I successfully setup a mini Ray cluster (1 head + 1 worker, each with 4 CPU cores) manually. However, I failed to set up it automatically using the Apache Ray autoscaler. The head node starts correctly while the worker node never joins the cluster. Below is my YAML configuration for the autoscaler. Is there anything I did wrong?
cluster_name: my_ray_cluster
min_workers: 8
initial_workers: 8
max_workers: 8
provider:
type: local
head_ip: 10.148.186.178
worker_ips: [10.148.186.18]
auth:
ssh_user: USER_NAME
ssh_private_key: ~/.ssh/id_rsa
# Files or directories to copy to the head and worker nodes.
file_mounts: {
# "/path1/on/remote/machine": "/path1/on/local/machine",
# "/path2/on/remote/machine": "/path2/on/local/machine",
}
head_setup_commands:
- pip3 install ray[debug,dashboard]
setup_commands:
- pip3 install ray[debug,dashboard]
# Command to start ray on the head node. You don't need to change this.
head_start_ray_commands:
- ray stop
- ray start --head --redis-port=6379
worker_start_ray_commands:
- ray stop
- ray start --address=10.148.186.178:6379
Related
I am trying to setup a K3S cluster for learning purposes but I am having trouble connecting the master node with agents. I have looked several tutorials and discussions on this but I can't find a solution. I know I am probably missing something obvious (due to my lack of knowledge), but still help would be much appreciated.
I am using two AWS t2.micro instances with default configuration.
When ssh into the master and installed K3S using
curl -sfL https://get.k3s.io | sh -s - --no-deploy traefik --write-kubeconfig-mode 644 --node-name k3s-master-01
with kubectl get nodes, I am able to see the master
NAME STATUS ROLES AGE VERSION
k3s-master-01 Ready control-plane,master 13s v1.23.6+k3s1
So far it seems I am doing things right. From what I understand, I am supposed to configure the kubeconfig file. So, I accessed it by using
cat /etc/rancher/k3s/k3s.yaml
I copied the configuration file and the server info to match the private IP I took from AWS console, resulting in something like this
apiVersion: v1
clusters:
- cluster:
certificate-authority-data: <lots_of_info>
server: https://<master_private_IP>:6443
name: default
contexts:
- context:
cluster: default
user: default
name: default
current-context: default
kind: Config
preferences: {}
users:
- name: default
user:
client-certificate-data: <my_certificate_data>
client-key-data: <my_key_data>
Then, I ran vi ~/.kube/config, and there I pasted the kubeconfig file
Finally, I grabbed the token with cat /var/lib/rancher/k3s/server/node-token, ssh into the other machine and then run the following
curl -sfL https://get.k3s.io | K3S_NODE_NAME=k3s-worker-01 K3S_URL=https://<master_private_IP>:6443 K3S_TOKEN=<master_token> sh -
The output is
[INFO] Finding release for channel stable
[INFO] Using v1.23.6+k3s1 as release
[INFO] Downloading hash https://github.com/k3s-io/k3s/releases/download/v1.23.6+k3s1/sha256sum-amd64.txt
[INFO] Downloading binary https://github.com/k3s-io/k3s/releases/download/v1.23.6+k3s1/k3s
[INFO] Verifying binary download
[INFO] Installing k3s to /usr/local/bin/k3s
[INFO] Skipping installation of SELinux RPM
[INFO] Creating /usr/local/bin/kubectl symlink to k3s
[INFO] Creating /usr/local/bin/crictl symlink to k3s
[INFO] Creating /usr/local/bin/ctr symlink to k3s
[INFO] Creating killall script /usr/local/bin/k3s-killall.sh
[INFO] Creating uninstall script /usr/local/bin/k3s-agent-uninstall.sh
[INFO] env: Creating environment file /etc/systemd/system/k3s-agent.service.env
[INFO] systemd: Creating service file /etc/systemd/system/k3s-agent.service
[INFO] systemd: Enabling k3s-agent unit
Created symlink /etc/systemd/system/multi-user.target.wants/k3s-agent.service → /etc/systemd/system/k3s-agent.service.
[INFO] systemd: Starting k3s-agent
By this output, it looks like I have created an agent. However, when I run kubectl get nodes in the master, I still get
NAME STATUS ROLES AGE VERSION
k3s-master-01 Ready control-plane,master 12m v1.23.6+k3s1
What is the thing I was supposed to do in order to get the agent connected to the master? I am guess I am probably missing something simple, but I just can't seem to find the solution. I've read all the documentation but it is still not clear to me where I am making the mistake. I've tried saving the private master IP and token into the agent as environmental variables with export K3S_TOKEN=master_token and K3S_URL=master_private_IP and then simply running curl -sfL https://get.k3s.io | sh - but I still can't see the worker nodes when running kubectl get nodes
Any help would be appreciated.
It might be your VM instance firewall that prevents appropriate connection from your master to the worker node (and vice versa). Official rancher documentation advise to disable firewall for (Red Hat/CentOS) Enterprise Linux:
It is recommended to turn off firewalld:
systemctl disable firewalld --now
If enabled, it is required to disable nm-cloud-setup and reboot the node:
systemctl disable nm-cloud-setup.service nm-cloud-setup.timer reboot
If you are using Ubuntu on your VM's, there is a different firewall tool (ufw).
In my case, allowing 6443 and 443(not sure if required) port TCP connections worked fine.
Allow port 6443 and TCP connection in all of your cluster machines:
sudo ufw allow 6443/tcp
Then apply k3s installation script in your worker node(s):
curl -sfL https://get.k3s.io | K3S_NODE_NAME=k3s-worker-1 K3S_URL=https://<k3s-master-1 IP>:6443 K3S_TOKEN=<k3s-master-1 TOKEN> sh -
This should work. If not, you can try adding additional allow rule for 443 tcp port as well.
A few options to check.
Check Journalctl for errors
journalctl -u k3s-agent.service -n 300 -xn
If using RaspberryPi for a worker node, make sure you have
cgroup_enable=cpuset cgroup_enable=memory cgroup_memory=1
as the very end of your /boot/cmdline.txt file. DO NOT PUT THIS VALUE ON A NEW LINE! Should just be appended to the end of the line.
If your master node(s) have self-signed certs, make sure you copy the master node's self signed cert to your worker node(s). In linux or raspberry pi copy cert to /usr/local/share/ca-certificates, then issue an
sudo update-ca-certificates
on the worker node
Don't forget to reboot the worker node after you make these changes!
Hope this helps someone!
I am running slurm 20.11.8 on a system with 1 login node and 3 compute nodes. The status information I can find is below.
$ slurmd -V
slurm 20.11.8
$ sinfo -N
NODELIST NODES PARTITION STATE
pauli-node-01 1 normal* idle
pauli-node-02 1 normal* idle
pauli-node-03 1 normal* idle
$ systemctl status slurmctld
● slurmctld.service - Slurm controller daemon
Loaded: loaded (/usr/lib/systemd/system/slurmctld.service; enabled; vendor preset: disabled)
Active: active (running) since Tue 2021-10-05 22:04:10 CDT; 10h ago
Main PID: 11802 (slurmctld)
Tasks: 7
Memory: 6.7M
CGroup: /system.slice/slurmctld.service
└─11802 /usr/sbin/slurmctld -D
Oct 05 22:04:10 pauli.mer.utexas.edu systemd[1]: Started Slurm controller daemon.
Here is my configuration file:
$ cat /etc/slurm/slurm.conf
# slurm.conf file generated by configurator.html.
# Put this file on all nodes of your cluster.
# See the slurm.conf man page for more information.
#
SlurmctldHost=pauli
#SlurmctldHost=
#
#DisableRootJobs=NO
#EnforcePartLimits=NO
#Epilog=
#EpilogSlurmctld=
#FirstJobId=1
#MaxJobId=999999
#GresTypes=
#GroupUpdateForce=0
#GroupUpdateTime=600
#JobFileAppend=0
#JobRequeue=1
#JobSubmitPlugins=1
#KillOnBadExit=0
#LaunchType=launch/slurm
#Licenses=foo*4,bar
#MailProg=/bin/mail
#MaxJobCount=5000
#MaxStepCount=40000
#MaxTasksPerNode=128
MpiDefault=none
#MpiParams=ports=#-#
#PluginDir=
#PlugStackConfig=
#PrivateData=jobs
ProctrackType=proctrack/cgroup
#Prolog=
#PrologFlags=
#PrologSlurmctld=
#PropagatePrioProcess=0
#PropagateResourceLimits=
#PropagateResourceLimitsExcept=
#RebootProgram=
ReturnToService=1
SlurmctldPidFile=/var/run/slurmctld.pid
SlurmctldPort=6817
SlurmdPidFile=/var/run/slurmd.pid
SlurmdPort=6818
SlurmdSpoolDir=/var/spool/slurmd
SlurmUser=slurm
#SlurmdUser=root
#SrunEpilog=
#SrunProlog=
StateSaveLocation=/var/spool/slurmctld
SwitchType=switch/none
#TaskEpilog=
TaskPlugin=task/affinity
#TaskProlog=
#TopologyPlugin=topology/tree
#TmpFS=/tmp
#TrackWCKey=no
#TreeWidth=
#UnkillableStepProgram=
#UsePAM=0
#
#
# TIMERS
#BatchStartTimeout=10
#CompleteWait=0
#EpilogMsgTime=2000
#GetEnvTimeout=2
#HealthCheckInterval=0
#HealthCheckProgram=
InactiveLimit=0
KillWait=30
#MessageTimeout=10
#ResvOverRun=0
MinJobAge=300
#OverTimeLimit=0
SlurmctldTimeout=120
SlurmdTimeout=300
#UnkillableStepTimeout=60
#VSizeFactor=0
Waittime=0
#
#
# SCHEDULING
#DefMemPerCPU=0
#MaxMemPerCPU=0
#SchedulerTimeSlice=30
SchedulerType=sched/backfill
SelectType=select/cons_tres
SelectTypeParameters=CR_CPU
#
#
# JOB PRIORITY
#PriorityFlags=
#PriorityType=priority/basic
#PriorityDecayHalfLife=
#PriorityCalcPeriod=
#PriorityFavorSmall=
#PriorityMaxAge=
#PriorityUsageResetPeriod=
#PriorityWeightAge=
#PriorityWeightFairshare=
#PriorityWeightJobSize=
#PriorityWeightPartition=
#PriorityWeightQOS=
#
#
# LOGGING AND ACCOUNTING
#AccountingStorageEnforce=0
AccountingStorageHost=pauli
#AccountingStoragePass=abcdef
#AccountingStoragePort=1234
AccountingStorageType=accounting_storage/none
AccountingStorageUser=slurm
AccountingStoreJobComment=YES
ClusterName=cluster
#DebugFlags=
JobCompHost=pauli
#JobCompLoc=slurm_comp_db
#JobCompPass=abcdef
#JobCompPort=1234
JobCompType=jobcomp/none
JobCompUser=slurm
#JobContainerType=job_container/none
JobAcctGatherFrequency=30
JobAcctGatherType=jobacct_gather/none
SlurmctldDebug=info
#SlurmctldLogFile=
SlurmdDebug=info
#SlurmdLogFile=
#SlurmSchedLogFile=
#SlurmSchedLogLevel=
#
#
# POWER SAVE SUPPORT FOR IDLE NODES (optional)
#SuspendProgram=
#ResumeProgram=
#SuspendTimeout=
#ResumeTimeout=
#ResumeRate=
#SuspendExcNodes=
#SuspendExcParts=
#SuspendRate=
#SuspendTime=
#
#
# COMPUTE NODES
NodeName=pauli-node-0[1-2] CPUs=64 Sockets=2 CoresPerSocket=16 ThreadsPerCore=2 State=UNKNOWN
NodeName=pauli-node-03 CPUs=40 Sockets=2 CoresPerSocket=10 ThreadsPerCore=2 State=UNKNOWN
PartitionName=normal Nodes=pauli-node-0[1-3] Default=YES MaxTime=INFINITE State=UP
When I try to run a simple job on 1 node with srun (along the lines of Slurm: Quick Start User Guide), the job just hangs. Does anyone know where I should look for diagnostic information to figure out why the job hangs?
$ srun -N1 -n1 -l hostname
One of the first things to check is the network connectivity and making sure no firewall is in the way. You can check that with
scontrol ping
on the control nodes. Also, srun has a -v option that can tell you where it is blocked (you can add multiple of such options to increase the verbosity).
And of course, the log files for both the controller and the slurmd's may contain information. Again, the log level can be increased, with scontrol setdebug.
The usual suspects, besides the firewall, are often SELinux, netmasks, IP routes. Make also sure the clocks are in sync and munge is running OK.
SOLVED. The firewall for all compute nodes must be either "off" or configured to trust the other nodes in the system.
See Compute node firewall must be off
I was able to run (on Linux RedHat)
firewall-cmd --zone=trusted --add-source=10.xxx.xxx.xxx --add-source=10.xxx.xxx.xxx --add-source=10.xxx.xxx.xxx
on each compute node in order to avoid turning off the firewall altogether. I think the reason the problem came up recently is that the firewall was probably deactivated, and after a recent reboot of the system, the firewall came back up.
Thanks #damienfrancois for helping point me to firewall problems.
Observed behavior
I started with one node Openshift cluster and it successfully deployed master/node and gluster volume. Now I extend Openshift cluster and it was successfully.
but on extending glusterfs volume with below
[glusterfs]
10.1.1.1 glusterfs_devices='[ "/dev/vdb" ]'
10.1.1.2 glusterfs_devices='[ "/dev/vdb" ]' openshift_node_labels="type=upgrade"
ansible-playbook -i inventory2.ini /usr/share/ansible/openshift-ansible/playbooks/openshift-glusterfs/config.yml -e openshift_upgrade_nodes_label="type=upgrade"
it only added 10.1.1.2 as peer but volume still has only one brick
Following customization done to start deploy gluster from 1 node {--durability none}
openshift-ansible/roles/openshift_storage_glusterfs/tasks/heketi_init_db.yml
- name: Create heketi DB volume
command: "{{ glusterfs_heketi_client }} setup-openshift-heketi-storage --image {{ glusterfs_heketi_image }} --listfile /tmp/heketi-storage.json **--durability none**"
register: setup_storage
>gluster peer status
Number of Peers: 1
Hostname: 10.1.1.2
Uuid: 1b8159e4-99e2-4f4d-ad95-e97bc8655d32
State: Peer in Cluster (Connected)
gluster volume info
Volume Name: heketidbstorage
Type: Distribute
Volume ID: 769419b9-d28f-4cdd-a8f3-708b6b738f65
Status: Started
Snapshot Count: 0
Number of Bricks: 1
Transport-type: tcp
Bricks:
Brick1: 10.1.1.1:/var/lib/heketi/mounts/vg_4187bfa3eb090ceffea9c53b156ddbd4/brick_80401b43be8c3c8a74417b18ad574524/brick
Options Reconfigured:
transport.address-family: inet
nfs.disable: on
Expected/desired behavior
I am expecting that on addition of every new node it should create new brick too
Details on how to reproduce (minimal and precise)
Add nodes in gluster cluster with below commands
ansible-playbook -i inventory2.ini /usr/share/ansible/openshift-ansible/playbooks/openshift-glusterfs/config.yml -e openshift_upgrade_nodes_label="type=upgrade"
Information about the environment:
Heketi version used (e.g. v6.0.0 or master): OpenShift 3.10
Operating system used: CentOS
Heketi compiled from sources, as a package (rpm/deb), or container: Container
If container, which container image: docker.io/heketi/heketi:latest
Using kubernetes, openshift, or direct install: Openshift
If kubernetes/openshift, is gluster running inside kubernetes/openshift or outside: outside
If kubernetes/openshift, how was it deployed (gk-deploy, openshift-ansible, other, custom): openshift-ansible
Just adding a node/server does not mean that the brick will also be added to existing
gluster volume.
You have to add that brick, hosted on new node, to existing volume.
command -
"gluster volume add-brick host:brick-path commit force"
Not sure if you have provided this command in your automation script or not.
I have setup a K8S cluster (1 master and 2 slaves) using Kubeadm on my laptop.
Deployed 6 replicas of a pod. 3 of them got deployed to each of the slaves.
Did a shutdown of one of the slave.
It took ~6 minutes for the 3 pods to be scheduled on the running node.
Initially, I thought that it had to do something with the K8S setup. After some digging found out, it's because of the defaults in the K8S for Controller Manager and Kubelet as mentioned here. It made sense. I checked out the K8S documentation on where to change the configuration properties and also checked the configuration files on the cluster node, but couldn't figure it out.
kubelet: node-status-update-frequency=4s (from 10s)
controller-manager: node-monitor-period=2s (from 5s)
controller-manager: node-monitor-grace-period=16s (from 40s)
controller-manager: pod-eviction-timeout=30s (from 5m)
Could someone point out what needs to be done to make the above-mentioned configuration changes permanent and also the different options for the same?
On the kubelet change this file on all your nodes:
/var/lib/kubelet/kubeadm-flags.env
Add the option at the end or anywhere on this line:
KUBELET_KUBEADM_ARGS=--cgroup-driver=cgroupfs --cni-bin-dir=/opt/cni/bin
--cni-conf-dir=/etc/cni/net.d --network-plugin=cni
--resolv-conf=/run/systemd/resolve/resolv.conf
--node-status-update-frequency=10s <== add this
On your kube-controller-manager change on the master the following file:
/etc/kubernetes/manifests/kube-controller-manager.yaml
In this section:
containers:
- command:
- kube-controller-manager
- --address=127.0.0.1
- --allocate-node-cidrs=true
- --cloud-provider=aws
- --cluster-cidr=192.168.0.0/16
- --cluster-signing-cert-file=/etc/kubernetes/pki/ca.crt
- --cluster-signing-key-file=/etc/kubernetes/pki/ca.key
- --controllers=*,bootstrapsigner,tokencleaner
- --kubeconfig=/etc/kubernetes/controller-manager.conf
- --leader-elect=true
- --node-cidr-mask-size=24
- --root-ca-file=/etc/kubernetes/pki/ca.crt
- --service-account-private-key-file=/etc/kubernetes/pki/sa.key
- --use-service-account-credentials=true
- –-node-monitor-period=5s <== add this line
On your master do a sudo systemctl restart docker
On all your nodes do a sudo systemctl restart kubelet
You should have the new configs take effect.
Hope it helps.
I'm using "oc cluster up" to start my Openshift Origin environment. I can see, however, that once I shutdown the cluster my projects aren't persisted at restart. Is there a way to make them persistent ?
Thanks
There are a couple ways to do this. oc cluster up doesn't have a primary use case of persisting resources.
There are couple ways to do it:
Leverage capturing etcd as described in the oc cluster up README
There is a wrapper tool, that makes it easy to do this.
There is now an example in the cluster up --help command, it is bound to stay up to date so check that first
oc cluster up --help
...
Examples:
# Start OpenShift on a new docker machine named 'openshift'
oc cluster up --create-machine
# Start OpenShift using a specific public host name
oc cluster up --public-hostname=my.address.example.com
# Start OpenShift and preserve data and config between restarts
oc cluster up --host-data-dir=/mydata --use-existing-config
So specifically in v1.3.2 use --host-data-dir and --use-existing-config
Assuming you are using docker machine with vm such as virtual box, the easiest way I found is taking a vm snapshot WHILE vm and openshift cluster are up and running. This snapshot will backup memory in addition to disk therefore you can restore entire cluster later on by restoring the vm snapshot, then run docker-machine start ...
btw, as of latest os image openshift/origin:v3.6.0-rc.0 and oc cli, --host-data-dir=/mydata as suggested in the other answer doesn't work for me.
I'm using:
VirtualBox 5.1.26
Kubernetes v1.5.2+43a9be4
openshift v1.5.0+031cbe4
Didn't work for me using --host-data-dir (and others) :
oc cluster up --logging=true --metrics=true --docker-machine=openshift --use-existing-config=true --host-data-dir=/vm/data --host-config-dir=/vm/config --host-pv-dir=/vm/pv --host-volumes-dir=/vm/volumes
With output:
-- Checking OpenShift client ... OK
-- Checking Docker client ...
Starting Docker machine 'openshift'
Started Docker machine 'openshift'
-- Checking Docker version ...
WARNING: Cannot verify Docker version
-- Checking for existing OpenShift container ... OK
-- Checking for openshift/origin:v1.5.0 image ... OK
-- Checking Docker daemon configuration ... OK
-- Checking for available ports ... OK
-- Checking type of volume mount ...
Using Docker shared volumes for OpenShift volumes
-- Creating host directories ... OK
-- Finding server IP ...
Using docker-machine IP 192.168.99.100 as the host IP
Using 192.168.99.100 as the server IP
-- Starting OpenShift container ...
Starting OpenShift using container 'origin'
FAIL
Error: could not start OpenShift container "origin"
Details:
Last 10 lines of "origin" container log:
github.com/openshift/origin/vendor/github.com/coreos/pkg/capnslog.(*PackageLogger).Panicf(0xc4202a1600, 0x42b94c0, 0x1f, 0xc4214d9f08, 0x2, 0x2)
/go/src/github.com/openshift/origin/_output/local/go/src/github.com/openshift/origin/vendor/github.com/coreos/pkg/capnslog/pkg_logger.go:75 +0x16a
github.com/openshift/origin/vendor/github.com/coreos/etcd/mvcc/backend.newBackend(0xc4209f84c0, 0x33, 0x5f5e100, 0x2710, 0xc4214d9fa8)
/go/src/github.com/openshift/origin/_output/local/go/src/github.com/openshift/origin/vendor/github.com/coreos/etcd/mvcc/backend/backend.go:106 +0x341
github.com/openshift/origin/vendor/github.com/coreos/etcd/mvcc/backend.NewDefaultBackend(0xc4209f84c0, 0x33, 0x461e51, 0xc421471200)
/go/src/github.com/openshift/origin/_output/local/go/src/github.com/openshift/origin/vendor/github.com/coreos/etcd/mvcc/backend/backend.go:100 +0x4d
github.com/openshift/origin/vendor/github.com/coreos/etcd/etcdserver.NewServer.func1(0xc4204bf640, 0xc4209f84c0, 0x33, 0xc421079a40)
/go/src/github.com/openshift/origin/_output/local/go/src/github.com/openshift/origin/vendor/github.com/coreos/etcd/etcdserver/server.go:272 +0x39
created by github.com/openshift/origin/vendor/github.com/coreos/etcd/etcdserver.NewServer
/go/src/github.com/openshift/origin/_output/local/go/src/github.com/openshift/origin/vendor/github.com/coreos/etcd/etcdserver/server.go:274 +0x345
Openshift writes to the directories /vm/... (also defined in VirtualBox) but successfully won't start.
See [https://github.com/openshift/origin/issues/12602][1]
Worked for me too, using Virtual Box Snapshots and restoring them.
To make it persistent after each shutdown you need to provide base-dir parameter.
$ mkdir ~/openshift-config
$ oc cluster up --base-dir=~/openshift-config
From help
$ oc cluster up --help
...
Options:
--base-dir='': Directory on Docker host for cluster up configuration
--enable=[*]: A list of components to enable. '*' enables all on-by-default components, 'foo' enables the component named 'foo', '-foo' disables the component named 'foo'.
--forward-ports=false: Use Docker port-forwarding to communicate with origin container. Requires 'socat' locally.
--http-proxy='': HTTP proxy to use for master and builds
--https-proxy='': HTTPS proxy to use for master and builds
--image='openshift/origin-${component}:${version}': Specify the images to use for OpenShift
--no-proxy=[]: List of hosts or subnets for which a proxy should not be used
--public-hostname='': Public hostname for OpenShift cluster
--routing-suffix='': Default suffix for server routes
--server-loglevel=0: Log level for OpenShift server
--skip-registry-check=false: Skip Docker daemon registry check
--write-config=false: Write the configuration files into host config dir
But you shouln't use it, because "cluster up" is removed in version 4.0.0. More here: https://github.com/openshift/origin/pull/21399