@tutti-os/nexight-devops 0.5.2 → 0.5.4

package/README.md

@@ -48,6 +48,7 @@ nexight-devops ppe-lane status --all
 nexight-devops release prepare
 nexight-devops release hotfix
 nexight-devops release approve
+nexight-devops release rollback
 nexight-devops release rollback tutti-server
 ```
 
@@ -64,11 +65,11 @@ nexight-devops release rollback tutti-server
 - `admin computer log-recall` exposes the lower-level task operations for troubleshooting and recovery.
 - `datafinder` validates access, reads dashboards/reports, runs trusted local JavaScript queries, and exposes fixed reliability queries.
 - `diagnose shared-agent` runs a bounded, read-only CloudWatch playbook that correlates room descriptor results, Owner lease observations, client version, and HTTP failures. It verifies the production AWS account before querying and supports stable `--json` output.
-- `ppe-lane deploy` deploys selected backend repositories and the optional `zk-web/canvas` staging artifact to a PPE lane.
-- `ppe-lane destroy` checks existing lane instances and triggers destroy workflows after a lane-name confirmation.
-- `ppe-lane status` queries AWS Cloud Map for one lane, or all lanes with `--all`.
+- `ppe-lane deploy` deploys selected backend repositories, including `tutti-server`, and the optional `zk-web/canvas` staging artifact to a PPE lane.
+- `ppe-lane destroy` checks existing lane instances and triggers destroy workflows after a lane-name confirmation. `tutti-server` must be selected explicitly for destruction.
+- `ppe-lane status` queries AWS Cloud Map for one lane, or all lanes with `--all`, including `tutti-server`.
 - `release prepare`, `release hotfix`, and `release approve` run the controlled release lifecycle through GitHub Actions and approval issues.
-- `release rollback tutti-server` lists the current production image and recent successful Helm deployments (up to ten entries), then restores a selected image through GitHub Actions and waits for health checks. Use `--revision <number> --yes` for scripted selection and confirmation.
+- `release rollback` lets you choose `tutti-server` interactively; `release rollback tutti-server` selects it directly. The command reads the current production image and up to ten recent Helm deployments directly when local read access to the production cluster is available. Otherwise it retrieves the verified list through GitHub Actions. The selected image is always revalidated and restored by GitHub Actions, followed by health checks. Use `--revision <number> --yes` with the explicit repository for scripted selection and confirmation.
 
 ## Controlled Release
 

package/dist/cli.js

@@ -4018,6 +4018,17 @@ var ppeLaneCanvasTarget = {
 
 // src/constants/ppe-lane-repos.ts
 var ppeLaneRepos = [
+  {
+    key: "tutti-server",
+    label: "tutti-server",
+    repoFullName: "tutti-lab/tutti-server",
+    deployWorkflowFile: "deploy-lane.yml",
+    destroyWorkflowFile: "destroy-lane.yml",
+    defaultRef: "main",
+    defaultDestroySelection: false,
+    cloudMapServiceId: "srv-izinli73rwxpf2dr",
+    serviceName: "tutti-server"
+  },
   {
     key: "zk-agent-engine",
     label: "zk-agent-engine",
@@ -4463,7 +4474,11 @@ var runPpeLaneDestroy = async () => {
   const lane = validatePpeLaneName(await ui.input({ message: t("ppeLane.destroy.enterLane") }));
   const selectedKeys = await ui.checkbox({
     message: t("ppeLane.destroy.selectRepos"),
-    choices: ppeLaneRepos.map((repo) => ({ value: repo.key, name: repo.label, checked: true }))
+    choices: ppeLaneRepos.map((repo) => ({
+      value: repo.key,
+      name: repo.label,
+      checked: repo.defaultDestroySelection !== false
+    }))
   });
   if (selectedKeys.length === 0) {
     throw new CliError(t("ppeLane.destroy.atLeastOneRepo"), 1);
@@ -5085,7 +5100,103 @@ var registerReleasePrepareCommand = (program) => {
 import { mkdtemp as mkdtemp3, readFile as readFile3, rm as rm2 } from "fs/promises";
 import { tmpdir } from "os";
 import { join as join5 } from "path";
-var REPO = "tutti-lab/tutti-server";
+
+// src/constants/rollback-targets.ts
+var tuttiServerRollbackTarget = {
+  key: "tutti-server",
+  label: "tutti-server",
+  repoFullName: "tutti-lab/tutti-server"
+};
+var rollbackTargets = [tuttiServerRollbackTarget];
+
+// src/commands/release/rollback-local.ts
+var CLUSTER = "arn:aws:eks:us-east-1:250509935467:cluster/nexight-ojo-us-east-1";
+var REPOSITORY = "250509935467.dkr.ecr.us-east-1.amazonaws.com/nexight/tutti/tutti-server";
+var DIGEST = /^sha256:[0-9a-f]{64}$/;
+var MANIFEST_IMAGE = new RegExp(`image:\\s*["']?${REPOSITORY.replaceAll(".", "\\.")}@(sha256:[0-9a-f]{64})`, "g");
+var read = async (command, args) => {
+  const result = await runCommand(command, args);
+  if (result.exitCode !== 0) {
+    throw new CliError(result.stderr.trim() || `${command} read failed`, 1);
+  }
+  return result.stdout.trim();
+};
+var digestFromManifest = (manifest) => {
+  const matches = [...manifest.matchAll(MANIFEST_IMAGE)].map((match) => match[1]);
+  const digests = new Set(matches);
+  if (digests.size !== 1) {
+    throw new CliError("Helm revision has no unambiguous Tutti image", 1);
+  }
+  return [...digests][0];
+};
+var buildLocalRollbackCatalog = (history, deployment, manifests) => {
+  if (!Array.isArray(history) || history.length === 0) {
+    throw new CliError("Production Helm history is empty", 1);
+  }
+  const sorted = [...history].sort((left, right) => right.revision - left.revision);
+  const latest = sorted[0];
+  if (!latest || !["deployed", "failed"].includes(latest.status)) {
+    throw new CliError("Latest Helm revision needs inspection", 1);
+  }
+  const resource = deployment;
+  const images = resource?.spec?.template?.spec?.containers?.filter((container) => container.name === "tutti-server").map((container) => container.image);
+  if (images?.length !== 1 || !images[0]?.startsWith(`${REPOSITORY}@`)) {
+    throw new CliError("Production deployment has no unambiguous Tutti image", 1);
+  }
+  const currentDigest = images[0].slice(REPOSITORY.length + 1);
+  if (!DIGEST.test(currentDigest)) {
+    throw new CliError("Production deployment image is not digest-pinned", 1);
+  }
+  const entries = [latest, ...sorted.filter(
+    (item) => item.revision !== latest.revision && ["deployed", "superseded"].includes(item.status)
+  )].slice(0, 10);
+  const candidates = entries.map((item) => {
+    const manifest = manifests.get(item.revision);
+    if (!manifest) {
+      throw new CliError(`Missing manifest for Helm revision ${item.revision}`, 1);
+    }
+    const digest = digestFromManifest(manifest);
+    return {
+      revision: item.revision,
+      deployedAt: item.updated,
+      digest,
+      version: `sha256:${digest.slice(7, 19)}`,
+      current: digest === currentDigest
+    };
+  });
+  if (candidates[0]?.digest !== currentDigest) {
+    throw new CliError("Current image does not match the latest Helm revision", 1);
+  }
+  return { environment: "production", currentDigest, currentRevision: latest.revision, candidates };
+};
+var fetchLocalRollbackCatalog = async () => {
+  const cluster = await read("kubectl", ["config", "view", "--minify", "-o", "jsonpath={.clusters[0].name}"]);
+  if (cluster !== CLUSTER) {
+    throw new CliError("Current kubectl context is not the production cluster", 1);
+  }
+  const [historyRaw, deploymentRaw] = await Promise.all([
+    read("helm", ["history", "tutti-server", "-n", "tutti", "--max", "100", "-o", "json"]),
+    read("kubectl", ["get", "deployment", "tutti-server", "-n", "tutti", "-o", "json"])
+  ]);
+  const history = JSON.parse(historyRaw);
+  const deployment = JSON.parse(deploymentRaw);
+  if (!Array.isArray(history) || history.length === 0) {
+    throw new CliError("Production Helm history is empty", 1);
+  }
+  const sorted = [...history].sort((left, right) => right.revision - left.revision);
+  const latest = sorted[0];
+  const entries = [latest, ...sorted.filter(
+    (item) => item.revision !== latest.revision && ["deployed", "superseded"].includes(item.status)
+  )].slice(0, 10);
+  const manifests = new Map(await Promise.all(entries.map(async (item) => [
+    item.revision,
+    await read("helm", ["get", "manifest", "tutti-server", "-n", "tutti", "--revision", String(item.revision)])
+  ])));
+  return buildLocalRollbackCatalog(history, deployment, manifests);
+};
+
+// src/commands/release/rollback.ts
+var REPO = tuttiServerRollbackTarget.repoFullName;
 var WORKFLOW = "rollback.yml";
 var SHA256 = /^sha256:[0-9a-f]{64}$/;
 var SINGAPORE_TIME = new Intl.DateTimeFormat("sv-SE", {
@@ -5106,7 +5217,7 @@ var displayChoiceTime = (raw) => {
   const date = new Date(raw);
   return Number.isNaN(date.getTime()) ? raw : SINGAPORE_CHOICE_TIME.format(date);
 };
-var buildRollbackChoices = (catalog) => catalog.candidates.filter((candidate) => !candidate.current).sort((left, right) => Date.parse(right.deployedAt) - Date.parse(left.deployedAt)).map((candidate) => ({
+var buildRollbackChoices = (catalog) => catalog.candidates.filter((candidate) => !candidate.current && candidate.digest !== catalog.currentDigest).sort((left, right) => Date.parse(right.deployedAt) - Date.parse(left.deployedAt)).map((candidate) => ({
   name: `${displayChoiceTime(candidate.deployedAt)} SGT  #${candidate.revision}  ${candidate.version}`,
   value: candidate.revision
 }));
@@ -5194,11 +5305,22 @@ var fetchCatalog = async (runId) => {
     await rm2(dir, { recursive: true, force: true });
   }
 };
+var readCatalog = async () => {
+  const progress = ui.spinner("\u76F4\u63A5\u8BFB\u53D6\u751F\u4EA7 Helm \u5386\u53F2...").start();
+  try {
+    return await fetchLocalRollbackCatalog();
+  } catch {
+  } finally {
+    progress.stop();
+  }
+  logger.warn("\u672C\u673A\u65E0\u6CD5\u5B8C\u6574\u8BFB\u53D6\u751F\u4EA7\u90E8\u7F72\u5386\u53F2\uFF0C\u6539\u7528 GitHub Actions \u67E5\u8BE2\u3002");
+  const listRun = await dispatch("list");
+  return fetchCatalog(listRun.runId);
+};
 var runTuttiRollback = async (options) => {
   await ensurePrerequisites({ needsGh: true });
   await ensureNoActiveDeployment();
-  const listRun = await dispatch("list");
-  const catalog = await fetchCatalog(listRun.runId);
+  const catalog = await readCatalog();
   const choices = buildRollbackChoices(catalog);
   if (choices.length === 0) {
     throw new CliError("No previous retained production image is available to restore", 1);
@@ -5215,7 +5337,9 @@ var runTuttiRollback = async (options) => {
       pageSize: choices.length
     });
   }
-  const selected = catalog.candidates.find((candidate) => candidate.revision === revision && !candidate.current);
+  const selected = catalog.candidates.find(
+    (candidate) => candidate.revision === revision && !candidate.current && candidate.digest !== catalog.currentDigest
+  );
   if (!selected) {
     throw new CliError(`Revision ${revision} is not an eligible recent deployment`, 1);
   }
@@ -5224,6 +5348,7 @@ var runTuttiRollback = async (options) => {
   logger.info(`  ${catalog.currentDigest}`);
   logger.info(`Target: Helm revision ${selected.revision} / ${displayTime(selected.deployedAt)} / ${selected.version}`);
   logger.info(`  ${selected.digest}`);
+  logger.info("GitHub Actions will recheck the live deployment and target image before changing production.");
   if (!options.yes && !await ui.confirm({ message: "Restore this image to production?", default: false })) {
     logger.warn("Rollback cancelled");
     return;
@@ -5239,8 +5364,12 @@ var runTuttiRollback = async (options) => {
   logger.info(run.htmlUrl);
 };
 var registerReleaseRollbackCommand = (program) => {
-  program.command("rollback").description("Restore a recent production image").argument("<repository>", "Currently supported: tutti-server").option("--revision <number>", "Select a recent Helm revision without the chooser").option("--yes", "Skip the final confirmation").action(async (repository, options) => {
-    if (repository !== "tutti-server") {
+  program.command("rollback").description("Restore a recent production image").argument("[repository]", "Rollback target; omit to choose interactively").option("--revision <number>", "Select a recent Helm revision without the chooser").option("--yes", "Skip the final confirmation").action(async (repository, options) => {
+    const selected = repository ?? await ui.select({
+      message: "Select a repository to roll back",
+      choices: rollbackTargets.map((target) => ({ name: target.label, value: target.key }))
+    });
+    if (selected !== tuttiServerRollbackTarget.key) {
       throw new CliError("Only tutti-server is currently supported", 1);
     }
     await runTuttiRollback(options);