Skip to content

Latest commit

 

History

History
90 lines (62 loc) · 2.8 KB

README.md

File metadata and controls

90 lines (62 loc) · 2.8 KB

go-work

This a repository to demonstrate how to have cluster of workers in go.

Design

The Leader performs two main functions: heartbeat management i.e. the health of the workers in the cluster, and scheduling new work based the workers available.

flowchart LR

subgraph leader
	c[cluster]
	s[scheduler]
end

w1["worker..n"]

w1-- health -->c
w1-- join / leave -->c
w1-- work complete -->s

s-- start/cancel work -->w1
c-- members -->s
Loading

There are 3 core contracts:

  • cluster - the cluster manages joining and leaving the cluster and the health of workers.
  • scheduler - the scheduler manages the receiving of work request and the dishing out to the workers.
  • worker - the worker receives commands to start (or stop) work, and then reports health and completion of work to the cluster and schedule respectively.

All the communication is over gRPC (non-TLS in this example) using client and server RPC calls (non streaming).

Testing

The test coverage in the codebase is relatively high, however, there are a few gaps that can be closed, the plan is to close these in the not too distant future.

Future

  • Write a CLI to get a list of members, tasks and schedule work
    • The system hangs after work is scheduled and you try to list tasks
    • Only one worker is scheduled to work no matter if more are free
  • Put the comms over mTLS (outside of the implementation code, i.e. NGINX or Envoy)
  • Covert the heartbeat comms between the cluster leader and the members (workers) over a stream
  • Create a consumer library to enable custom code to become a worker
  • Get it working in ECS
  • Implement RAFT and have multiple leaders
  • The current task list is an in-memory linked list, make more resilient
  • There are some TODOs in the code, work to remove them (mainly tests)

Note

I implemented streaming of the heartbeat and the code for the client was unpleasant, therefore, I decided to not merge the changes. I have left the branch intact in dev/streaming for the curious.

Useful Commands

List all the operations available on the Leader server.

grpcurl -plaintext :50051 describe cluster.v1.LeaderService

List all the operations available on the Scheduler server.

grpcurl -plaintext :50052 describe scheduler.v1.SchedulerService

Get the list of worker members in the cluster.

grpcurl -plaintext :50051 cluster.v1.LeaderService.Members

Get the list of tasks pushed through the leader.

grpcurl -plaintext :50052 scheduler.v1.SchedulerService.GetTasks

Add a task to the scheduler.

grpcurl -plaintext -d '{"task": {"id":"1","payload":"do some work"}}' :50052 scheduler.v1.SchedulerService.ScheduleTask

Cancel a task from the scheduler.

grpcurl -plaintext -d '{"id":"1"}' :50052 scheduler.v1.SchedulerService.CancelTask