package qos.v1alpha1.qos

Get desktop application:
View/edit binary Protocol Buffers messages

gRPC service for the interface between Peloton and QoS advisor

• rpc GetHostMetrics (GetHostMetricsRequest, GetHostMetricsResponse)

  qos.proto:71

  Get the metrics of all hosts in the cluster

  message GetHostMetricsRequest

  qos.proto:13

  (message has no fields)

  message GetHostMetricsResponse

  qos.proto:15

  • map<string, Metrics> hosts = 1

    Map from host name to metric(s) that serve as an abstraction for load

• rpc GetServicesImpacted (GetServicesImpactedRequest, GetServicesImpactedResponse)

  qos.proto:75

  Get all service instances in the cluster along with associated CPU and memory metrics. This API returns a map of host name to services. If a host has at least one impacted service, the API will return all sevices with their associated impact scores and actual resource usage.

  message GetServicesImpactedRequest

  qos.proto:20

  (message has no fields)

  message GetServicesImpactedResponse

  qos.proto:22

  • map<string, ServiceInstances> instances = 1

    Map from host name to list of service instances along with resource requirements

Flexible representation of load

Used in: GetHostMetricsResponse

• sint32 score = 1

  Score ranges from 0 (least loaded/utilized) to 100 (most loaded/utilized)

Resources represents the set of relevant resources

Used in: ServiceInstance, ServiceInstances

• double cpu = 1

  Number of CPUs (can be fractional, e.g., 3.25 cores)

• uint64 memory = 2

  Amount of memory (bytes)

Abstraction for representing service instances

Used in: ServiceInstances

• string service_id = 1

  Service identifier - this will be the Peloton job UUID

• string instance_id = 2

  Instance identifier (can be Mesos task id or pod name or pod id)

• map<string, string> tags = 3

  Tags will contain information about some QoS specific metadata. These are not the labels in the pod.

• optional Resources resource_usage = 4

  Resources used by this service instance

• sint32 impact_score = 5

  Impact score ranges from 0 to 100 and is a measure of the extent to which a service instance is unable to get the configured resources such as CPU and memory. Higher values mean the service instance is impacted more and 0 means no impact.

List of service instances on a host. Each service instance has an impact score. The impact score ranges from 0 to 100 and is a measure of the extent to which a service instance is unable to get the configured resources such as CPU and memory.

Used in: GetServicesImpactedResponse

• repeated ServiceInstance instances = 1

  List of service instances

• optional Resources resources_needed = 2

  Resources that need to be freed up on the host