-core-migration-replay-checkpoint-recovery-20260822-046",
|
||||
...extra
|
||||
] = process.argv.slice(2);
|
||||
if (extra.length || !/^[A-Za-z0-9._-]{1,96}$/.test(patchId)) {
|
||||
throw new Error(
|
||||
"usage: build-device-control-core-migration-replay-checkpoint-recovery-artifact.mjs [patch-id]",
|
||||
);
|
||||
}
|
||||
|
||||
const entries = [
|
||||
"services/device-control-core/migrations/014_device_registry_profile_commands.sql",
|
||||
"deployment/device-control-core-migration-replay-checkpoint-recovery-v2.json",
|
||||
];
|
||||
const stage = await mkdtemp(join(tmpdir(), "nodedc-control-core-replay-checkpoint-"));
|
||||
const payload = join(stage, "payload");
|
||||
const target = resolve(artifactDir, `nodedc-device-plane-${patchId}.tgz`);
|
||||
|
||||
try {
|
||||
const descriptor = JSON.parse(await readFile(resolve(sourceRoot, entries[1]), "utf8"));
|
||||
const targetBytes = await readFile(resolve(sourceRoot, entries[0]));
|
||||
const targetSha256 = createHash("sha256").update(targetBytes).digest("hex");
|
||||
if (
|
||||
descriptor.schemaVersion
|
||||
!== "nodedc.device-plane.device-control-core-migration-replay-checkpoint-recovery.v2"
|
||||
|| descriptor.mode !== "terminal-044-replay-checkpoint-forward-repair"
|
||||
|| descriptor.failedRecovery?.patchId
|
||||
!== "device-control-core-migration-replay-recovery-20260822-044"
|
||||
|| descriptor.failedRecovery?.artifactSha256
|
||||
!== "b893d8c90f98943797d32f486d6477d58a3be69eb1291e28c4a4bbd2e96774b7"
|
||||
|| descriptor.failedRecovery?.startedApply !== false
|
||||
|| descriptor.sourceTarget.path !== entries[0]
|
||||
|| descriptor.sourceTarget.sha256 !== targetSha256
|
||||
|| descriptor.databaseRowMutation !== "none"
|
||||
|| descriptor.runtimeAction !== "build+recreate-device-control-core-only"
|
||||
) {
|
||||
throw new Error(
|
||||
"device_control_core_migration_replay_checkpoint_recovery_descriptor_mismatch",
|
||||
);
|
||||
}
|
||||
|
||||
for (const entry of entries) {
|
||||
await mkdir(dirname(join(payload, entry)), { recursive: true });
|
||||
await cp(resolve(sourceRoot, entry), join(payload, entry), { force: true });
|
||||
}
|
||||
await writeFile(
|
||||
join(stage, "manifest.env"),
|
||||
`id=${patchId}\ncomponent=device-plane\ntype=app-overlay\n`,
|
||||
"utf8",
|
||||
);
|
||||
await writeFile(join(stage, "files.txt"), `${entries.join("\n")}\n`, "utf8");
|
||||
await mkdir(artifactDir, { recursive: true });
|
||||
const tar = spawnSync(
|
||||
"python3",
|
||||
["-c", canonicalTarScript(), target, stage],
|
||||
{ encoding: "utf8", maxBuffer: 16 * 1024 * 1024 },
|
||||
);
|
||||
if (tar.status !== 0) throw new Error(`tar_failed:${tar.stderr || tar.stdout}`);
|
||||
const sha256 = createHash("sha256")
|
||||
.update(await readFile(target))
|
||||
.digest("hex");
|
||||
console.log(JSON.stringify({
|
||||
ok: true,
|
||||
patchId,
|
||||
component: "device-plane",
|
||||
artifact: target,
|
||||
sha256,
|
||||
entries,
|
||||
build: ["nodedc/device-control-core:local"],
|
||||
services: ["device-control-core"],
|
||||
transition: descriptor.mode,
|
||||
databaseRowMutation: descriptor.databaseRowMutation,
|
||||
runtimeAction: descriptor.runtimeAction,
|
||||
}, null, 2));
|
||||
} finally {
|
||||
await rm(stage, { recursive: true, force: true });
|
||||
}
|
||||
|
||||
function canonicalTarScript() {
|
||||
return [
|
||||
"import gzip,io,pathlib,sys,tarfile",
|
||||
"root=pathlib.Path(sys.argv[2])",
|
||||
"with open(sys.argv[1],'wb') as out:",
|
||||
" with gzip.GzipFile(filename='',mode='wb',fileobj=out,compresslevel=9,mtime=0) as gz:",
|
||||
" with tarfile.open(fileobj=gz,mode='w',format=tarfile.PAX_FORMAT) as tar:",
|
||||
" for top in ('manifest.env','files.txt','payload'):",
|
||||
" p=root/top; paths=[p]+(sorted(p.rglob('*')) if p.is_dir() else [])",
|
||||
" for x in paths:",
|
||||
" info=tar.gettarinfo(str(x),arcname=x.relative_to(root).as_posix()); info.uid=info.gid=0; info.uname=info.gname='root'; info.mtime=0; info.mode=0o755 if info.isdir() else 0o644",
|
||||
" with (open(x,'rb') if info.isfile() else io.BytesIO()) as src: tar.addfile(info,src if info.isfile() else None)",
|
||||
].join("\n");
|
||||
}
|
||||
@@ -94,7 +94,16 @@ class DeviceControlCoreMigrationReplayAuditArtifactTest(unittest.TestCase):
|
||||
def test_database_audit_returns_all_mismatches_without_mutation(self):
|
||||
database_result = mock.Mock(
|
||||
returncode=0,
|
||||
stdout="2\t0\tfalse\tfalse\tfalse\ttrue\n",
|
||||
stdout=(
|
||||
"2\t0\tfalse\tfalse\tfalse\t"
|
||||
+ json.dumps(
|
||||
sorted(
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_REPLAY_011_COMMAND_KINDS
|
||||
),
|
||||
separators=(",", ":"),
|
||||
)
|
||||
+ "\n"
|
||||
),
|
||||
stderr="",
|
||||
)
|
||||
with mock.patch.object(
|
||||
@@ -118,7 +127,10 @@ class DeviceControlCoreMigrationReplayAuditArtifactTest(unittest.TestCase):
|
||||
self.assertFalse(evidence["constraintCoversFinalKinds"])
|
||||
self.assertFalse(evidence["hostTelemetryTableAbsent"])
|
||||
self.assertTrue(evidence["constraintMatchesReplay011Kinds"])
|
||||
self.assertEqual(evidence["constraintPhase"], "replay-011")
|
||||
self.assertTrue(evidence["constraintMatchesKnownReplayCheckpoint"])
|
||||
self.assertFalse(evidence["recovery044Ready"])
|
||||
self.assertFalse(evidence["checkpointRecoveryReady"])
|
||||
self.assertFalse(evidence["finalStateReady"])
|
||||
command = run.call_args.args[0]
|
||||
query = command[command.index("-c") + 1]
|
||||
@@ -134,7 +146,14 @@ class DeviceControlCoreMigrationReplayAuditArtifactTest(unittest.TestCase):
|
||||
def test_recovery_rejection_prints_bounded_database_evidence(self):
|
||||
database_result = mock.Mock(
|
||||
returncode=0,
|
||||
stdout="0\t0\ttrue\ttrue\ttrue\tfalse\n",
|
||||
stdout=(
|
||||
"0\t0\ttrue\ttrue\ttrue\t"
|
||||
+ json.dumps(
|
||||
sorted(RUNNER.DEVICE_PLANE_CONTROL_CORE_FINAL_COMMAND_KINDS),
|
||||
separators=(",", ":"),
|
||||
)
|
||||
+ "\n"
|
||||
),
|
||||
stderr="",
|
||||
)
|
||||
output = io.StringIO()
|
||||
@@ -159,6 +178,8 @@ class DeviceControlCoreMigrationReplayAuditArtifactTest(unittest.TestCase):
|
||||
"constraint_covers_final_kinds=true",
|
||||
"host_telemetry_table_absent=true",
|
||||
"constraint_matches_replay_011_kinds=false",
|
||||
"constraint_phase=final-016",
|
||||
"constraint_matches_final_016_kinds=true",
|
||||
"recovery_final_state_ready=false",
|
||||
):
|
||||
self.assertIn(marker, rendered)
|
||||
@@ -197,7 +218,11 @@ class DeviceControlCoreMigrationReplayAuditArtifactTest(unittest.TestCase):
|
||||
"constraintCoversFinalKinds": True,
|
||||
"hostTelemetryTableAbsent": True,
|
||||
"constraintMatchesReplay011Kinds": True,
|
||||
"constraintPhase": "replay-011",
|
||||
"constraintMatchesKnownReplayCheckpoint": True,
|
||||
"constraintMatchesFinalKinds": False,
|
||||
"recovery044Ready": False,
|
||||
"checkpointRecoveryReady": False,
|
||||
"finalStateReady": False,
|
||||
}
|
||||
failure_evidence = {
|
||||
|
||||
+442
@@ -0,0 +1,442 @@
|
||||
#!/usr/bin/env python3
|
||||
import hashlib
|
||||
import importlib.machinery
|
||||
import importlib.util
|
||||
import inspect
|
||||
import json
|
||||
import os
|
||||
import re
|
||||
import subprocess
|
||||
import tarfile
|
||||
import tempfile
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
from unittest import mock
|
||||
|
||||
|
||||
SCRIPT_DIR = Path(__file__).resolve().parent
|
||||
RUNNER_PATH = Path(
|
||||
os.environ.get(
|
||||
"NODEDC_DEPLOY_RUNNER",
|
||||
SCRIPT_DIR.parents[2] / "platform/infra/deploy-runner/nodedc-deploy",
|
||||
)
|
||||
)
|
||||
BUILDER = (
|
||||
SCRIPT_DIR
|
||||
/ "build-device-control-core-migration-replay-checkpoint-recovery-artifact.mjs"
|
||||
)
|
||||
LEGACY_BUILDER = (
|
||||
SCRIPT_DIR
|
||||
/ "build-device-control-core-migration-replay-recovery-artifact.mjs"
|
||||
)
|
||||
MIGRATION_ROOT = SCRIPT_DIR.parents[1] / "services/device-control-core/migrations"
|
||||
|
||||
|
||||
def load_runner():
|
||||
loader = importlib.machinery.SourceFileLoader(
|
||||
"nodedc_control_core_migration_replay_checkpoint_runner_under_test",
|
||||
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 database_stdout(
|
||||
kinds,
|
||||
*,
|
||||
invalid=0,
|
||||
triggering=8,
|
||||
validated="false",
|
||||
covers_final="false",
|
||||
telemetry_absent="true",
|
||||
):
|
||||
return "\t".join(
|
||||
(
|
||||
str(invalid),
|
||||
str(triggering),
|
||||
validated,
|
||||
covers_final,
|
||||
telemetry_absent,
|
||||
json.dumps(sorted(kinds), separators=(",", ":")),
|
||||
)
|
||||
) + "\n"
|
||||
|
||||
|
||||
class DeviceControlCoreMigrationReplayCheckpointRecoveryArtifactTest(
|
||||
unittest.TestCase
|
||||
):
|
||||
def build(self, artifact_dir):
|
||||
environment = os.environ.copy()
|
||||
environment["NODEDC_DEPLOY_ARTIFACT_DIR"] = str(artifact_dir)
|
||||
result = subprocess.run(
|
||||
[
|
||||
"node",
|
||||
str(BUILDER),
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_REPLAY_CHECKPOINT_RECOVERY_PATCH_ID,
|
||||
],
|
||||
check=True,
|
||||
capture_output=True,
|
||||
text=True,
|
||||
env=environment,
|
||||
)
|
||||
return json.loads(result.stdout)
|
||||
|
||||
def test_artifact_is_deterministic_exact_new_identity(self):
|
||||
with tempfile.TemporaryDirectory(
|
||||
prefix="nodedc-control-core-checkpoint-recovery-",
|
||||
) as directory:
|
||||
artifact_dir = Path(directory)
|
||||
first = self.build(artifact_dir)
|
||||
first_bytes = Path(first["artifact"]).read_bytes()
|
||||
second = self.build(artifact_dir)
|
||||
second_bytes = Path(second["artifact"]).read_bytes()
|
||||
|
||||
self.assertEqual(first_bytes, second_bytes)
|
||||
self.assertEqual(
|
||||
first["sha256"],
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_REPLAY_CHECKPOINT_RECOVERY_ARTIFACT_SHA256,
|
||||
)
|
||||
self.assertEqual(first["services"], ["device-control-core"])
|
||||
self.assertEqual(first["databaseRowMutation"], "none")
|
||||
self.assertEqual(
|
||||
first["entries"],
|
||||
list(
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_REPLAY_CHECKPOINT_RECOVERY_ENTRIES
|
||||
),
|
||||
)
|
||||
with tarfile.open(first["artifact"], "r:gz") as archive:
|
||||
descriptor = json.loads(
|
||||
archive.extractfile(
|
||||
"payload/"
|
||||
+ RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_REPLAY_CHECKPOINT_RECOVERY_REL
|
||||
).read().decode("utf-8")
|
||||
)
|
||||
migration = archive.extractfile(
|
||||
"payload/" + RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_014_REL
|
||||
).read()
|
||||
self.assertEqual(
|
||||
descriptor,
|
||||
RUNNER.expected_device_plane_control_core_migration_replay_checkpoint_recovery_descriptor(),
|
||||
)
|
||||
self.assertEqual(
|
||||
hashlib.sha256(migration).hexdigest(),
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_014_TARGET_SHA256,
|
||||
)
|
||||
self.assertIn(b")) not valid;", migration.lower())
|
||||
self.assertIsNone(
|
||||
re.search(
|
||||
rb"(?im)^\s*(?:delete|update|insert|truncate)\b",
|
||||
migration,
|
||||
)
|
||||
)
|
||||
|
||||
def test_runner_selects_only_core_and_exact_build(self):
|
||||
entries = (
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_REPLAY_CHECKPOINT_RECOVERY_ENTRIES
|
||||
)
|
||||
self.assertEqual(
|
||||
RUNNER.component_services("device-plane", entries),
|
||||
("device-control-core",),
|
||||
)
|
||||
builds = RUNNER.component_builds("device-plane", entries)
|
||||
self.assertEqual(
|
||||
builds,
|
||||
((
|
||||
RUNNER.DEVICE_PLANE_ROOT,
|
||||
(
|
||||
"build",
|
||||
"--no-cache",
|
||||
"--network=host",
|
||||
"-f",
|
||||
"services/device-control-core/Dockerfile",
|
||||
"-t",
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_IMAGE,
|
||||
".",
|
||||
),
|
||||
),),
|
||||
)
|
||||
|
||||
def test_checkpoint_sets_match_every_committed_replay_migration(self):
|
||||
for phase, expected in RUNNER.DEVICE_PLANE_CONTROL_CORE_REPLAY_CHECKPOINTS:
|
||||
migration_number = phase.split("-", 1)[1]
|
||||
migration = next(MIGRATION_ROOT.glob(f"{migration_number}_*.sql"))
|
||||
command_kinds = tuple(re.findall(r"'([^']+)'", migration.read_text()))
|
||||
with self.subTest(phase=phase):
|
||||
self.assertEqual(command_kinds, expected)
|
||||
self.assertIn("not valid", migration.read_text().lower())
|
||||
|
||||
def test_every_exact_replay_checkpoint_is_accepted(self):
|
||||
for phase, kinds in RUNNER.DEVICE_PLANE_CONTROL_CORE_REPLAY_CHECKPOINTS:
|
||||
database_result = mock.Mock(
|
||||
returncode=0,
|
||||
stdout=database_stdout(kinds),
|
||||
stderr="",
|
||||
)
|
||||
with self.subTest(phase=phase), mock.patch.object(
|
||||
RUNNER,
|
||||
"device_plane_service_container_ids",
|
||||
return_value=("p" * 64,),
|
||||
), mock.patch.object(
|
||||
RUNNER.subprocess,
|
||||
"run",
|
||||
return_value=database_result,
|
||||
):
|
||||
evidence = (
|
||||
RUNNER.collect_device_plane_control_core_migration_replay_database_evidence(
|
||||
expected_state="replay-checkpoint-predecessor",
|
||||
)
|
||||
)
|
||||
self.assertEqual(evidence["constraintPhase"], phase)
|
||||
self.assertTrue(
|
||||
evidence["constraintMatchesKnownReplayCheckpoint"]
|
||||
)
|
||||
self.assertTrue(evidence["checkpointRecoveryReady"])
|
||||
self.assertFalse(evidence["finalStateReady"])
|
||||
|
||||
def test_unknown_missing_and_validated_checkpoint_states_are_rejected(self):
|
||||
cases = (
|
||||
(
|
||||
"unknown",
|
||||
database_stdout(("owner_scope.ensure", "unknown.ensure")),
|
||||
),
|
||||
(
|
||||
"missing",
|
||||
database_stdout((), validated="missing"),
|
||||
),
|
||||
(
|
||||
"validated-checkpoint",
|
||||
database_stdout(
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_REPLAY_005_COMMAND_KINDS,
|
||||
validated="true",
|
||||
),
|
||||
),
|
||||
)
|
||||
for name, stdout in cases:
|
||||
database_result = mock.Mock(
|
||||
returncode=0,
|
||||
stdout=stdout,
|
||||
stderr="",
|
||||
)
|
||||
with self.subTest(name=name), mock.patch.object(
|
||||
RUNNER,
|
||||
"device_plane_service_container_ids",
|
||||
return_value=("p" * 64,),
|
||||
), mock.patch.object(
|
||||
RUNNER.subprocess,
|
||||
"run",
|
||||
return_value=database_result,
|
||||
), self.assertRaises(RUNNER.DeployError):
|
||||
RUNNER.collect_device_plane_control_core_migration_replay_database_evidence(
|
||||
expected_state="replay-checkpoint-predecessor",
|
||||
)
|
||||
|
||||
def test_final_acceptance_requires_exact_migration_016_set(self):
|
||||
database_result = mock.Mock(
|
||||
returncode=0,
|
||||
stdout=database_stdout(
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_FINAL_COMMAND_KINDS,
|
||||
validated="true",
|
||||
covers_final="true",
|
||||
),
|
||||
stderr="",
|
||||
)
|
||||
with mock.patch.object(
|
||||
RUNNER,
|
||||
"device_plane_service_container_ids",
|
||||
return_value=("p" * 64,),
|
||||
), mock.patch.object(
|
||||
RUNNER.subprocess,
|
||||
"run",
|
||||
return_value=database_result,
|
||||
):
|
||||
evidence = (
|
||||
RUNNER.collect_device_plane_control_core_migration_replay_database_evidence(
|
||||
expected_state="final",
|
||||
)
|
||||
)
|
||||
self.assertEqual(evidence["constraintPhase"], "final-016")
|
||||
self.assertTrue(evidence["constraintMatchesFinalKinds"])
|
||||
self.assertTrue(evidence["finalStateReady"])
|
||||
self.assertFalse(evidence["checkpointRecoveryReady"])
|
||||
|
||||
def test_terminal_044_failure_evidence_is_exact_and_preapply(self):
|
||||
with tempfile.TemporaryDirectory(
|
||||
prefix="nodedc-control-core-recovery-044-failure-",
|
||||
) as directory:
|
||||
root = Path(directory)
|
||||
artifact_dir = root / "artifact"
|
||||
failed_dir = root / "failed"
|
||||
runner_tmp = root / "tmp"
|
||||
state_file = root / "applied.jsonl"
|
||||
failed_state_file = root / "failed.jsonl"
|
||||
artifact_dir.mkdir()
|
||||
failed_dir.mkdir()
|
||||
runner_tmp.mkdir()
|
||||
environment = os.environ.copy()
|
||||
environment["NODEDC_DEPLOY_ARTIFACT_DIR"] = str(artifact_dir)
|
||||
subprocess.run(
|
||||
[
|
||||
"node",
|
||||
str(LEGACY_BUILDER),
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_REPLAY_RECOVERY_PATCH_ID,
|
||||
],
|
||||
check=True,
|
||||
capture_output=True,
|
||||
text=True,
|
||||
env=environment,
|
||||
)
|
||||
source = (
|
||||
artifact_dir
|
||||
/ RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_REPLAY_RECOVERY_ARTIFACT
|
||||
)
|
||||
failed_artifact = (
|
||||
failed_dir
|
||||
/ RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_REPLAY_RECOVERY_FAILED_ARTIFACT
|
||||
)
|
||||
source.replace(failed_artifact)
|
||||
state_file.write_text("", encoding="utf-8")
|
||||
record = {
|
||||
"artifact": failed_artifact.name,
|
||||
"backup_id": None,
|
||||
"component": "device-plane",
|
||||
"failed_at": (
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_REPLAY_RECOVERY_FAILED_AT
|
||||
),
|
||||
"id": RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_REPLAY_RECOVERY_PATCH_ID,
|
||||
"message": (
|
||||
"Device Control Core migration recovery database invariant mismatch"
|
||||
),
|
||||
"rollback_status": "not-required",
|
||||
"sha256": (
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_REPLAY_RECOVERY_ARTIFACT_SHA256
|
||||
),
|
||||
"started_apply": False,
|
||||
"status": "failed",
|
||||
}
|
||||
failed_state_file.write_text(
|
||||
json.dumps(record) + "\n",
|
||||
encoding="utf-8",
|
||||
)
|
||||
|
||||
with mock.patch.object(RUNNER, "FAILED_DIR", failed_dir), mock.patch.object(
|
||||
RUNNER,
|
||||
"STATE_FILE",
|
||||
state_file,
|
||||
), mock.patch.object(
|
||||
RUNNER,
|
||||
"FAILED_STATE_FILE",
|
||||
failed_state_file,
|
||||
), mock.patch.object(RUNNER, "TMP_DIR", runner_tmp):
|
||||
evidence = (
|
||||
RUNNER.validate_device_plane_control_core_migration_replay_recovery_failure()
|
||||
)
|
||||
self.assertEqual(evidence["record"], record)
|
||||
|
||||
def test_checkpoint_recovery_preflight_reaches_database_collector(self):
|
||||
with tempfile.TemporaryDirectory(
|
||||
prefix="nodedc-control-core-checkpoint-preflight-",
|
||||
) as directory:
|
||||
root = Path(directory)
|
||||
migration = (
|
||||
root / RUNNER.DEVICE_PLANE_CONTROL_CORE_MIGRATION_014_REL
|
||||
)
|
||||
migration.parent.mkdir(parents=True)
|
||||
migration.write_bytes(b"migration-014-predecessor")
|
||||
core = {
|
||||
"containerId": "c" * 64,
|
||||
"imageId": RUNNER.DEVICE_PLANE_CONTROL_CORE_V3_PREAPPLY_IMAGE_ID,
|
||||
"status": "restarting",
|
||||
"health": "unhealthy",
|
||||
"restartCount": 280,
|
||||
}
|
||||
runtime = {"core": core, "current": {"services": [core]}}
|
||||
database = {
|
||||
"constraintPhase": "replay-007",
|
||||
"checkpointRecoveryReady": True,
|
||||
}
|
||||
descriptor = (
|
||||
RUNNER.expected_device_plane_control_core_migration_replay_checkpoint_recovery_descriptor()
|
||||
)
|
||||
|
||||
with mock.patch.object(
|
||||
RUNNER,
|
||||
"validate_device_plane_control_core_migration_replay_checkpoint_recovery_payload",
|
||||
return_value=descriptor,
|
||||
), mock.patch.object(
|
||||
RUNNER,
|
||||
"validate_device_plane_control_core_double_failure_evidence",
|
||||
return_value={
|
||||
"firstBackup": root / "first",
|
||||
"secondBackup": root / "second",
|
||||
},
|
||||
), mock.patch.object(
|
||||
RUNNER,
|
||||
"validate_device_plane_control_core_migration_replay_recovery_failure",
|
||||
return_value={"record": {"started_apply": False}},
|
||||
), mock.patch.object(
|
||||
RUNNER,
|
||||
"component_root",
|
||||
return_value=root,
|
||||
), mock.patch.object(
|
||||
RUNNER,
|
||||
"DEVICE_PLANE_CONTROL_CORE_MIGRATION_014_PREDECESSOR_SHA256",
|
||||
hashlib.sha256(migration.read_bytes()).hexdigest(),
|
||||
), mock.patch.object(
|
||||
RUNNER,
|
||||
"validate_device_plane_control_core_migration_replay_preserved_runtime",
|
||||
return_value=runtime,
|
||||
), mock.patch.object(
|
||||
RUNNER,
|
||||
"inspect_optional_local_image",
|
||||
return_value=RUNNER.DEVICE_PLANE_CONTROL_CORE_V3_PREAPPLY_IMAGE_ID,
|
||||
), mock.patch.object(
|
||||
RUNNER,
|
||||
"collect_device_plane_control_core_migration_replay_database_evidence",
|
||||
return_value=database,
|
||||
) as collector:
|
||||
evidence = (
|
||||
RUNNER.validate_device_plane_control_core_migration_replay_checkpoint_recovery_evidence(
|
||||
root
|
||||
)
|
||||
)
|
||||
|
||||
self.assertEqual(evidence["database"], database)
|
||||
collector.assert_called_once_with(
|
||||
expected_state="replay-checkpoint-predecessor",
|
||||
)
|
||||
|
||||
def test_v2_rollback_and_acceptance_are_registered(self):
|
||||
rollback_source = inspect.getsource(RUNNER.rollback_device_plane_apply)
|
||||
health_source = inspect.getsource(RUNNER.run_healthchecks)
|
||||
preflight_source = inspect.getsource(
|
||||
RUNNER.validate_device_plane_control_core_migration_replay_checkpoint_recovery_evidence
|
||||
)
|
||||
acceptance_source = inspect.getsource(
|
||||
RUNNER.accept_device_plane_control_core_migration_replay_checkpoint_recovery
|
||||
)
|
||||
self.assertIn(
|
||||
"is_device_plane_control_core_migration_replay_checkpoint_recovery_slice",
|
||||
rollback_source,
|
||||
)
|
||||
self.assertIn(
|
||||
"accept_device_plane_control_core_migration_replay_checkpoint_recovery",
|
||||
health_source,
|
||||
)
|
||||
self.assertIn(
|
||||
'expected_state="replay-checkpoint-predecessor"',
|
||||
preflight_source,
|
||||
)
|
||||
self.assertIn(
|
||||
"validate_device_plane_control_core_migration_replay_recovery_failure",
|
||||
preflight_source,
|
||||
)
|
||||
self.assertIn('expected_state="final"', acceptance_source)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -139,7 +139,16 @@ class DeviceControlCoreMigrationReplayRecoveryArtifactTest(unittest.TestCase):
|
||||
def test_database_preflight_is_read_only_and_covers_live_receipts(self):
|
||||
database_result = mock.Mock(
|
||||
returncode=0,
|
||||
stdout="0\t8\tfalse\tfalse\ttrue\ttrue\n",
|
||||
stdout=(
|
||||
"0\t8\tfalse\tfalse\ttrue\t"
|
||||
+ json.dumps(
|
||||
sorted(
|
||||
RUNNER.DEVICE_PLANE_CONTROL_CORE_REPLAY_011_COMMAND_KINDS
|
||||
),
|
||||
separators=(",", ":"),
|
||||
)
|
||||
+ "\n"
|
||||
),
|
||||
stderr="",
|
||||
)
|
||||
with mock.patch.object(
|
||||
@@ -162,7 +171,9 @@ class DeviceControlCoreMigrationReplayRecoveryArtifactTest(unittest.TestCase):
|
||||
self.assertFalse(evidence["constraintValidated"])
|
||||
self.assertFalse(evidence["constraintCoversFinalKinds"])
|
||||
self.assertTrue(evidence["constraintMatchesReplay011Kinds"])
|
||||
self.assertEqual(evidence["constraintPhase"], "replay-011")
|
||||
self.assertTrue(evidence["recovery044Ready"])
|
||||
self.assertTrue(evidence["checkpointRecoveryReady"])
|
||||
self.assertFalse(evidence["finalStateReady"])
|
||||
command = run.call_args.args[0]
|
||||
query = command[command.index("-c") + 1]
|
||||
@@ -175,8 +186,7 @@ class DeviceControlCoreMigrationReplayRecoveryArtifactTest(unittest.TestCase):
|
||||
)
|
||||
)
|
||||
self.assertIn("regexp_matches", query)
|
||||
for command_kind in RUNNER.DEVICE_PLANE_CONTROL_CORE_REPLAY_011_COMMAND_KINDS:
|
||||
self.assertIn(f"'{command_kind}'", query)
|
||||
self.assertIn("array_to_json", query)
|
||||
|
||||
def test_runner_predecessor_kind_set_matches_migration_011(self):
|
||||
migration = MIGRATION_011.read_text(encoding="utf-8")
|
||||
@@ -190,7 +200,14 @@ class DeviceControlCoreMigrationReplayRecoveryArtifactTest(unittest.TestCase):
|
||||
def test_database_acceptance_requires_final_validated_constraint(self):
|
||||
database_result = mock.Mock(
|
||||
returncode=0,
|
||||
stdout="0\t8\ttrue\ttrue\ttrue\tfalse\n",
|
||||
stdout=(
|
||||
"0\t8\ttrue\ttrue\ttrue\t"
|
||||
+ json.dumps(
|
||||
sorted(RUNNER.DEVICE_PLANE_CONTROL_CORE_FINAL_COMMAND_KINDS),
|
||||
separators=(",", ":"),
|
||||
)
|
||||
+ "\n"
|
||||
),
|
||||
stderr="",
|
||||
)
|
||||
with mock.patch.object(
|
||||
@@ -211,13 +228,20 @@ class DeviceControlCoreMigrationReplayRecoveryArtifactTest(unittest.TestCase):
|
||||
self.assertTrue(evidence["constraintValidated"])
|
||||
self.assertTrue(evidence["constraintCoversFinalKinds"])
|
||||
self.assertFalse(evidence["constraintMatchesReplay011Kinds"])
|
||||
self.assertEqual(evidence["constraintPhase"], "final-016")
|
||||
self.assertTrue(evidence["constraintMatchesFinalKinds"])
|
||||
self.assertFalse(evidence["recovery044Ready"])
|
||||
self.assertFalse(evidence["checkpointRecoveryReady"])
|
||||
self.assertTrue(evidence["finalStateReady"])
|
||||
|
||||
def test_database_preflight_rejects_ambiguous_constraint_shape(self):
|
||||
database_result = mock.Mock(
|
||||
returncode=0,
|
||||
stdout="0\t8\tfalse\tfalse\ttrue\tfalse\n",
|
||||
stdout=(
|
||||
"0\t8\tfalse\tfalse\ttrue\t"
|
||||
+ json.dumps(["owner_scope.ensure", "unknown.ensure"])
|
||||
+ "\n"
|
||||
),
|
||||
stderr="",
|
||||
)
|
||||
with mock.patch.object(
|
||||
|
||||
Reference in New Issue
Block a user