ais: improve logging around initalization

Signed-off-by: Janusz Marcinkiewicz <januszm@nvidia.com>
NVIDIA · Nov 28, 2024 · 25940e6 · 25940e6
1 parent b5d471b
commit 25940e6
Show file tree

Hide file tree

Showing 3 changed files with 29 additions and 31 deletions.
diff --git a/ais/proxy.go b/ais/proxy.go
@@ -126,23 +126,22 @@ func initPID(config *cmn.Config) (pid string) {
 	// 1. ID from env
 	if pid = envDaemonID(apc.Proxy); pid != "" {
 		if err := cos.ValidateDaemonID(pid); err != nil {
-			nlog.Errorf("Warning: %v", err)
+			nlog.Errorln("Daemon ID loaded from env is invalid:", err)
 		}
+		nlog.Infof("Initialized %s from env", meta.Pname(pid))
 		return
 	}
 
-	// 2. proxy, K8s
+	// 2. ID from Kubernetes Node name.
 	if k8s.IsK8s() {
-		// NOTE: always generate i.e., compute
-		if net.ParseIP(k8s.NodeName) != nil { // does not parse as IP
-			nlog.Warningf("using K8s node name %q, an IP addr, to compute _persistent_ proxy ID", k8s.NodeName)
-		}
-		return cos.HashK8sProxyID(k8s.NodeName)
+		pid = cos.HashK8sProxyID(k8s.NodeName)
+		nlog.Infof("Initialized %s from K8s node name %q", meta.Pname(pid), k8s.NodeName)
+		return pid
 	}
 
-	// 3. try to read ID
+	// 3. Read ID from persistent file.
 	if pid = readProxyID(config); pid != "" {
-		nlog.Infof("p[%s] from %q", pid, fname.ProxyID)
+		nlog.Infof("Initialized %s from file %q", meta.Pname(pid), fname.ProxyID)
 		return
 	}
 
@@ -154,7 +153,7 @@ func initPID(config *cmn.Config) (pid string) {
 	// store ID on disk
 	err = os.WriteFile(filepath.Join(config.ConfigDir, fname.ProxyID), []byte(pid), cos.PermRWR)
 	debug.AssertNoErr(err)
-	nlog.Infof("p[%s] ID randomly generated", pid)
+	nlog.Infof("Initialized %s with randomly generated ID", meta.Pname(pid))
 	return
 }
 

diff --git a/ais/target.go b/ais/target.go
@@ -278,24 +278,24 @@ func (t *target) initHostIP(config *cmn.Config) {
 func initTID(config *cmn.Config) (tid string, generated bool) {
 	if tid = envDaemonID(apc.Target); tid != "" {
 		if err := cos.ValidateDaemonID(tid); err != nil {
-			nlog.Errorln("Warning:", err)
+			nlog.Errorln("Daemon ID loaded from env is invalid:", err)
 		}
+		nlog.Infof("Initialized %s from env", meta.Tname(tid))
 		return tid, false
 	}
 
-	var err error
-	if tid, err = fs.LoadNodeID(config.FSP.Paths); err != nil {
+	if tid, err := fs.LoadNodeID(config.FSP.Paths); err != nil {
 		cos.ExitLog(err) // FATAL
-	}
-	if tid != "" {
+	} else if tid != "" {
+		nlog.Infof("Initialized %s from file", meta.Tname(tid))
 		return tid, false
 	}
 
 	// this target: generate random ID
 	tid = genDaemonID(apc.Target, config)
-	err = cos.ValidateDaemonID(tid)
+	err := cos.ValidateDaemonID(tid)
 	debug.AssertNoErr(err)
-	nlog.Infoln(meta.Tname(tid) + ": ID randomly generated")
+	nlog.Infof("Initialized %s with randomly generated ID", meta.Tname(tid))
 	return tid, true
 }
 

diff --git a/cmn/k8s/k8s.go b/cmn/k8s/k8s.go
@@ -10,6 +10,7 @@ import (
 	"strings"
 
 	"github.com/NVIDIA/aistore/api/env"
+	"github.com/NVIDIA/aistore/cmn/cos"
 	"github.com/NVIDIA/aistore/cmn/debug"
 	"github.com/NVIDIA/aistore/cmn/nlog"
 	v1 "k8s.io/api/core/v1"
@@ -44,8 +45,8 @@ func Init() {
 
 	var (
 		pod      *v1.Pod
-		nodeName = os.Getenv(env.AIS.K8sNode)
 		podName  = os.Getenv(env.AIS.K8sPod)
+		nodeName = os.Getenv(env.AIS.K8sNode)
 	)
 	if podName != "" {
 		debug.Func(func() {
@@ -57,36 +58,34 @@ func Init() {
 	}
 	nlog.Infof("Checking pod: %q, node: %q", podName, nodeName)
 
-	// node name specified - proceed directly to check
-	if nodeName != "" {
-		goto checkNode
-	}
 	if podName == "" {
-		nlog.Infoln("environment (above) not set =>", nonK8s)
+		if nodeName != "" {
+			// If the Pod is not set but the Node is, we should continue checking.
+			goto checkNode
+		}
+		nlog.Infof("Env %q and %q are not set => %s", env.AIS.K8sNode, env.AIS.K8sPod, nonK8s)
 		return
 	}
 
-	// check POD
+	// Check Pod.
 	pod, err = client.Pod(podName)
 	if err != nil {
-		nlog.Errorf("Failed to get pod %q: %v", podName, err)
+		cos.ExitLogf("Failed to get Pod %q, err: %v", podName, err)
 		return
 	}
 	nodeName = pod.Spec.NodeName
-	nlog.Infoln("pod.Spec: Node", nodeName, "Hostname", pod.Spec.Hostname, "HostNetwork", pod.Spec.HostNetwork)
+	nlog.Infoln("Pod spec", "name", podName, "namespace", pod.Namespace, "node", nodeName, "hostname", pod.Spec.Hostname, "host_network", pod.Spec.HostNetwork)
 	_ppvols(pod.Spec.Volumes)
 
-checkNode: // always check Node
+checkNode:
+	// Check Node.
 	node, err := client.Node(nodeName)
 	if err != nil {
-		nlog.Errorf("Failed to get Node %q: %v", nodeName, err)
+		cos.ExitLogf("Failed to get Node %q, err: %v", nodeName, err)
 		return
 	}
 
 	NodeName = node.Name
-	if node.Namespace != "" {
-		nlog.Infoln("Node", NodeName, "Namespace", node.Namespace)
-	}
 }
 
 func _ppvols(volumes []v1.Volume) {