Major I set up a Jenkins master on Azure container service (Kubernetes). I got a lot of warning since the Jenkins master was set up and never end:
Sep 15, 2017 8:33:23 AM hudson.TcpSlaveAgentListener$ConnectionHandler run WARNING: Connection #600 failed java.io.EOFException at java.io.DataInputStream.readFully(DataInputStream.java:197) at java.io.DataInputStream.readFully(DataInputStream.java:169) at hudson.TcpSlaveAgentListener$ConnectionHandler.run(TcpSlaveAgentListener.java:213) Sep 15, 2017 8:33:23 AM hudson.TcpSlaveAgentListener$ConnectionHandler run WARNING: Connection #602 failed java.io.EOFException at java.io.DataInputStream.readFully(DataInputStream.java:197) at java.io.DataInputStream.readFully(DataInputStream.java:169) at hudson.TcpSlaveAgentListener$ConnectionHandler.run(TcpSlaveAgentListener.java:213) Sep 15, 2017 8:33:27 AM hudson.TcpSlaveAgentListener$ConnectionHandler run WARNING: Connection #603 failed java.io.EOFException at java.io.DataInputStream.readFully(DataInputStream.java:197) at java.io.DataInputStream.readFully(DataInputStream.java:169) at hudson.TcpSlaveAgentListener$ConnectionHandler.run(TcpSlaveAgentListener.java:213)
Although there is a lot of warning, but everything is OK. I can even use Azure VM plugin to set up jnlp slaves.
I think this issue related to port: 50000 because if I expose port 50000 as a cluster port, there is no error. However, if I set port 50000 as LoadBalancer (we have to do this), the errors shows above.
Here is my kube files:
kind: PersistentVolumeClaim
apiVersion: v1
metadata:
name: azdisk
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 5Gi
storageClassName: azuredisk
---
kind: Deployment
apiVersion: apps/v1beta1
metadata:
name: jenkins-1
spec:
replicas: 1
template:
metadata:
name: jenkins-1
labels:
app: jenkins-1
spec:
containers:
- name: jenkins-container
image: zackliu1995/jenkins
volumeMounts:
- name: azure
mountPath: /var/jenkins_home
securityContext:
privileged: true
ports:
- name: port8080
containerPort: 8080
protocol: TCP
- name: port50000
containerPort: 50000
protocol: TCP
- name: port22
containerPort: 22
volumes:
- name: azure
persistentVolumeClaim:
claimName: azdisk
---
apiVersion: v1
kind: Service
metadata:
name: jenkins-srv
spec:
selector:
app: jenkins-1
ports:
- name: http
port: 80
protocol: TCP
targetPort: 8080
- name: slave
port: 50000
protocol: TCP
targetPort: 50000
- name: ssh
port: 22
targetPort: 22
type: LoadBalancer
I tried official Jenkins image, it caused the same issue.
[JENKINS-46893] Huge amount of TcpSlaveAgentListener EOFException on Kubernetes
Chenyang Liu
created issue -
| Component/s | New: remoting [ 15489 ] |
Chenyang Liu
added a comment - Because we want to make Jenkins as service for different users.
As we know, LoadBalancer will apply for a new public IP in Azure automatically and our user can use this new IP to visit his Jenkins.
Do you know the reason why there are so much warnings when the Jenkins start to run. But in fact there is no issue to connect slave.
Chenyang Liu
added a comment - Because we want to make Jenkins as service for different users.
As we know, LoadBalancer will apply for a new public IP in Azure automatically and our user can use this new IP to visit his Jenkins.
Do you know the reason why there are so much warnings when the Jenkins start to run. But in fact there is no issue to connect slave. I would guess there is a broken agent. Or maybe VMWare plugin forcefully disconnects the agent. It usually happen with onRecvClosed() warning, but this one may also happen if the data transfer is in progress. In order to say more, I would need agent logs at least
Chenyang Liu
added a comment - Oh, you probably misunderstood the issue. There is no agent but only a Jenkins master. This issue happened since the very beginning that I haven't even finished my wizard (haven't unlocked the Jenkins and haven't installed the suggest plugins) and of cause haven't installed VM agent plugin.
So, I think maybe it caused by Jenkins's self check... I don't know, so strange.
Chenyang Liu
added a comment - Oh, you probably misunderstood the issue. There is no agent but only a Jenkins master. This issue happened since the very beginning that I haven't even finished my wizard (haven't unlocked the Jenkins and haven't installed the suggest plugins) and of cause haven't installed VM agent plugin.
So, I think maybe it caused by Jenkins's self check... I don't know, so strange. I can add more diagnostics, but there is no such self-check in Jenkins for sure.
The errors may be also coming from Remoting-based CLI, non-terminated Jenkins Maven project runs on instances. There are also other components from vendors which may be trying to connect to the master via Remoting.
Does a load balancer connect to the port in question as a health check of sorts?
Perhaps rtyler or olblak can take a look at this as our resident Kubernetes on Azure experts.
Chenyang Liu
added a comment - That's probably the case, Azure Load Balancer does have health probes for every 5 second by default.
I will check it later on Monday.
Chenyang Liu
added a comment - That's probably the case, Azure Load Balancer does have health probes for every 5 second by default.
I will check it later on Monday.
Why do you have to set up a port as a load balancer?
If you want to setup high availability or load balancing for Jenkins, the recommended approach is to expose ports of underlying Jenkins master instances and let agents connect to them directly using the "hudson.TcpSlaveAgentListener.hostName" system property (it will be returned to agents in HttpResponse headers)