fix: keep pod log streams alive

changelogs/2.5.2.md

@@ -0,0 +1,10 @@
+# 2.5.2-rc.1
+
+## Fixed
+
+- Keep server-side Kubernetes log streams active with a 5-second idle heartbeat, below common intermediary idle timeouts.
+- Retry transient Kubernetes log socket and stream closures; permanent TLS certificate errors remain terminal.
+
+## Changed
+
+- Prefix CLI log lines with the originating pod name instead of the target service name.

command/logs.ts

@@ -86,7 +86,7 @@ export async function runLogs(
       write(service ?? event.pod ?? project, event.message, true);
       continue;
     }
-    write(event.targetName || service || project, event.message, false);
+    write(event.pod || event.targetName || service || project, event.message, false);
   }
 }
 

command/main.ts

@@ -37,7 +37,7 @@ function cloneCommand<T extends object>(command: T): T {
 export const main = defineCommand({
   meta: {
     name: "kuber",
-    version: "2.5.1",
+    version: "2.5.2-rc.1",
     description: "Docker Compose -> K8s translation layer",
   },
   args: {

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dmgnr/kuber",
-  "version": "2.5.1",
+  "version": "2.5.2-rc.1",
   "description": "Docker Compose to Kubernetes translation layer",
   "bin": {
     "kuber": "dist/index.js"

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 &&
+    typeof code === "string"
-      (status === undefined || status === 408 || status === 429 || status >= 500),
+      ? 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);

tests/command/logs.test.ts

@@ -58,7 +58,7 @@ describe("logs API command runner", () => {
       },
     ]);
     expect(output).toEqual([
-      ["web", "ready", false],
+      ["web-1", "ready", false],
       ["web api", "disconnected", true],
     ]);
   });

tests/server/log-service.test.ts

@@ -650,7 +650,7 @@ describe("LogService", () => {
       expect(error).toMatchObject({
         pod: "web-abc",
         container: "web",
-        retryable: false,
+        retryable: true,
       });
       expect(error?.message).toMatch(
         /socket.*connection.*closed|socket hang up|fetch failed/i,