Platform Resource Management

[drniiken]

GOAL
As DevOps, I want to have full control of hardware resources consumed by the platform. I also need good monitoring/logging/debugging tools to keep track of issues and problems.

Lifetime of Services

• if a user logs out of the platform, resources should automatically be freed
• after a long time of inactivity, resources should automatically be freed

Protocol for Problem Handling

• what to do if a node fails
• what to do if a service fails

Monitoring Tools

• access to swarm logs
• access to node logs
• access to service logs
• look into available tools, i.e., docker EE

Hardware Allocation

• automation of cloud resource management
• integration into scheduler

[oetiker]

resources should probably be logged 'per user' so that we know who is causing the damage

[esraneufeld]

high priority (to be addressed for now):

• Lifetime of Services

[sanderegg]

State:

• book-keeping of services started by user/study for each opened browser tab:

  1. by monitoring websocket connection/disconnection (1 websocket per browser tab)
  2. store resources in external InMemory data store "Redis" to allow scaling of webserver in case of heavy load

• allows auto-de-allocation of services when:

  1. browser tab is closed
  2. user logs-out
  3. user disconnects for X amount of time
  4. user is inactive for Y amount of time
  5. anonymous user closes tabs or is inactive

• deleting of project deletes all project resources, e.g. services, internal data and database entries

[sanderegg]

State:

• book-keeping of services started by user/study for each opened browser tab:
  • by monitoring websocket connection/disconnection (1 websocket per browser tab)
  • store resources in external InMemory data store "Redis" to allow scaling of webserver in case of heavy load
• dynamic services are now auto-deallocated when:
  • browser tab is closed
  • user actively logs out (will also logout any other browser tab open of the same user)
  • user disconnects for more than 15 minutes (network down), during that time the services are kept up
  • user closes a project by going to the dashboard (unless another tab is still opened)
• deleting a project deletes the project resources
• deleting a node inside a project deletes the nodes resources
• Refreshing the browser page while a project is open now re-opens the same project instead of bringing the user to the dashboard
• Added a service to overview the Redis database state called redis-commander