fix ssh tcp pool transient diagnostics

scripts/src/jobs.ts

@@ -38,6 +38,7 @@ export interface JobProgressSummary {
   eventsObserved: number;
   slow: boolean;
   warnings: string[];
+  diagnostics?: Record<string, unknown> | null;
   timings: Record<string, number>;
   summary: string;
   nextCommand: string | null;
@@ -437,6 +438,7 @@ function summarizeRuntimeLaneTriggerJobProgress(job: JobRecord, stdoutTail: stri
     stageElapsedSeconds,
     lastEventAgeSeconds,
   });
+  const tcpPoolDiagnostics = job.status === "succeeded" ? null : sshTcpPoolDiagnosticsFromJobText(`${stderrTail}\n${stdoutTail}`);
   const slow = warnings.length > 0;
   return {
     kind: "hwlab-runtime-lane-trigger",
@@ -452,6 +454,7 @@ function summarizeRuntimeLaneTriggerJobProgress(job: JobRecord, stdoutTail: stri
     eventsObserved: events.length,
     slow,
     warnings,
+    diagnostics: tcpPoolDiagnostics,
     timings,
     summary: [
       job.status,
@@ -462,6 +465,7 @@ function summarizeRuntimeLaneTriggerJobProgress(job: JobRecord, stdoutTail: stri
       elapsedSeconds !== null ? `elapsed=${elapsedSeconds}s` : null,
       stageElapsedSeconds !== null && job.status === "running" ? `stageElapsed=${stageElapsedSeconds}s` : null,
       lastEventAgeSeconds !== null && job.status === "running" ? `lastEventAge=${lastEventAgeSeconds}s` : null,
+      tcpPoolDiagnostics !== null ? `sshPool=${String(tcpPoolDiagnostics.failureKind ?? "transient")}` : null,
       slow ? "visibility-warning" : null,
     ].filter(Boolean).join(" "),
     nextCommand: pipelineRun
@@ -472,6 +476,47 @@ function summarizeRuntimeLaneTriggerJobProgress(job: JobRecord, stdoutTail: stri
   };
 }
 
+function sshTcpPoolDiagnosticsFromJobText(text: string): Record<string, unknown> | null {
+  const hint = lastJsonLinePayload(text, "UNIDESK_SSH_TCP_POOL_HINT");
+  if (hint !== null) return hint;
+  const error = lastJsonLinePayload(text, "UNIDESK_SSH_ERROR");
+  if (error !== null && typeof error.failureKind === "string" && error.failureKind.startsWith("provider-data-")) {
+    return {
+      code: "ssh-tcp-pool-transient",
+      failureKind: error.failureKind,
+      providerId: error.providerId ?? null,
+      dataChannelId: error.dataChannelId ?? null,
+      dataPool: error.dataPool ?? null,
+      diagnostics: { fullHealth: "bun scripts/cli.ts debug health" },
+    };
+  }
+  if (/ssh tcp data channel closed/iu.test(text)) {
+    return { code: "ssh-tcp-pool-transient", failureKind: "provider-data-channel-closed", diagnostics: { fullHealth: "bun scripts/cli.ts debug health" } };
+  }
+  if (/requested ssh tcp data channel is not ready|ssh tcp data channel is not available/iu.test(text)) {
+    return { code: "ssh-tcp-pool-transient", failureKind: "provider-data-channel-missing", diagnostics: { fullHealth: "bun scripts/cli.ts debug health" } };
+  }
+  if (/provider ssh tcp data pool has no idle channel/iu.test(text)) {
+    return { code: "ssh-tcp-pool-transient", failureKind: "provider-data-pool-exhausted", diagnostics: { fullHealth: "bun scripts/cli.ts debug health" } };
+  }
+  return null;
+}
+
+function lastJsonLinePayload(text: string, prefix: string): Record<string, unknown> | null {
+  const lines = text.split(/\r?\n/u);
+  for (let index = lines.length - 1; index >= 0; index -= 1) {
+    const line = lines[index]?.trim() ?? "";
+    if (!line.startsWith(`${prefix} `)) continue;
+    try {
+      const parsed = JSON.parse(line.slice(prefix.length + 1)) as unknown;
+      if (typeof parsed === "object" && parsed !== null && !Array.isArray(parsed)) return parsed as Record<string, unknown>;
+    } catch {
+      return null;
+    }
+  }
+  return null;
+}
+
 function genericJobProgress(job: JobRecord, stderrTailOverride?: string): JobProgressSummary {
   const nowMs = Date.now();
   const stderrTail = stderrTailOverride ?? tailFile(job.stderrFile, 96_000);