[Compass-ci] Re: [PATCH compass-ci] sched: refactor sched class for the lkp cluster sync

On Mon, Nov 09, 2020 at 03:20:32PM +0800, Cao Xueliang wrote:

...

• # all request states:
• # wait_ready | abort | failed | finished | wait_finish |
• # write_state | roles_ip
• def request_cluster_state(env)

The function should be the first function in the file.

Got it.

Thanks, RenWen

Thanks, Xueliang

...

• request_state = env.params.query["state"]
• job_id = env.params.query["job_id"]
• cluster_id = @redis.hash_get("sched/id2cluster", job_id).not_nil!
• cluster_state = ""
• states = {"abort" => "abort",

•          "finished"    => "finish",
  

•          "failed"      => "abort",
  

•          "wait_ready"  => "ready",
  

•          "wait_finish" => "finish"}
  

• case request_state
• when "abort", "finished", "failed"

•  # update node state only
  

•  update_cluster_state(cluster_id, job_id, {"state" => states[request_state]})
  

• when "wait_ready"

•  update_cluster_state(cluster_id, job_id, {"state" => states[request_state]})
  

•  @block_helper.block_until_finished(cluster_id) {
  

•    cluster_state = sync_cluster_state(cluster_id, job_id, states[request_state])
  

•    cluster_state == "ready" || cluster_state == "abort"
  

•  }
  

•  return cluster_state
  

• when "wait_finish"

•  update_cluster_state(cluster_id, job_id, {"state" => states[request_state]})
  

•  while 1
  

•    sleep(10)
  

•    cluster_state = sync_cluster_state(cluster_id, job_id, states[request_state])
  

•    break if (cluster_state == "finish" || cluster_state == "abort")
  

•  end
  

•  return cluster_state
  

• when "write_state"

•  node_roles = env.params.query["node_roles"]
  

•  node_ip = env.params.query["ip"]
  

•  direct_ips = env.params.query["direct_ips"]
  

•  direct_macs = env.params.query["direct_macs"]
  

•  job_info = {"roles"       => node_roles,
  

•              "ip"          => node_ip,
  

•              "direct_ips"  => direct_ips,
  

•              "direct_macs" => direct_macs}
  

•  update_cluster_state(cluster_id, job_id, job_info)
  

• when "roles_ip"

•  role = "server"
  

•  role_state = get_role_state(cluster_id, role)
  

•  raise "Missing #{role} state in cluster state" unless role_state
  

•  return "server=#{role_state["ip"]}\n" \
  

•         "direct_server_ips=#{role_state["direct_ips"]}"
  

• end
• # show cluster state
• return @redis.hash_get("sched/cluster_state", cluster_id)
• end
• # get the node state of role from cluster_state
• private def get_role_state(cluster_id, role)
• cluster_state = get_cluster_state(cluster_id)
• cluster_state.each_value do |role_state|

•  return role_state if role_state["roles"] == role
  

• end
• end
• # node_state: "finish" | "ready"
• def sync_cluster_state(cluster_id, job_id, node_state)
• cluster_state = get_cluster_state(cluster_id)
• cluster_state.each_value do |host_state|

•  state = host_state["state"]
  

•  return "abort" if state == "abort"
  

• end
• cluster_state.each_value do |host_state|

•  state = host_state["state"]
  

•  next if "#{state}" == "#{node_state}"
  

•  return "retry"
  

• end
• # cluster state is node state when all nodes are normal
• return node_state
• end

+end

2.23.0