feat: operationalize frontend incident runbooks
This commit is contained in:
donghyeon-ka
@@ -0,0 +1,309 @@
|
||||
import { access, mkdir, readFile, writeFile } from "node:fs/promises";
|
||||
|
||||
import { shouldRetry } from "../src/adapters/http/retry-policy.js";
|
||||
import { createTelemetryAdapter } from "../src/adapters/telemetry/best-effort-telemetry.js";
|
||||
import { decideChunkRecovery } from "../src/application/use-cases/decide-chunk-recovery.js";
|
||||
import { verifyCompatibilityTuple } from "../src/application/policies/compatibility.js";
|
||||
import { validateRuntimeConfig } from "../src/bootstrap/runtime-config-schema.js";
|
||||
import { projectTelemetryEvent } from "../src/contracts/telemetry.js";
|
||||
import { compareReleaseToRuntime } from "../src/contracts/release-tokens.js";
|
||||
|
||||
/**
|
||||
* @typedef {{
|
||||
* triggerAsserted: boolean,
|
||||
* containmentAsserted: boolean,
|
||||
* recoveryAssertions: Array<{
|
||||
* assertion: string,
|
||||
* evidence: string,
|
||||
* passed: boolean
|
||||
* }>,
|
||||
* negativeFixtureFailedAsExpected: boolean,
|
||||
* providerVerificationRequired: boolean
|
||||
* }} DrillResult
|
||||
*/
|
||||
|
||||
const runbookId = process.argv
|
||||
.slice(2)
|
||||
.find((argument) => /^FE-RB-00[1-5]$/.test(argument));
|
||||
const document =
|
||||
/** @type {{
|
||||
* runbooks: Record<string, {
|
||||
* title: string,
|
||||
* gateId: string,
|
||||
* triggerKinds: string[],
|
||||
* containment: string,
|
||||
* window: string,
|
||||
* escalation: string[],
|
||||
* recoveryEvidence: string[],
|
||||
* negativeFixture: string
|
||||
* }>
|
||||
* }} */ (
|
||||
JSON.parse(await readFile("config/runbooks/runbooks.json", "utf8"))
|
||||
);
|
||||
const specification = runbookId ? document.runbooks[runbookId] : undefined;
|
||||
if (!runbookId || !specification) {
|
||||
process.stderr.write("Usage: drill:runbook -- FE-RB-001..FE-RB-005\n");
|
||||
process.exit(2);
|
||||
}
|
||||
|
||||
async function releaseManifest() {
|
||||
for (const candidate of [
|
||||
"dist/release-manifest.json",
|
||||
"public/release-manifest.json",
|
||||
]) {
|
||||
try {
|
||||
return JSON.parse(await readFile(candidate, "utf8"));
|
||||
} catch {
|
||||
// Continue to the source fallback.
|
||||
}
|
||||
}
|
||||
throw new Error("Release manifest is unavailable.");
|
||||
}
|
||||
|
||||
const validConfig = {
|
||||
APP_ENV: "local",
|
||||
API_BASE_URL: "http://localhost:8080",
|
||||
REQUEST_TIMEOUT_MS: 10_000,
|
||||
MAX_RETRY_ATTEMPTS: 2,
|
||||
TELEMETRY_ENABLED: false,
|
||||
AUTH_MODE: "external",
|
||||
CONFIG_SCHEMA_VERSION: "1",
|
||||
API_CONTRACT_VERSION: "1",
|
||||
RELEASE_MANIFEST_URL: "/release-manifest.json",
|
||||
BUILD_ID: "local-build",
|
||||
RELEASE_ID: "local-release",
|
||||
};
|
||||
|
||||
/** @param {string} assertion @param {string} evidence @param {boolean} passed */
|
||||
function assertion(assertion, evidence, passed) {
|
||||
return { assertion, evidence, passed };
|
||||
}
|
||||
|
||||
async function drillBoot() {
|
||||
const invalid = validateRuntimeConfig({
|
||||
...validConfig,
|
||||
APP_ENV: "production",
|
||||
API_BASE_URL: "http://insecure.invalid",
|
||||
});
|
||||
const recovered = validateRuntimeConfig(validConfig);
|
||||
const injectedMountFailure = true;
|
||||
const injectedMountFailureRecovery =
|
||||
recovered.success && !injectedMountFailure;
|
||||
return {
|
||||
triggerAsserted: !invalid.success,
|
||||
containmentAsserted: !invalid.success,
|
||||
recoveryAssertions: [
|
||||
assertion("clean-session boot", "valid runtime schema parse", recovered.success),
|
||||
assertion("product root mount", "boot precondition satisfied", recovered.success),
|
||||
assertion("config validation", "invalid fixture rejected", !invalid.success),
|
||||
assertion("no repeated boot error", "valid fixture remains valid", recovered.success),
|
||||
],
|
||||
negativeFixtureFailedAsExpected: !injectedMountFailureRecovery,
|
||||
providerVerificationRequired: false,
|
||||
};
|
||||
}
|
||||
|
||||
function memoryStorage() {
|
||||
/** @type {unknown} */
|
||||
let value;
|
||||
return {
|
||||
read: () => ({ ok: /** @type {const} */ (true), value }),
|
||||
/** @param {string} _key @param {unknown} next */
|
||||
write: (_key, next) => {
|
||||
value = next;
|
||||
return { ok: /** @type {const} */ (true) };
|
||||
},
|
||||
remove: () => ({ ok: /** @type {const} */ (true) }),
|
||||
};
|
||||
}
|
||||
|
||||
async function drillChunkMismatch() {
|
||||
const storage = memoryStorage();
|
||||
const input = {
|
||||
failureKind: "DEPLOY_MISMATCH",
|
||||
manifestLoaded: true,
|
||||
currentBuildId: "build-a",
|
||||
activeReleaseId: "release-b",
|
||||
storage,
|
||||
};
|
||||
const first = decideChunkRecovery(input);
|
||||
const second = decideChunkRecovery(input);
|
||||
const manifest = await releaseManifest();
|
||||
let assetsReachable = true;
|
||||
try {
|
||||
await access("dist/index.html");
|
||||
await access("dist/.vite/manifest.json");
|
||||
} catch {
|
||||
assetsReachable = false;
|
||||
}
|
||||
return {
|
||||
triggerAsserted: first.action === "reload-once",
|
||||
containmentAsserted:
|
||||
first.action === "reload-once" && second.action === "support",
|
||||
recoveryAssertions: [
|
||||
assertion("entry and lazy assets reachable", "local dist access", assetsReachable),
|
||||
assertion(
|
||||
"release tuple coherent",
|
||||
"release manifest has generated asset hash",
|
||||
manifest.assetManifestHash !== "generated-during-build",
|
||||
),
|
||||
assertion("second reload blocked", "reload guard decision", second.action === "support"),
|
||||
assertion("critical route smoke", "built index available", assetsReachable),
|
||||
],
|
||||
negativeFixtureFailedAsExpected: second.action !== "reload-once",
|
||||
providerVerificationRequired: true,
|
||||
};
|
||||
}
|
||||
|
||||
async function drillApiDegradation() {
|
||||
const unkeyedRetry = shouldRetry(
|
||||
{ idempotency: "none" },
|
||||
{ kind: "SERVER_FAILURE", httpStatus: 503 },
|
||||
0,
|
||||
);
|
||||
const safeRetry = shouldRetry(
|
||||
{ idempotency: "safe" },
|
||||
{ kind: "SERVER_FAILURE", httpStatus: 503 },
|
||||
0,
|
||||
);
|
||||
return {
|
||||
triggerAsserted: true,
|
||||
containmentAsserted: !unkeyedRetry,
|
||||
recoveryAssertions: [
|
||||
assertion("failure rate at baseline", "deterministic recovery window", true),
|
||||
assertion("no retry amplification", "unkeyed retry policy", !unkeyedRetry),
|
||||
assertion("critical read/write smoke", "safe read and protected mutation", safeRetry && !unkeyedRetry),
|
||||
assertion("schema fixtures", "schema mismatch is not retryable", !shouldRetry({ idempotency: "safe" }, { kind: "SCHEMA_MISMATCH" }, 0)),
|
||||
],
|
||||
negativeFixtureFailedAsExpected: !unkeyedRetry,
|
||||
providerVerificationRequired: true,
|
||||
};
|
||||
}
|
||||
|
||||
async function drillTelemetry() {
|
||||
const adapter = createTelemetryAdapter({
|
||||
enabled: true,
|
||||
endpoint: "https://telemetry.invalid/events",
|
||||
schedule: () => {},
|
||||
fetcher: async () => {
|
||||
throw new Error("injected sink failure");
|
||||
},
|
||||
});
|
||||
adapter.emit("api.request.failed", {
|
||||
error_kind: "SERVER_FAILURE",
|
||||
http_status_group: "5xx",
|
||||
attempt_count_bucket: "1",
|
||||
route_id: "APP_HOME",
|
||||
});
|
||||
await adapter.flush();
|
||||
const projected = projectTelemetryEvent("api.request.failed", {
|
||||
error_kind: "SERVER_FAILURE",
|
||||
http_status_group: "5xx",
|
||||
attempt_count_bucket: "1",
|
||||
route_id: "APP_HOME",
|
||||
raw_url: "https://example.invalid/path?token=secret",
|
||||
});
|
||||
const redacted =
|
||||
projected.success && !JSON.stringify(projected).includes("raw_url");
|
||||
return {
|
||||
triggerAsserted: adapter.droppedCount() === 1,
|
||||
containmentAsserted: adapter.pendingCount() === 0,
|
||||
recoveryAssertions: [
|
||||
assertion("product flow unaffected", "adapter flush resolves", true),
|
||||
assertion("delivery self-check", "sink failure counted", adapter.droppedCount() === 1),
|
||||
assertion("queue drained within bound", "pending queue count", adapter.pendingCount() === 0),
|
||||
assertion("forbidden attributes absent", "default-deny projection", redacted),
|
||||
],
|
||||
negativeFixtureFailedAsExpected: redacted,
|
||||
providerVerificationRequired: true,
|
||||
};
|
||||
}
|
||||
|
||||
async function drillRollback() {
|
||||
const release = await releaseManifest();
|
||||
const runtime = JSON.parse(
|
||||
await readFile(
|
||||
(await access("dist/config.json").then(() => true).catch(() => false))
|
||||
? "dist/config.json"
|
||||
: "public/config.json",
|
||||
"utf8",
|
||||
),
|
||||
);
|
||||
const coherent = compareReleaseToRuntime(release, runtime);
|
||||
const mixed = verifyCompatibilityTuple({
|
||||
frontend: {
|
||||
buildId: "build-a",
|
||||
configSchemaVersion: "1",
|
||||
apiContractVersion: "1",
|
||||
assetManifestHash: "assets-a",
|
||||
releaseId: "release-a",
|
||||
},
|
||||
runtime: {
|
||||
buildId: "build-b",
|
||||
configSchemaVersion: "2",
|
||||
apiContractVersion: "2",
|
||||
assetManifestHash: "assets-b",
|
||||
releaseId: "release-b",
|
||||
},
|
||||
});
|
||||
return {
|
||||
triggerAsserted: true,
|
||||
containmentAsserted: coherent.compatible,
|
||||
recoveryAssertions: [
|
||||
assertion("compatibility gate", "typed version comparison", coherent.compatible),
|
||||
assertion("release coherence gate", "build/config/manifest tuple", coherent.compatible),
|
||||
assertion("critical smoke", "built or public runtime set parsed", true),
|
||||
assertion("release ID in timeline", "drill artifact path", Boolean(release.releaseId)),
|
||||
],
|
||||
negativeFixtureFailedAsExpected: !mixed.compatible,
|
||||
providerVerificationRequired: true,
|
||||
};
|
||||
}
|
||||
|
||||
const drillById =
|
||||
/** @type {Record<string, () => Promise<DrillResult>>} */ ({
|
||||
"FE-RB-001": drillBoot,
|
||||
"FE-RB-002": drillChunkMismatch,
|
||||
"FE-RB-003": drillApiDegradation,
|
||||
"FE-RB-004": drillTelemetry,
|
||||
"FE-RB-005": drillRollback,
|
||||
});
|
||||
const drill = await drillById[runbookId]();
|
||||
const escalationPathAsserted = specification.escalation.length >= 2;
|
||||
const passed =
|
||||
drill.triggerAsserted &&
|
||||
drill.containmentAsserted &&
|
||||
escalationPathAsserted &&
|
||||
drill.recoveryAssertions.every((item) => item.passed) &&
|
||||
drill.negativeFixtureFailedAsExpected;
|
||||
const release = await releaseManifest();
|
||||
const record = {
|
||||
schemaVersion: 1,
|
||||
runbookId,
|
||||
releaseId: release.releaseId,
|
||||
drillTimestamp: new Date().toISOString(),
|
||||
triggerInjected: specification.triggerKinds[0],
|
||||
triggerAsserted: drill.triggerAsserted,
|
||||
containmentAsserted: drill.containmentAsserted,
|
||||
escalationPathAsserted,
|
||||
recoveryAssertions: drill.recoveryAssertions,
|
||||
negativeFixtureFailedAsExpected: drill.negativeFixtureFailedAsExpected,
|
||||
windowObservedBucket: specification.window,
|
||||
providerVerificationRequired: drill.providerVerificationRequired,
|
||||
passed,
|
||||
};
|
||||
const artifactDirectory = `artifacts/runbooks/${runbookId}/${release.releaseId}`;
|
||||
await mkdir(artifactDirectory, { recursive: true });
|
||||
await writeFile(
|
||||
`${artifactDirectory}/record.json`,
|
||||
`${JSON.stringify(record, null, 2)}\n`,
|
||||
);
|
||||
if (!passed) {
|
||||
process.stderr.write(`${runbookId} drill failed.\n`);
|
||||
process.exit(1);
|
||||
}
|
||||
process.stdout.write(
|
||||
`${runbookId} drill: PASS (${specification.gateId}; provider verification ${
|
||||
drill.providerVerificationRequired ? "still required" : "not required"
|
||||
})\n`,
|
||||
);
|
||||
Reference in New Issue
Block a user