fix: keep pod log streams alive

server/kubernetes-logs.ts

@@ -40,10 +40,19 @@ function logError(error: unknown): KubernetesLogError {
     error && typeof error === "object" && "code" in error
       ? error.code
       : undefined;
+  const transientNetworkCodes = new Set([
+    "ECONNRESET", "ETIMEDOUT", "EPIPE", "ECONNREFUSED", "EHOSTUNREACH",
+    "ENETUNREACH", "ECONNABORTED", "EAI_AGAIN", "UND_ERR_SOCKET",
+    "UND_ERR_CONNECT_TIMEOUT",
+  ]);
+  const message = error instanceof Error ? error.message : "";
   return new KubernetesLogError(
     error instanceof Error ? error.message : "Kubernetes log request failed",
-    code === undefined &&
-      (status === undefined || status === 408 || status === 429 || status >= 500),
+    typeof code === "string"
+      ? transientNetworkCodes.has(code)
+      : status === undefined && /socket hang up|socket closed/i.test(message)
+        ? true
+        : status === undefined || status === 408 || status === 429 || status >= 500,
   );
 }
 

server/log-service.ts

@@ -438,7 +438,7 @@ export class LogService {
     const discoveryInterval = positiveInteger(input.discoveryIntervalMs, 2_000);
     const heartbeatInterval = positiveInteger(
       input.heartbeatIntervalMs,
-      15_000,
+      5_000,
     );
     const retryInterval = positiveInteger(input.retryIntervalMs, 1_000);
     const queue = new BoundedAsyncQueue<LogEvent>(capacity);
@@ -489,7 +489,11 @@ export class LogService {
             signal,
             (event) => queue.push(event),
           );
-          if (!signal.aborted) completed.add(key);
+          if (!signal.aborted)
+            throw new KubernetesLogError(
+              "Kubernetes log stream ended unexpectedly",
+              true,
+            );
         } catch (error) {
           if (!isAbort(signal)) {
             await pushError(error, source);