test: add remote control stress budgets

scripts/stress-remote-control.mjs

@@ -20,6 +20,9 @@ function parseArgs(argv) {
     timeoutMs: 45_000,
     skipChain: false,
     skipRuntime: false,
+    reportJson: null,
+    maxChainP95Ms: null,
+    maxRuntimeP95Ms: null,
   };
 
   for (const arg of argv) {
@@ -31,7 +34,12 @@ function parseArgs(argv) {
       options.runtimeConcurrency = positiveInt(arg.split("=")[1], options.runtimeConcurrency);
     } else if (arg.startsWith("--poll-ms=")) options.pollMs = positiveInt(arg.split("=")[1], options.pollMs);
     else if (arg.startsWith("--timeout-ms=")) options.timeoutMs = positiveInt(arg.split("=")[1], options.timeoutMs);
-    else if (arg === "--help" || arg === "-h") {
+    else if (arg.startsWith("--report-json=")) options.reportJson = arg.slice("--report-json=".length);
+    else if (arg.startsWith("--max-chain-p95-ms=")) {
+      options.maxChainP95Ms = positiveInt(arg.split("=")[1], options.maxChainP95Ms);
+    } else if (arg.startsWith("--max-runtime-p95-ms=")) {
+      options.maxRuntimeP95Ms = positiveInt(arg.split("=")[1], options.maxRuntimeP95Ms);
+    } else if (arg === "--help" || arg === "-h") {
       options.help = true;
     }
   }
@@ -453,6 +461,29 @@ function hasFailure(summary) {
   return summary.failed > 0;
 }
 
+function findThresholdFailures(summaries, options) {
+  const failures = [];
+  const chain = summaries.find((summary) => summary.name === "chain");
+  if (chain && options.maxChainP95Ms && chain.latencyMs.p95 > options.maxChainP95Ms) {
+    failures.push({
+      name: "chain_p95_latency",
+      actualMs: chain.latencyMs.p95,
+      maxMs: options.maxChainP95Ms,
+    });
+  }
+
+  const runtime = summaries.find((summary) => summary.name === "runtime");
+  if (runtime && options.maxRuntimeP95Ms && runtime.latencyMs.p95 > options.maxRuntimeP95Ms) {
+    failures.push({
+      name: "runtime_p95_latency",
+      actualMs: runtime.latencyMs.p95,
+      maxMs: options.maxRuntimeP95Ms,
+    });
+  }
+
+  return failures;
+}
+
 function printHelp() {
   console.log(`Usage: node scripts/stress-remote-control.mjs [options]
 
@@ -462,6 +493,9 @@ Options:
   --runtime-concurrency=N     direct runtime concurrency, default 24
   --poll-ms=N                 local-agent task poll interval, default 5
   --timeout-ms=N              chain stress timeout, default 45000
+  --report-json=PATH          write the full stress report to PATH
+  --max-chain-p95-ms=N        fail when local-agent chain p95 latency is above N
+  --max-runtime-p95-ms=N      fail when direct runtime p95 latency is above N
   --skip-chain                skip local-agent chain stress
   --skip-runtime              skip direct runtime stress
 `);
@@ -473,6 +507,8 @@ if (options.help) {
   process.exit(0);
 }
 
+const startedAt = new Date();
+const startedMs = Date.now();
 const summaries = [];
 if (!options.skipChain) {
   summaries.push(await runChainStress(options));
@@ -481,7 +517,23 @@ if (!options.skipRuntime) {
   summaries.push(await runRuntimeStress(options));
 }
 
-console.log(JSON.stringify({ ok: summaries.every((summary) => !hasFailure(summary)), summaries }, null, 2));
-if (summaries.some(hasFailure)) {
+const thresholdFailures = findThresholdFailures(summaries, options);
+const report = {
+  ok: summaries.every((summary) => !hasFailure(summary)) && thresholdFailures.length === 0,
+  startedAt: startedAt.toISOString(),
+  finishedAt: new Date().toISOString(),
+  durationMs: Date.now() - startedMs,
+  options,
+  summaries,
+  thresholdFailures,
+};
+
+if (options.reportJson) {
+  await mkdir(path.dirname(path.resolve(options.reportJson)), { recursive: true });
+  await writeFile(path.resolve(options.reportJson), `${JSON.stringify(report, null, 2)}\n`, "utf8");
+}
+
+console.log(JSON.stringify(report, null, 2));
+if (!report.ok) {
   process.exitCode = 1;
 }