Log

Project log

Scheduler should be capable of:

• allocate nodes
• dis-allocate nodes
• gather metrics
• have a policy/rule

Apparently nodes are going to be dependent of the underling platform.

Timeline

1. See what's available on the market regarding auto-scalers

• PoC Autoscaler based on Marathon (DC/OS)
• PoC SwarmScale

2. Deployment

Deploy RancherOS on amazon EC2

• Reference image is ami-58fe0238 on us-west-2 (Oregon)
• Create security group sg-65884803 named zenscaler-test
• Deploy RancherOS using webui + Rancher
• Deploy RancherOS using aws-cli

$ aws ec2 run-instances --image-id ami-58fe0238 --count 1 --instance-type t2.small --security-groups zenscaler-test
small --security-groups zenscaler-test
301517625970	r-c936e20d
INSTANCES	0	x86_64		False	xen	ami-58fe0238	i-18e503b7	t2.small	2016-06-02T14:03:36.000Z	ip-172-31-35-166.us-west-2.compute.internal	172.31.35.166		/dev/sda1ebs	True		subnet-79e9511c	hvm	vpc-1b7add7e
MONITORING	disabled
NETWORKINTERFACES		02:e4:81:7b:02:29	eni-e8b7cb92	301517625970	ip-172-31-35-166.us-west-2.compute.internal	172.31.35.166	True	in-use	subnet-79e9511c	vpc-1b7add7e
ATTACHMENT	2016-06-02T14:03:36.000Z	eni-attach-7ea2bb89	True	0	attaching
GROUPS	sg-65884803	zenscaler-test
PRIVATEIPADDRESSES	True	ip-172-31-35-166.us-west-2.compute.internal	172.31.35.166
PLACEMENT	us-west-2b		default
SECURITYGROUPS	sg-65884803	zenscaler-test
STATE	0	pending
STATEREASON	pending	pending

Deploy DC/OS

DC/OS setup to AWS is described here and use a template on CloudFormation.

Tested

Deploy CoreOS+Kubernetes

• Done using the official coreOS repo for AWS (you'll need kubectl)
• Also a WIP as an ansible script in zenscaler/deploy/kubernetes

Deploy basic Debian + swarm

• Got an (untested) reference.
• Otherwise done locally, see zenscaler/deploy/swarm.

3. Define base architecture

  +-----------+
  |   Scaler  |--- (spawn nodes)
  +-----+-----+
        |
        |              [NODE_1]
+---------------+     /
| Load balancer |----+ - [NODE_2]
+-------+-------+     \ 
        |             [NODE_3]
  +-----+-----+
  | Core node |
  +-----------+

3. Ansible

Tech stack overview

Roles

• Bare stuff AWS, baremetal, DO, VMs
• OSs RancherOS, CoreOS, DC/OS, debian, CentOS 7
• Deployment Ansible
• Infra management Rancher, Apache Meso, Kuberneties, Docker swarm
• Service management etcd, zookeeper
• Task management/Scheduler Marathon
• Proxy træfik, HAProxy, nginx
• Storage Torus (beta)

Affinity stack

• CoreOS+Kubernetes+etcd

Ideas

• Docker Swarm allows affinity, to be correlated with resources use?

Other resources

• The end of pets