Blame · Line-by-line history
admin-diagnose.tsx
Each line is annotated with the commit that last touched it. Click any SHA to jump to that commit and see the surrounding change.
| 826eccf | 1 | /** |
| 2 | * /admin/diagnose — comprehensive AI health scan. | |
| 3 | * | |
| 4 | * Single page the site admin opens to see, at a glance, every config knob | |
| 5 | * the platform depends on and whether it is wired up. Each row is one | |
| 6 | * check; status is green / yellow / red with a one-line "what to do". | |
| 7 | * | |
| 8 | * Categories covered: | |
| 9 | * - Email delivery (EMAIL_PROVIDER, RESEND_API_KEY) | |
| 10 | * - AI (ANTHROPIC_API_KEY presence) | |
| 11 | * - GateTest integration (URL + API key) | |
| 12 | * - Service worker SHA (BUILD_SHA pinned vs dev-stable fallback) | |
| 13 | * - Database (DATABASE_URL well-formed, latest migration applied) | |
| 14 | * - Canonical URL (APP_BASE_URL matches request host) | |
| 15 | * - Self-host (SELF_HOST_REPO declared + post-receive hook present) | |
| 16 | * - Auto-merge (branch_protection.enable_auto_merge for main) | |
| 17 | * - Synthetic monitor (any RED checks in last hour) | |
| 18 | * - Email smoke (POST /admin/diagnose/test-email fires a test) | |
| 19 | * | |
| 20 | * Gating: requireAuth + isSiteAdmin via the same gate() pattern as | |
| 21 | * /admin/ops + /admin/status. No data leaks to non-admins. | |
| 22 | */ | |
| 23 | ||
| 24 | import { Hono } from "hono"; | |
| 115c66b | 25 | import { eq, and, desc, gt, sql } from "drizzle-orm"; |
| 826eccf | 26 | import { readdir } from "fs/promises"; |
| 27 | import { join } from "path"; | |
| 28 | import { | |
| 29 | branchProtection, | |
| 30 | repositories, | |
| 31 | syntheticChecks, | |
| 32 | users, | |
| 115c66b | 33 | workflowRuns, |
| 826eccf | 34 | } from "../db/schema"; |
| 115c66b | 35 | import { platformDeploys } from "../db/schema-deploys"; |
| 826eccf | 36 | import { db } from "../db"; |
| 37 | import { Layout } from "../views/layout"; | |
| 38 | import { softAuth } from "../middleware/auth"; | |
| 39 | import type { AuthEnv } from "../middleware/auth"; | |
| 40 | import { isSiteAdmin } from "../lib/admin"; | |
| 41 | import { config } from "../lib/config"; | |
| 42 | import { sendEmail } from "../lib/email"; | |
| 43 | import { latestMigration } from "../lib/post-deploy-smoke"; | |
| 115c66b | 44 | import { getLastTick, getTickCount } from "../lib/autopilot"; |
| 826eccf | 45 | |
| 46 | type CheckStatus = "green" | "yellow" | "red"; | |
| 47 | ||
| 48 | interface CheckResult { | |
| 49 | category: string; | |
| 50 | name: string; | |
| 51 | status: CheckStatus; | |
| 52 | detail: string; | |
| 53 | fix?: string; | |
| 54 | } | |
| 55 | ||
| 56 | const diagnose = new Hono<AuthEnv>(); | |
| 57 | diagnose.use("*", softAuth); | |
| 58 | ||
| 59 | async function gate(c: any): Promise<{ user: any } | Response> { | |
| 60 | const user = c.get("user"); | |
| 61 | if (!user) return c.redirect("/login?next=/admin/diagnose"); | |
| 62 | if (!(await isSiteAdmin(user.id))) { | |
| 63 | return c.html( | |
| 64 | <Layout title="Forbidden" user={user}> | |
| 65 | <div class="empty-state"> | |
| 66 | <h2>403 — Not a site admin</h2> | |
| 67 | <p>You don't have permission to view this page.</p> | |
| 68 | </div> | |
| 69 | </Layout>, | |
| 70 | 403 | |
| 71 | ); | |
| 72 | } | |
| 73 | return { user }; | |
| 74 | } | |
| 75 | ||
| 76 | // ─── Individual checks ─────────────────────────────────────────────────── | |
| 77 | ||
| 78 | function checkEmail(): CheckResult { | |
| 79 | if (config.emailProvider !== "resend") { | |
| 80 | return { | |
| 81 | category: "Email", | |
| 82 | name: "Provider", | |
| 83 | status: "red", | |
| 84 | detail: `EMAIL_PROVIDER=${config.emailProvider} — verification + magic-link emails go to stderr, not inboxes.`, | |
| 85 | fix: "Set EMAIL_PROVIDER=resend in /etc/gluecron.env, then `systemctl restart gluecron`.", | |
| 86 | }; | |
| 87 | } | |
| 88 | if (!config.resendApiKey) { | |
| 89 | return { | |
| 90 | category: "Email", | |
| 91 | name: "Provider", | |
| 92 | status: "red", | |
| 93 | detail: "EMAIL_PROVIDER=resend but RESEND_API_KEY is empty — every send will fail.", | |
| 94 | fix: "Add RESEND_API_KEY=re_xxx to /etc/gluecron.env, then restart.", | |
| 95 | }; | |
| 96 | } | |
| 97 | return { | |
| 98 | category: "Email", | |
| 99 | name: "Provider", | |
| 100 | status: "green", | |
| 101 | detail: `Resend wired (from: ${config.emailFrom}). Use the test button below to confirm.`, | |
| 102 | }; | |
| 103 | } | |
| 104 | ||
| 105 | function checkAnthropic(): CheckResult { | |
| 106 | if (!config.anthropicApiKey) { | |
| 107 | return { | |
| 108 | category: "AI", | |
| 109 | name: "Anthropic API key", | |
| 110 | status: "yellow", | |
| 111 | detail: "ANTHROPIC_API_KEY unset — AI PR review + AI deploy-failure analysis disabled.", | |
| 112 | fix: "Add ANTHROPIC_API_KEY=sk-ant-xxx to /etc/gluecron.env.", | |
| 113 | }; | |
| 114 | } | |
| 115 | return { | |
| 116 | category: "AI", | |
| 117 | name: "Anthropic API key", | |
| 118 | status: "green", | |
| 119 | detail: `Key present (length ${config.anthropicApiKey.length}).`, | |
| 120 | }; | |
| 121 | } | |
| 122 | ||
| 123 | function checkGateTest(): CheckResult { | |
| 124 | const hasUrl = !!process.env.GATETEST_URL; | |
| 125 | const hasKey = !!process.env.GATETEST_API_KEY; | |
| 126 | if (!hasUrl && !hasKey) { | |
| 127 | return { | |
| 128 | category: "GateTest", | |
| 129 | name: "Scanner integration", | |
| 130 | status: "yellow", | |
| 131 | detail: "Unconfigured — push-time GateTest scans skip silently.", | |
| 132 | fix: "Set GATETEST_URL + GATETEST_API_KEY in /etc/gluecron.env to enable per-push scans.", | |
| 133 | }; | |
| 134 | } | |
| 135 | if (hasUrl && !hasKey) { | |
| 136 | return { | |
| 137 | category: "GateTest", | |
| 138 | name: "Scanner integration", | |
| 139 | status: "red", | |
| 140 | detail: "GATETEST_URL set but GATETEST_API_KEY empty — calls will 401.", | |
| 141 | fix: "Add GATETEST_API_KEY to /etc/gluecron.env.", | |
| 142 | }; | |
| 143 | } | |
| 144 | return { | |
| 145 | category: "GateTest", | |
| 146 | name: "Scanner integration", | |
| 147 | status: "green", | |
| 148 | detail: `Configured — pushes POST to ${process.env.GATETEST_URL}.`, | |
| 149 | }; | |
| 150 | } | |
| 151 | ||
| 152 | function checkBuildSha(): CheckResult { | |
| 153 | const sha = process.env.BUILD_SHA?.trim(); | |
| 154 | if (sha) { | |
| 155 | return { | |
| 156 | category: "Deploy", | |
| 157 | name: "BUILD_SHA pinned", | |
| 158 | status: "green", | |
| 159 | detail: `${sha.slice(0, 12)} — service worker rotates per deploy.`, | |
| 160 | }; | |
| 161 | } | |
| 162 | return { | |
| 163 | category: "Deploy", | |
| 164 | name: "BUILD_SHA pinned", | |
| 165 | status: "yellow", | |
| 166 | detail: "BUILD_SHA unset — falling back to dev-stable. Browsers won't see new deploys reflected in the SW cache.", | |
| 167 | fix: "Latest scripts/self-deploy.sh + hetzner-deploy.yml pin this automatically; trigger a deploy.", | |
| 168 | }; | |
| 169 | } | |
| 170 | ||
| 171 | function checkAppBaseUrl(c: any): CheckResult { | |
| 172 | const expected = config.appBaseUrl; | |
| 173 | const host = c.req.header("host") || ""; | |
| 174 | const proto = | |
| 175 | c.req.header("x-forwarded-proto") || | |
| 176 | (c.req.url.startsWith("https://") ? "https" : "http"); | |
| 177 | const actual = `${proto}://${host}`; | |
| 178 | if (!expected || expected === "http://localhost:3000") { | |
| 179 | return { | |
| 180 | category: "Config", | |
| 181 | name: "APP_BASE_URL canonical", | |
| 182 | status: "yellow", | |
| 183 | detail: `APP_BASE_URL is "${expected}" — outbound email links + WebAuthn origin will be wrong.`, | |
| 184 | fix: "Set APP_BASE_URL=https://gluecron.com in /etc/gluecron.env.", | |
| 185 | }; | |
| 186 | } | |
| 187 | if (host && !expected.endsWith(host)) { | |
| 188 | return { | |
| 189 | category: "Config", | |
| 190 | name: "APP_BASE_URL canonical", | |
| 191 | status: "yellow", | |
| 192 | detail: `APP_BASE_URL=${expected} but request arrived at ${actual}. WebAuthn passkeys issued for one host can't be used at the other.`, | |
| 193 | fix: "Align APP_BASE_URL with the host you actually serve from.", | |
| 194 | }; | |
| 195 | } | |
| 196 | return { | |
| 197 | category: "Config", | |
| 198 | name: "APP_BASE_URL canonical", | |
| 199 | status: "green", | |
| 200 | detail: expected, | |
| 201 | }; | |
| 202 | } | |
| 203 | ||
| 204 | function checkDatabase(): CheckResult { | |
| 205 | const url = config.databaseUrl; | |
| 206 | if (!url) { | |
| 207 | return { | |
| 208 | category: "Database", | |
| 209 | name: "Connection string", | |
| 210 | status: "red", | |
| 211 | detail: "DATABASE_URL unset — every page that queries the DB will 500.", | |
| 212 | fix: "Set DATABASE_URL in /etc/gluecron.env.", | |
| 213 | }; | |
| 214 | } | |
| 215 | let masked = url; | |
| 216 | try { | |
| 217 | const u = new URL(url); | |
| 218 | masked = `${u.protocol}//${u.username ? "***" : ""}@${u.host}${u.pathname}`; | |
| 219 | } catch { | |
| 220 | // unparseable | |
| 221 | return { | |
| 222 | category: "Database", | |
| 223 | name: "Connection string", | |
| 224 | status: "red", | |
| 225 | detail: "DATABASE_URL is not a valid URL.", | |
| 226 | fix: "Fix the URL format: postgres://user:pass@host:port/dbname", | |
| 227 | }; | |
| 228 | } | |
| 229 | return { | |
| 230 | category: "Database", | |
| 231 | name: "Connection string", | |
| 232 | status: "green", | |
| 233 | detail: masked, | |
| 234 | }; | |
| 235 | } | |
| 236 | ||
| 237 | async function checkMigrations(): Promise<CheckResult> { | |
| 238 | try { | |
| 239 | const drizzleDir = join(process.cwd(), "drizzle"); | |
| 240 | const files = (await readdir(drizzleDir)).filter((f) => f.endsWith(".sql")); | |
| 241 | const latest = latestMigration(files); | |
| 242 | if (!latest) { | |
| 243 | return { | |
| 244 | category: "Database", | |
| 245 | name: "Migrations applied", | |
| 246 | status: "yellow", | |
| 247 | detail: "No migration files found in drizzle/.", | |
| 248 | }; | |
| 249 | } | |
| 250 | const rows = (await db.execute( | |
| 251 | `SELECT name FROM _migrations ORDER BY name DESC LIMIT 1` as never | |
| 252 | )) as any; | |
| 253 | const list = rows?.rows ?? (Array.isArray(rows) ? rows : []); | |
| 254 | const applied: string | undefined = list[0]?.name; | |
| 255 | if (!applied) { | |
| 256 | return { | |
| 257 | category: "Database", | |
| 258 | name: "Migrations applied", | |
| 259 | status: "red", | |
| 260 | detail: "_migrations table is empty.", | |
| 261 | fix: "Run `bun run db:migrate` on the box.", | |
| 262 | }; | |
| 263 | } | |
| 264 | if (applied !== latest) { | |
| 265 | return { | |
| 266 | category: "Database", | |
| 267 | name: "Migrations applied", | |
| 268 | status: "red", | |
| 269 | detail: `DB at ${applied}, drizzle/ has ${latest}.`, | |
| 270 | fix: "Run `bun run db:migrate` on the box, or redeploy (the workflow runs it).", | |
| 271 | }; | |
| 272 | } | |
| 273 | return { | |
| 274 | category: "Database", | |
| 275 | name: "Migrations applied", | |
| 276 | status: "green", | |
| 277 | detail: `Latest: ${applied}`, | |
| 278 | }; | |
| 279 | } catch (err) { | |
| 280 | return { | |
| 281 | category: "Database", | |
| 282 | name: "Migrations applied", | |
| 283 | status: "yellow", | |
| 284 | detail: `Couldn't read migration state: ${(err as Error).message.slice(0, 100)}`, | |
| 285 | }; | |
| 286 | } | |
| 287 | } | |
| 288 | ||
| 289 | async function checkAutoMerge(): Promise<CheckResult> { | |
| a3b6378 | 290 | // Resolve which repo to check. SELF_HOST_REPO is the canonical |
| 291 | // "this is the platform's own repo" pointer — falling back to | |
| 292 | // `ccantynz/Gluecron.com` keeps the legacy default behaviour. | |
| 293 | // Without this, the check used to hardcode `ccantynz` and report | |
| 294 | // "Owner not found" for installs where the canonical owner is | |
| 295 | // `ccantynz-alt` or anything else. | |
| 296 | const selfRepo = process.env.SELF_HOST_REPO || "ccantynz/Gluecron.com"; | |
| 297 | const [ownerName, repoName] = selfRepo.includes("/") | |
| 298 | ? selfRepo.split("/") | |
| 299 | : [selfRepo, "Gluecron.com"]; | |
| 826eccf | 300 | try { |
| 301 | const [owner] = await db | |
| 302 | .select({ id: users.id }) | |
| 303 | .from(users) | |
| a3b6378 | 304 | .where(eq(users.username, ownerName)) |
| 826eccf | 305 | .limit(1); |
| 306 | if (!owner) { | |
| 307 | return { | |
| 308 | category: "Auto-merge", | |
| 309 | name: "main protection", | |
| 310 | status: "yellow", | |
| a3b6378 | 311 | detail: `Owner user '${ownerName}' not found in users table (looked up via SELF_HOST_REPO).`, |
| 312 | fix: "Set SELF_HOST_REPO=<actual-owner>/<repo> in /etc/gluecron.env, or register the owner.", | |
| 826eccf | 313 | }; |
| 314 | } | |
| 315 | const [repo] = await db | |
| 316 | .select({ id: repositories.id }) | |
| 317 | .from(repositories) | |
| 318 | .where( | |
| 319 | and( | |
| 320 | eq(repositories.ownerId, owner.id), | |
| a3b6378 | 321 | eq(repositories.name, repoName) |
| 826eccf | 322 | ) |
| 323 | ) | |
| 324 | .limit(1); | |
| 325 | if (!repo) { | |
| 326 | return { | |
| 327 | category: "Auto-merge", | |
| 328 | name: "main protection", | |
| 329 | status: "yellow", | |
| a3b6378 | 330 | detail: `Repository row for ${ownerName}/${repoName} not found. Either the platform repo isn't registered in its own DB, or SELF_HOST_REPO points at the wrong owner/name.`, |
| 331 | fix: `Create the repo at /new (owner=${ownerName}, name=${repoName}), or correct SELF_HOST_REPO in /etc/gluecron.env.`, | |
| 826eccf | 332 | }; |
| 333 | } | |
| 334 | const [bp] = await db | |
| 335 | .select({ enableAutoMerge: branchProtection.enableAutoMerge }) | |
| 336 | .from(branchProtection) | |
| 337 | .where( | |
| 338 | and( | |
| 339 | eq(branchProtection.repositoryId, repo.id), | |
| 340 | eq(branchProtection.pattern, "main") | |
| 341 | ) | |
| 342 | ) | |
| 343 | .limit(1); | |
| 344 | if (!bp) { | |
| 345 | return { | |
| 346 | category: "Auto-merge", | |
| 347 | name: "main protection", | |
| 348 | status: "yellow", | |
| 349 | detail: "No branch_protection row for main yet.", | |
| a3b6378 | 350 | fix: `Visit /${ownerName}/${repoName}/gates/protection to configure.`, |
| 826eccf | 351 | }; |
| 352 | } | |
| 353 | return { | |
| 354 | category: "Auto-merge", | |
| 355 | name: "main protection", | |
| 356 | status: bp.enableAutoMerge ? "green" : "yellow", | |
| 357 | detail: bp.enableAutoMerge | |
| 358 | ? "Auto-merge ENABLED on main." | |
| 359 | : "Auto-merge DISABLED on main.", | |
| 360 | fix: bp.enableAutoMerge | |
| 361 | ? undefined | |
| 362 | : "Visit /admin/ops to enable.", | |
| 363 | }; | |
| 364 | } catch (err) { | |
| 365 | return { | |
| 366 | category: "Auto-merge", | |
| 367 | name: "main protection", | |
| 368 | status: "yellow", | |
| 369 | detail: `Couldn't read: ${(err as Error).message.slice(0, 100)}`, | |
| 370 | }; | |
| 371 | } | |
| 372 | } | |
| 373 | ||
| 374 | async function checkSyntheticMonitor(): Promise<CheckResult> { | |
| 375 | try { | |
| 376 | const oneHourAgo = new Date(Date.now() - 60 * 60 * 1000); | |
| 377 | const reds = await db | |
| 378 | .select({ name: syntheticChecks.checkName }) | |
| 379 | .from(syntheticChecks) | |
| 380 | .where( | |
| 381 | and( | |
| 382 | eq(syntheticChecks.status, "red"), | |
| 383 | gt(syntheticChecks.checkedAt, oneHourAgo) | |
| 384 | ) | |
| 385 | ) | |
| 386 | .orderBy(desc(syntheticChecks.checkedAt)) | |
| 387 | .limit(10); | |
| 388 | if (reds.length === 0) { | |
| 389 | return { | |
| 390 | category: "Monitor", | |
| 391 | name: "Synthetic checks (1h)", | |
| 392 | status: "green", | |
| 393 | detail: "All synthetic checks green in the last hour.", | |
| 394 | }; | |
| 395 | } | |
| 396 | const names = Array.from(new Set(reds.map((r) => r.name))).slice(0, 5); | |
| 397 | return { | |
| 398 | category: "Monitor", | |
| 399 | name: "Synthetic checks (1h)", | |
| 400 | status: "red", | |
| 401 | detail: `Red in last hour: ${names.join(", ")}.`, | |
| 402 | fix: "Open /admin/status for the full row table.", | |
| 403 | }; | |
| 404 | } catch (err) { | |
| 405 | return { | |
| 406 | category: "Monitor", | |
| 407 | name: "Synthetic checks (1h)", | |
| 408 | status: "yellow", | |
| 409 | detail: `Couldn't read: ${(err as Error).message.slice(0, 100)}`, | |
| 410 | }; | |
| 411 | } | |
| 412 | } | |
| 413 | ||
| 414 | function checkSelfHost(): CheckResult { | |
| 415 | const repo = process.env.SELF_HOST_REPO; | |
| 416 | if (!repo) { | |
| 417 | return { | |
| 418 | category: "Self-host", | |
| 419 | name: "Bootstrap", | |
| 420 | status: "yellow", | |
| 421 | detail: "SELF_HOST_REPO unset — pushes to this repo don't trigger self-deploy.", | |
| 422 | fix: "Run scripts/self-host-bootstrap.ts on the box and add SELF_HOST_REPO=ccantynz/Gluecron.com to /etc/gluecron.env.", | |
| 423 | }; | |
| 424 | } | |
| 425 | return { | |
| 426 | category: "Self-host", | |
| 427 | name: "Bootstrap", | |
| 428 | status: "green", | |
| 429 | detail: `Self-hosting ${repo} — push to main fires self-deploy.sh.`, | |
| 430 | }; | |
| 431 | } | |
| 432 | ||
| 115c66b | 433 | // ─── New checks (2026-05-16 reliability sweep) ─────────────────────────── |
| 434 | ||
| 435 | /** | |
| 436 | * Is the autopilot loop ticking on schedule? If not, half the platform's | |
| 437 | * self-healing breaks silently — mirror sync, advisory rescans, scheduled | |
| 438 | * workflows, auto-merge sweep, stale-sweep all skip. | |
| 439 | */ | |
| 440 | function checkAutopilot(): CheckResult { | |
| 441 | if (process.env.AUTOPILOT_DISABLED === "1") { | |
| 442 | return { | |
| 443 | category: "Autopilot", | |
| 444 | name: "Background loop", | |
| 445 | status: "yellow", | |
| 446 | detail: "AUTOPILOT_DISABLED=1 — background maintenance loop is OFF.", | |
| 447 | fix: "Remove or unset AUTOPILOT_DISABLED in /etc/gluecron.env to re-enable.", | |
| 448 | }; | |
| 449 | } | |
| 450 | const total = getTickCount(); | |
| 451 | const tick = getLastTick(); | |
| 452 | const intervalRaw = process.env.AUTOPILOT_INTERVAL_MS; | |
| 453 | const intervalMs = | |
| 454 | intervalRaw && Number.isFinite(Number(intervalRaw)) && Number(intervalRaw) > 0 | |
| 455 | ? Number(intervalRaw) | |
| 456 | : 5 * 60 * 1000; | |
| 457 | // Allow 2x the interval before flagging — accounts for slow ticks. | |
| 458 | const staleMs = intervalMs * 2; | |
| 459 | if (!tick) { | |
| 460 | if (total === 0) { | |
| 461 | return { | |
| 462 | category: "Autopilot", | |
| 463 | name: "Background loop", | |
| 464 | status: "yellow", | |
| 465 | detail: `Loop is enabled but has not ticked yet. First tick fires after ${Math.round(intervalMs / 1000)}s.`, | |
| 466 | }; | |
| 467 | } | |
| 468 | return { | |
| 469 | category: "Autopilot", | |
| 470 | name: "Background loop", | |
| 471 | status: "red", | |
| 472 | detail: `${total} tick(s) recorded but last tick result is missing — loop may have crashed.`, | |
| 473 | fix: "Check journalctl -u gluecron for [autopilot] errors. Run a tick manually at /admin/autopilot.", | |
| 474 | }; | |
| 475 | } | |
| 476 | const finishedAt = new Date(tick.finishedAt).getTime(); | |
| 477 | const ageMs = Date.now() - finishedAt; | |
| 478 | if (ageMs > staleMs) { | |
| 479 | return { | |
| 480 | category: "Autopilot", | |
| 481 | name: "Background loop", | |
| 482 | status: "red", | |
| 483 | detail: `Last tick was ${Math.round(ageMs / 1000)}s ago (interval is ${Math.round(intervalMs / 1000)}s). Loop is stalled.`, | |
| 484 | fix: "Run a tick manually at /admin/autopilot. Check journalctl for [autopilot] errors.", | |
| 485 | }; | |
| 486 | } | |
| 487 | const failed = tick.tasks.filter((t) => !t.ok).length; | |
| 488 | if (failed > 0) { | |
| 489 | return { | |
| 490 | category: "Autopilot", | |
| 491 | name: "Background loop", | |
| 492 | status: "yellow", | |
| 493 | detail: `Loop running but ${failed}/${tick.tasks.length} tasks failed in the last tick.`, | |
| 494 | fix: "Open /admin/autopilot for the per-task error list.", | |
| 495 | }; | |
| 496 | } | |
| 497 | return { | |
| 498 | category: "Autopilot", | |
| 499 | name: "Background loop", | |
| 500 | status: "green", | |
| 501 | detail: `Ticking on schedule (${total} tick${total === 1 ? "" : "s"} this process; last ${Math.round(ageMs / 1000)}s ago).`, | |
| 502 | }; | |
| 503 | } | |
| 504 | ||
| 505 | /** | |
| 506 | * When did we last successfully deploy? Stale deploys are an early | |
| 507 | * warning sign the deploy pipeline is broken silently (which is exactly | |
| 508 | * what happened on 2026-05-15 — 17 hours of failed deploys, no alert). | |
| 509 | */ | |
| 510 | async function checkRecentDeploy(): Promise<CheckResult> { | |
| 511 | try { | |
| 512 | const [latest] = await db | |
| 513 | .select({ | |
| 514 | sha: platformDeploys.sha, | |
| 515 | status: platformDeploys.status, | |
| 516 | startedAt: platformDeploys.startedAt, | |
| 517 | finishedAt: platformDeploys.finishedAt, | |
| 518 | error: platformDeploys.error, | |
| 519 | }) | |
| 520 | .from(platformDeploys) | |
| 521 | .orderBy(desc(platformDeploys.startedAt)) | |
| 522 | .limit(1); | |
| 523 | if (!latest) { | |
| 524 | return { | |
| 525 | category: "Deploy", | |
| 526 | name: "Latest deploy", | |
| 527 | status: "yellow", | |
| 528 | detail: "No deploys recorded yet. The hetzner-deploy.yml workflow posts events to /api/events/deploy/* — set DEPLOY_EVENT_TOKEN in the workflow env to enable.", | |
| 529 | }; | |
| 530 | } | |
| 531 | const ref = latest.finishedAt || latest.startedAt; | |
| 532 | const ageHours = (Date.now() - new Date(ref).getTime()) / (60 * 60 * 1000); | |
| 533 | const sha7 = (latest.sha || "").slice(0, 7); | |
| 534 | if (latest.status === "failed") { | |
| 535 | return { | |
| 536 | category: "Deploy", | |
| 537 | name: "Latest deploy", | |
| 538 | status: "red", | |
| 539 | detail: `Last deploy (${sha7}) FAILED ${ageHours.toFixed(1)}h ago: ${(latest.error || "no error message").slice(0, 200)}.`, | |
| 540 | fix: "Open /admin/deploys for the run timeline. Trigger a new deploy after fixing.", | |
| 541 | }; | |
| 542 | } | |
| 543 | if (latest.status === "in_progress") { | |
| 544 | return { | |
| 545 | category: "Deploy", | |
| 546 | name: "Latest deploy", | |
| 547 | status: "yellow", | |
| 548 | detail: `Deploy in progress (${sha7}, started ${ageHours.toFixed(1)}h ago).`, | |
| 549 | }; | |
| 550 | } | |
| 551 | if (latest.status === "succeeded" && ageHours > 48) { | |
| 552 | return { | |
| 553 | category: "Deploy", | |
| 554 | name: "Latest deploy", | |
| 555 | status: "yellow", | |
| 556 | detail: `Last deploy was ${sha7} ${ageHours.toFixed(1)}h ago. If you pushed to main since then, the deploy pipeline may have silently failed.`, | |
| 557 | fix: "Check the GitHub Actions Hetzner deploy run for the latest main commit.", | |
| 558 | }; | |
| 559 | } | |
| 560 | return { | |
| 561 | category: "Deploy", | |
| 562 | name: "Latest deploy", | |
| 563 | status: "green", | |
| 564 | detail: `${sha7} deployed cleanly ${ageHours.toFixed(1)}h ago.`, | |
| 565 | }; | |
| 566 | } catch (err) { | |
| 567 | return { | |
| 568 | category: "Deploy", | |
| 569 | name: "Latest deploy", | |
| 570 | status: "yellow", | |
| 571 | detail: `Couldn't read platform_deploys: ${(err as Error).message.slice(0, 100)}`, | |
| 572 | }; | |
| 573 | } | |
| 574 | } | |
| 575 | ||
| 576 | /** | |
| 577 | * Is the workflow worker draining the queue? A backed-up queue or a | |
| 578 | * stuck queued row means CI gates aren't firing. | |
| 579 | */ | |
| 580 | async function checkWorkflowQueue(): Promise<CheckResult> { | |
| 581 | try { | |
| 582 | const [queued] = await db | |
| 583 | .select({ n: sql<number>`count(*)::int` }) | |
| 584 | .from(workflowRuns) | |
| 585 | .where(eq(workflowRuns.status, "queued")); | |
| 586 | const queuedN = Number(queued?.n || 0); | |
| 587 | const [running] = await db | |
| 588 | .select({ n: sql<number>`count(*)::int` }) | |
| 589 | .from(workflowRuns) | |
| 590 | .where(eq(workflowRuns.status, "running")); | |
| 591 | const runningN = Number(running?.n || 0); | |
| 592 | if (queuedN > 25) { | |
| 593 | return { | |
| 594 | category: "Workflows", | |
| 595 | name: "Run queue", | |
| 596 | status: "red", | |
| 597 | detail: `${queuedN} runs queued (running: ${runningN}). The worker is backed up.`, | |
| 598 | fix: "Check journalctl for [workflow-runner] errors. Restart gluecron if persistent.", | |
| 599 | }; | |
| 600 | } | |
| 601 | if (queuedN > 5) { | |
| 602 | return { | |
| 603 | category: "Workflows", | |
| 604 | name: "Run queue", | |
| 605 | status: "yellow", | |
| 606 | detail: `${queuedN} runs queued, ${runningN} running. Worker may be slow.`, | |
| 607 | }; | |
| 608 | } | |
| 609 | return { | |
| 610 | category: "Workflows", | |
| 611 | name: "Run queue", | |
| 612 | status: "green", | |
| 613 | detail: `${queuedN} queued, ${runningN} running.`, | |
| 614 | }; | |
| 615 | } catch (err) { | |
| 616 | return { | |
| 617 | category: "Workflows", | |
| 618 | name: "Run queue", | |
| 619 | status: "yellow", | |
| 620 | detail: `Couldn't read workflow_runs: ${(err as Error).message.slice(0, 100)}`, | |
| 621 | }; | |
| 622 | } | |
| 623 | } | |
| 624 | ||
| 625 | /** | |
| 626 | * Crontech deploy webhook secret — without it, the webhook POSTs | |
| 627 | * unsigned and Crontech rejects with 401, but our hook side never sees | |
| 628 | * the rejection because the request is fire-and-forget. | |
| 629 | */ | |
| 630 | function checkCrontechWebhook(): CheckResult { | |
| 631 | const url = process.env.CRONTECH_DEPLOY_URL; | |
| 632 | const secret = process.env.CRONTECH_HMAC_SECRET; | |
| 633 | if (!url) { | |
| 634 | return { | |
| 635 | category: "Crontech", | |
| 636 | name: "Deploy webhook", | |
| 637 | status: "yellow", | |
| 638 | detail: "CRONTECH_DEPLOY_URL unset — pushes to the Crontech repo don't notify the deploy pipeline.", | |
| 639 | fix: "Optional integration. Set CRONTECH_DEPLOY_URL + CRONTECH_HMAC_SECRET if you want push-triggered Crontech deploys.", | |
| 640 | }; | |
| 641 | } | |
| 642 | if (!secret) { | |
| 643 | return { | |
| 644 | category: "Crontech", | |
| 645 | name: "Deploy webhook", | |
| 646 | status: "red", | |
| 647 | detail: "CRONTECH_DEPLOY_URL set but CRONTECH_HMAC_SECRET empty — webhook will be rejected as unsigned.", | |
| 648 | fix: "Add CRONTECH_HMAC_SECRET to /etc/gluecron.env (match the value configured on Crontech's side).", | |
| 649 | }; | |
| 650 | } | |
| 651 | return { | |
| 652 | category: "Crontech", | |
| 653 | name: "Deploy webhook", | |
| 654 | status: "green", | |
| 655 | detail: `Configured (POST to ${url}).`, | |
| 656 | }; | |
| 657 | } | |
| 658 | ||
| 826eccf | 659 | // ─── Page handler ──────────────────────────────────────────────────────── |
| 660 | ||
| 661 | function pill(status: CheckStatus): any { | |
| 662 | const map: Record<CheckStatus, { bg: string; fg: string; label: string }> = { | |
| 663 | green: { bg: "rgba(52,211,153,0.16)", fg: "#34d399", label: "✓ OK" }, | |
| 664 | yellow: { bg: "rgba(245,158,11,0.16)", fg: "#f59e0b", label: "! WARN" }, | |
| 665 | red: { bg: "rgba(248,113,113,0.16)", fg: "#f87171", label: "× FAIL" }, | |
| 666 | }; | |
| 667 | const s = map[status]; | |
| 668 | return ( | |
| 669 | <span | |
| 670 | style={`display:inline-flex;align-items:center;padding:2px 10px;border-radius:999px;font-size:12px;font-weight:600;background:${s.bg};color:${s.fg};white-space:nowrap`} | |
| 671 | > | |
| 672 | {s.label} | |
| 673 | </span> | |
| 674 | ); | |
| 675 | } | |
| 676 | ||
| 115c66b | 677 | async function runAllChecks(c: any): Promise<CheckResult[]> { |
| 678 | return [ | |
| 826eccf | 679 | checkEmail(), |
| 680 | checkAnthropic(), | |
| 681 | checkGateTest(), | |
| 682 | checkBuildSha(), | |
| 683 | checkAppBaseUrl(c), | |
| 684 | checkDatabase(), | |
| 685 | await checkMigrations(), | |
| 686 | await checkAutoMerge(), | |
| 687 | await checkSyntheticMonitor(), | |
| 688 | checkSelfHost(), | |
| 115c66b | 689 | // 2026-05-16 reliability sweep additions: |
| 690 | checkAutopilot(), | |
| 691 | await checkRecentDeploy(), | |
| 692 | await checkWorkflowQueue(), | |
| 693 | checkCrontechWebhook(), | |
| 826eccf | 694 | ]; |
| 115c66b | 695 | } |
| 696 | ||
| 697 | // JSON endpoint for programmatic monitoring. Same gate as the HTML page | |
| 698 | // (site-admin only) so deploy state isn't public. | |
| 699 | diagnose.get("/admin/diagnose.json", async (c) => { | |
| 700 | const g = await gate(c); | |
| 701 | if (g instanceof Response) return g; | |
| 702 | const results = await runAllChecks(c); | |
| 703 | const counts = { | |
| 704 | green: results.filter((r) => r.status === "green").length, | |
| 705 | yellow: results.filter((r) => r.status === "yellow").length, | |
| 706 | red: results.filter((r) => r.status === "red").length, | |
| 707 | }; | |
| 708 | const overall = | |
| 709 | counts.red > 0 ? "red" : counts.yellow > 0 ? "yellow" : "green"; | |
| 710 | return c.json({ | |
| 711 | ok: true, | |
| 712 | overall, | |
| 713 | counts, | |
| 714 | checks: results, | |
| 715 | asOf: new Date().toISOString(), | |
| 716 | }); | |
| 717 | }); | |
| 718 | ||
| 719 | // /admin/health alias — same handler, friendlier URL. The user expected | |
| 720 | // this to exist; making the expectation reality is cheaper than arguing | |
| 721 | // about naming. | |
| 722 | diagnose.get("/admin/health", async (c) => { | |
| 723 | return c.redirect("/admin/diagnose"); | |
| 724 | }); | |
| 725 | ||
| 726 | diagnose.get("/admin/diagnose", async (c) => { | |
| 727 | const g = await gate(c); | |
| 728 | if (g instanceof Response) return g; | |
| 729 | const { user } = g; | |
| 730 | ||
| 731 | const results: CheckResult[] = await runAllChecks(c); | |
| 826eccf | 732 | |
| 733 | const counts = { | |
| 734 | green: results.filter((r) => r.status === "green").length, | |
| 735 | yellow: results.filter((r) => r.status === "yellow").length, | |
| 736 | red: results.filter((r) => r.status === "red").length, | |
| 737 | }; | |
| 738 | const headline = | |
| 739 | counts.red > 0 | |
| 740 | ? `${counts.red} failing` | |
| 741 | : counts.yellow > 0 | |
| 742 | ? `${counts.yellow} needs attention` | |
| 743 | : "All green"; | |
| 744 | const headlineColor = | |
| 745 | counts.red > 0 | |
| 746 | ? "var(--red, #cf222e)" | |
| 747 | : counts.yellow > 0 | |
| 748 | ? "#d97706" | |
| 749 | : "var(--green, #2da44e)"; | |
| 750 | ||
| 751 | const flash = c.req.query("test_email"); | |
| 752 | ||
| 753 | return c.html( | |
| 754 | <Layout title="Diagnose — admin" user={user}> | |
| 755 | <div style="max-width:920px;margin:0 auto;padding:var(--space-5) var(--space-3)"> | |
| 756 | <div style="display:flex;align-items:center;justify-content:space-between;margin-bottom:var(--space-3)"> | |
| 757 | <h1 style="margin:0">Diagnose</h1> | |
| 758 | <a href="/admin" class="btn btn-sm"> | |
| 759 | Back to admin | |
| 760 | </a> | |
| 761 | </div> | |
| 762 | <p style="color:var(--text-muted);margin-bottom:var(--space-4)"> | |
| 763 | Read-only health scan of every config knob the platform depends on. | |
| 764 | One row per check. Green = wired, yellow = degraded, red = broken. | |
| 765 | </p> | |
| 766 | ||
| 767 | <div | |
| 768 | style={`background:var(--bg-elevated);border:1px solid var(--border);border-radius:8px;padding:var(--space-3) var(--space-4);margin-bottom:var(--space-4);display:flex;align-items:center;gap:var(--space-3)`} | |
| 769 | > | |
| 770 | <span | |
| 771 | style={`display:inline-block;width:14px;height:14px;border-radius:50%;background:${headlineColor}`} | |
| 772 | /> | |
| 773 | <strong style="font-size:18px">{headline}</strong> | |
| 774 | <span style="color:var(--text-muted);font-size:13px"> | |
| 775 | {counts.green} green · {counts.yellow} warn · {counts.red} fail | |
| 776 | </span> | |
| 777 | </div> | |
| 778 | ||
| 779 | {flash && ( | |
| 780 | <div | |
| 781 | class={flash === "ok" ? "banner" : "auth-error"} | |
| 782 | style="margin-bottom:var(--space-4)" | |
| 783 | > | |
| 784 | {flash === "ok" | |
| 785 | ? "Test email dispatched. If the provider is 'log' you'll see it in journalctl, not your inbox." | |
| 786 | : `Test email failed: ${decodeURIComponent(flash)}`} | |
| 787 | </div> | |
| 788 | )} | |
| 789 | ||
| 790 | <div | |
| 791 | style="background:var(--bg-elevated);border:1px solid var(--border);border-radius:8px;overflow:hidden;margin-bottom:var(--space-4)" | |
| 792 | > | |
| 793 | {results.map((r, i) => ( | |
| 794 | <div | |
| 795 | style={`padding:var(--space-3) var(--space-4);${i < results.length - 1 ? "border-bottom:1px solid var(--border);" : ""}display:grid;grid-template-columns:120px 80px 1fr;gap:var(--space-3);align-items:start`} | |
| 796 | > | |
| 797 | <div> | |
| 798 | <div style="font-size:11px;color:var(--text-muted);text-transform:uppercase;letter-spacing:0.04em"> | |
| 799 | {r.category} | |
| 800 | </div> | |
| 801 | <div style="font-weight:600;font-size:14px;margin-top:2px"> | |
| 802 | {r.name} | |
| 803 | </div> | |
| 804 | </div> | |
| 805 | <div>{pill(r.status)}</div> | |
| 806 | <div> | |
| 807 | <div style="font-size:13px;line-height:1.45">{r.detail}</div> | |
| 808 | {r.fix && ( | |
| 809 | <div | |
| 810 | style="font-size:12px;color:var(--text-muted);margin-top:4px;line-height:1.4" | |
| 811 | > | |
| 812 | → {r.fix} | |
| 813 | </div> | |
| 814 | )} | |
| 815 | </div> | |
| 816 | </div> | |
| 817 | ))} | |
| 818 | </div> | |
| 819 | ||
| 820 | <div | |
| 821 | style="background:var(--bg-elevated);border:1px solid var(--border);border-radius:8px;padding:var(--space-3) var(--space-4)" | |
| 822 | > | |
| 823 | <h3 | |
| 824 | style="margin:0 0 var(--space-2) 0;font-size:14px;letter-spacing:0.04em;text-transform:uppercase;color:var(--text-muted)" | |
| 825 | > | |
| 826 | Test email delivery | |
| 827 | </h3> | |
| 828 | <p style="font-size:13px;color:var(--text-muted);margin:0 0 var(--space-3) 0"> | |
| 829 | Fires a one-line test email to <strong>{user.email}</strong> using | |
| 830 | the configured provider. If EMAIL_PROVIDER=log it appears in | |
| 831 | journalctl; if resend, in your inbox in <30s. | |
| 832 | </p> | |
| 833 | <form method="post" action="/admin/diagnose/test-email" style="margin:0"> | |
| 834 | <input | |
| 835 | type="hidden" | |
| 836 | name="_csrf" | |
| 837 | value={(c.get("csrfToken") as string | undefined) || ""} | |
| 838 | /> | |
| 839 | <button type="submit" class="btn btn-sm btn-primary"> | |
| 840 | Send test email | |
| 841 | </button> | |
| 842 | </form> | |
| 843 | </div> | |
| 844 | </div> | |
| 845 | </Layout> | |
| 846 | ); | |
| 847 | }); | |
| 848 | ||
| 849 | diagnose.post("/admin/diagnose/test-email", async (c) => { | |
| 850 | const g = await gate(c); | |
| 851 | if (g instanceof Response) return g; | |
| 852 | const { user } = g; | |
| 853 | if (!user.email) { | |
| 854 | return c.redirect( | |
| 855 | `/admin/diagnose?test_email=${encodeURIComponent("admin has no email on record")}` | |
| 856 | ); | |
| 857 | } | |
| 858 | const stamp = new Date().toISOString(); | |
| 859 | const result = await sendEmail({ | |
| 860 | to: user.email, | |
| 861 | subject: "Gluecron — diagnose test email", | |
| 862 | text: | |
| 863 | `This is a test email from /admin/diagnose at ${stamp}.\n\n` + | |
| 864 | `If you received this in your inbox, EMAIL_PROVIDER=resend is wired correctly.\n` + | |
| 865 | `If you only see it in journalctl, EMAIL_PROVIDER is still 'log'.\n`, | |
| 866 | }); | |
| 867 | if (!result.ok) { | |
| 868 | return c.redirect( | |
| 869 | `/admin/diagnose?test_email=${encodeURIComponent(result.error || result.skipped || "unknown failure")}` | |
| 870 | ); | |
| 871 | } | |
| 872 | return c.redirect("/admin/diagnose?test_email=ok"); | |
| 873 | }); | |
| 874 | ||
| 875 | export default diagnose; |