apiblaze 0.20.3 → 0.20.4

package/dist/index.js

@@ -932,7 +932,7 @@ var import_commander = require("commander");
 var import_chalk51 = __toESM(require("chalk"));
 
 // package.json
-var version = "0.20.3";
+var version = "0.20.4";
 
 // src/index.ts
 init_types();
@@ -9888,7 +9888,7 @@ function printResidue(report, applied) {
   for (const e of ghosts?.errors ?? []) console.log(import_chalk50.default.red(`    error: ${e}`));
   console.log();
 }
-async function runOp(sub, opts = {}) {
+async function runOp(sub, opts = {}, view) {
   if (!loadCredentials()) {
     console.log(import_chalk50.default.dim("Not logged in. Run `apiblaze login`."));
     return;
@@ -9904,6 +9904,9 @@ async function runOp(sub, opts = {}) {
       console.log(`  ${import_chalk50.default.cyan("apiblaze op residue")}        external-store residue report (Upstash + Neon/OpenFGA, dry-run)`);
       console.log(`  ${import_chalk50.default.cyan("apiblaze op sweep")}          delete the orphans the report shows (asks first; ${import_chalk50.default.dim("-y to skip")})`);
       console.log(`  ${import_chalk50.default.cyan("apiblaze op credits")}        list credit wallets`);
+      console.log(`  ${import_chalk50.default.cyan("apiblaze op latency")}        which FEATURE is slow, and what inside it ${import_chalk50.default.dim("(--hours N --team T --project P)")}`);
+      console.log(`  ${import_chalk50.default.cyan("apiblaze op latency slow")}   the slowest requests, with a cf-ray to drill into`);
+      console.log(`  ${import_chalk50.default.cyan("apiblaze op latency llm")}    chat timing by provider/model, tokens, credit-reserve cost`);
       console.log(import_chalk50.default.dim(`  (to prune all non-CP data: run scripts/nuke-but-cp.sh --apply --sweep in the repo)
 `));
       return;
@@ -9951,6 +9954,93 @@ async function runOp(sub, opts = {}) {
       }
       return;
     }
+    // ── LATENCY (specs/latency/ §7.4) ─────────────────────────────────────
+    // The blame table first: level 1 (which feature) and level 2 (what inside
+    // it) in one row. `upstream` is excluded from the blame column on purpose —
+    // it is the biggest feature on almost every request, so blaming it would
+    // report "upstream, 94%" forever and tell an operator nothing. The
+    // summary's "us or them" split is where upstream shows up.
+    case "latency": {
+      const qs = new URLSearchParams();
+      const FILTERS = [
+        ["hours", "hours"],
+        ["team", "team"],
+        ["project", "project"],
+        ["apiVersion", "api_version"],
+        ["tenant", "tenant"],
+        ["route", "route"],
+        ["environment", "environment"],
+        ["statusClass", "status_class"],
+        ["minMs", "min_ms"]
+      ];
+      for (const [optKey, queryKey] of FILTERS) {
+        const v = opts[optKey];
+        if (v !== void 0 && v !== null && v !== "") qs.set(queryKey, String(v));
+      }
+      const q = qs.toString() ? `?${qs}` : "";
+      const which = view;
+      if (which === "slow") {
+        const data = await opCall({ method: "GET", path: `/operator/latency/slow${q}`, summary: "slowest requests" });
+        if (opts.json) return void console.log(JSON.stringify(data, null, 2));
+        const rows2 = data?.rows ?? [];
+        if (!rows2.length) return void console.log(import_chalk50.default.dim("No requests over the threshold in that window."));
+        console.log(import_chalk50.default.bold(`
+Slowest requests \u2014 last ${data.window_hours}h, over ${data.min_ms}ms
+`));
+        console.log(import_chalk50.default.dim("  total   ours   theirs  blame                 request (cf-ray)"));
+        for (const r of rows2.slice(0, 30)) {
+          console.log(
+            `  ${String(Math.round(r.duration_ms)).padStart(6)}  ${String(Math.round(r.gateway_ms)).padStart(5)}  ${String(Math.round(r.upstream_ttfb_ms)).padStart(6)}  ${import_chalk50.default.yellow(`${r.slow_gw_feature || "-"}\u2192${r.slow_dep || "-"}`.padEnd(20))}  ${import_chalk50.default.dim(r.request_id || "")}`
+          );
+        }
+        console.log(import_chalk50.default.dim(`
+  Take a cf-ray to \`apiblaze logs\` for the exact per-feature breakdown \u2014 that row is unsampled.
+`));
+        return;
+      }
+      if (which === "llm") {
+        const data = await opCall({ method: "GET", path: `/operator/latency/llm${q}`, summary: "llm latency" });
+        if (opts.json) return void console.log(JSON.stringify(data, null, 2));
+        const rows2 = data?.rows ?? [];
+        if (!rows2.length) return void console.log(import_chalk50.default.dim("No LLM traffic in that window."));
+        console.log(import_chalk50.default.bold(`
+LLM timing \u2014 last ${data.window_hours}h
+`));
+        console.log(import_chalk50.default.dim("  requests   gen p95   reserve p95   in/out tokens p95   model"));
+        for (const r of rows2) {
+          console.log(
+            `  ${String(Math.round(r.requests)).padStart(8)}   ${String(Math.round(r.gen_p95_ms)).padStart(6)}ms   ${String(Math.round(r.reserve_p95_ms)).padStart(9)}ms   ${String(Math.round(r.input_tokens_p95)).padStart(6)}/${String(Math.round(r.output_tokens_p95)).padEnd(6)}   ${r.model || "-"}`
+          );
+        }
+        console.log(import_chalk50.default.dim(`
+  ${data.note}
+`));
+        return;
+      }
+      const [blame, summary] = await Promise.all([
+        opCall({ method: "GET", path: `/operator/latency${q}`, summary: "latency blame table" }),
+        opCall({ method: "GET", path: `/operator/latency/summary${q}`, summary: "latency summary" })
+      ]);
+      if (opts.json) return void console.log(JSON.stringify({ blame, summary }, null, 2));
+      const ov = summary?.apiblaze_overhead_ms ?? {};
+      console.log(import_chalk50.default.bold(`
+Latency \u2014 last ${summary?.window_hours ?? "?"}h, ${Number(summary?.requests ?? 0).toLocaleString()} requests
+`));
+      console.log(`  ${import_chalk50.default.bold("apiblaze overhead")}  p50 ${String(ov.p50 ?? 0).padStart(5)}ms   p95 ${String(ov.p95 ?? 0).padStart(6)}ms   p99 ${String(ov.p99 ?? 0).padStart(6)}ms   ${import_chalk50.default.dim("\u2190 ours")}`);
+      console.log(`  ${import_chalk50.default.bold("upstream ttfb    ")}  ${" ".repeat(24)}p95 ${String(summary?.upstream_ttfb_p95_ms ?? 0).padStart(6)}ms   ${import_chalk50.default.dim("\u2190 theirs")}`);
+      console.log(import_chalk50.default.dim(`  (per-request percentiles \u2014 never subtract one from the other)
+`));
+      const rows = blame?.blame ?? [];
+      if (!rows.length) return void console.log(import_chalk50.default.dim("No latency rows in that window."));
+      console.log(import_chalk50.default.bold("  Which feature ate the time, and what inside it\n"));
+      console.log(import_chalk50.default.dim("   share   p95      feature \u2192 dependency"));
+      for (const r of rows.slice(0, 15)) {
+        const share = `${(r.share * 100).toFixed(1)}%`;
+        console.log(`   ${share.padStart(6)}  ${String(r.p95_ms).padStart(6)}ms  ${import_chalk50.default.yellow(r.feature)} ${import_chalk50.default.dim("\u2192")} ${import_chalk50.default.cyan(r.dep)}`);
+      }
+      console.log("");
+      return;
+    }
     default:
       console.log(import_chalk50.default.red(`Unknown op subcommand '${sub}'. Run \`apiblaze op\` for the menu.`));
   }
@@ -10114,7 +10204,7 @@ apikeys.command("list").description("List control-plane API keys in your team").
 apikeys.command("mint").description("Create a control-plane API key (secret shown once)").option("--desc <text>", "Description").option("--expires-days <n>", "Expiry in days (default 90 server-side)").option("--team <id|name>", "Team (defaults to active team)").option("--json", "Output machine-readable JSON").action(action((opts, cmd) => runKeyMint({ ...cmd.parent?.opts(), ...opts })));
 apikeys.command("revoke").description("Revoke a control-plane API key").argument("<keyId>", "Key id (see `apikeys list`)").option("--team <id|name>", "Team (defaults to active team)").action(action((keyId, opts, cmd) => runKeyRevoke(keyId, { ...cmd.parent?.opts(), ...opts })));
 program.addCommand(apikeys, { hidden: true });
-var op = new import_commander.Command("op").description("Operator menu (platform operators only)").argument("[action]", "residue | sweep | credits (omit for the menu)").option("-y, --yes", "Skip the sweep confirmation prompt").option("--json", "Output machine-readable JSON").action(action((sub, opts) => runOp(sub, opts)));
+var op = new import_commander.Command("op").description("Operator menu (platform operators only)").argument("[action]", "residue | sweep | credits | latency (omit for the menu)").argument("[view]", "for `latency`: slow | llm (omit for the blame table)").option("-y, --yes", "Skip the sweep confirmation prompt").option("--json", "Output machine-readable JSON").option("--hours <n>", "Look-back window in hours (default 1)").option("--team <id>", "Filter to one team").option("--project <name>", "Filter to one proxy").option("--api-version <v>", "Filter to one API version").option("--tenant <name>", "Filter to one tenant").option("--route <path>", "Filter to one route").option("--environment <env>", "Filter to one environment").option("--status-class <c>", "Filter to a status class (2xx/4xx/5xx)").option("--min-ms <n>", "For `latency slow`: minimum duration (default 1000)").action(action((sub, view, opts) => runOp(sub, opts, view)));
 program.addCommand(op, { hidden: true });
 var spec = program.command("spec").description("View or update a proxy's OpenAPI spec (or build one by chatting: apiblaze agent openapi)");
 spec.command("get").description("Print the current OpenAPI document").argument("<project>", "Project name or id").option("--team <id|name>", "Team the project is in").option("--apiversion <version>", "API version").option("--json", "Compact JSON output").action(action((project, opts) => runSpecGet(project, opts)));

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "apiblaze",
-  "version": "0.20.3",
+  "version": "0.20.4",
   "description": "APIblaze CLI — Chat with your APIs, Manage your API keys, users and groups with the APIblaze serverless proxy",
   "keywords": [
     "apiblaze",