various changes

k8s-dvfs/src/main/groovy/recreationaltech/plugin/client/K8sClient.groovy

@@ -21,6 +21,7 @@ import dev.failsafe.FailsafeException
 import dev.failsafe.RetryPolicy
 import dev.failsafe.event.EventListener
 import dev.failsafe.event.ExecutionAttemptedEvent
+import dev.failsafe.function.CheckedPredicate
 import dev.failsafe.function.CheckedSupplier
 import nextflow.exception.K8sOutOfCpuException
 import nextflow.exception.K8sOutOfMemoryException
@@ -371,6 +372,32 @@ class K8sClient {
         (resp?.spec as Map)?.nodeName as String
     }
 
+    /**
+     * Get the external IP address of a node
+     * @param nodeName The node name
+     * @return The external IP address of the node, or internal IP if external is not available
+     */
+    String getNodeExternalIp(String nodeName) {
+        assert nodeName
+        final K8sResponseJson resp = nodeDescribe(nodeName)
+        final status = resp.status as Map
+        final addresses = status?.addresses as List<Map>
+        
+        // Try to find ExternalIP first
+        for (Map address : addresses) {
+            if (address.type == 'ExternalIP') {
+                return address.address as String
+            }
+        }
+        // Fall back to InternalIP
+        for (Map address : addresses) {
+            if (address.type == 'InternalIP') {
+                return address.address as String
+            }
+        }
+        return null
+    }
+
     /**
      * Get pod current state object
      *
@@ -870,10 +897,17 @@ class K8sClient {
             trace('GET', action, resp.text)
             final podList = new K8sResponseJson(resp.text)
             final items = podList.items as List<Map>
-            
+
             long totalAllocated = 0L
             if (items) {
                 for (Map pod : items) {
+                    // Skip terminated/succeeded pods
+                    final status = pod.status as Map
+                    final phase = status?.phase as String
+                    if (phase in ['Succeeded', 'Failed', 'Unknown']) {
+                        continue
+                    }
+
                     final spec = pod.spec as Map
                     if (spec) {
                         final containers = spec.containers as List<Map>
@@ -936,6 +970,13 @@ class K8sClient {
             long totalAllocated = 0L
             if (items) {
                 for (Map pod : items) {
+                    // Skip terminated/succeeded pods
+                    final status = pod.status as Map
+                    final phase = status?.phase as String
+                    if (phase in ['Succeeded', 'Failed', 'Unknown']) {
+                        continue
+                    }
+
                     final spec = pod.spec as Map
                     if (spec) {
                         final containers = spec.containers as List<Map>
@@ -1192,12 +1233,12 @@ class K8sClient {
      * @param cond A predicate that determines when a retry should be triggered
      * @return The {@link dev.failsafe.RetryPolicy} instance
      */
-    protected <T> RetryPolicy<T> retryPolicy(Predicate<? extends Throwable> cond) {
+    protected <T> RetryPolicy<T> retryPolicy(CheckedPredicate<? extends Throwable> cond) {
         final cfg = config.retryConfig
         final listener = new EventListener<ExecutionAttemptedEvent<T>>() {
             @Override
             void accept(ExecutionAttemptedEvent<T> event) throws Throwable {
-                log.debug("K8s response error - attempt: ${event.attemptCount}; reason: ${event.lastFailure.message}")
+                log.debug("K8s response error - attempt: ${event.attemptCount}; reason: ${event.lastException?.message ?: 'unknown'}")
             }
         }
         return RetryPolicy.<T>builder()
@@ -1219,9 +1260,8 @@ class K8sClient {
      */
     protected <T> T apply(CheckedSupplier<T> action) {
         // define the retry condition
-        final cond = new Predicate<? extends Throwable>() {
-            @Override
-            boolean test(Throwable t) {
+        final cond = [
+            test: { Throwable t ->
                 if ( t instanceof K8sResponseException && t.response.code in RETRY_CODES )
                     return true
                 if( t instanceof SocketException || t.cause instanceof SocketException )
@@ -1230,7 +1270,7 @@ class K8sClient {
                     return true
                 return false
             }
-        }
+        ] as CheckedPredicate<Throwable>
         // create the retry policy object
         final policy = retryPolicy(cond)
         // apply the action with and throw the original cause