# NOTE: These rules were kindly contributed by the SoundCloud engineering team.

### Container resources ###

cluster_namespace_controller_pod_container:spec_memory_limit_bytes =
  sum by (cluster,namespace,controller,pod_name,container_name) (
    label_replace(
      container_spec_memory_limit_bytes{container_name!=""},
      "controller", "$1",
      "pod_name", "^(.*)-[a-z0-9]+"
    )
  )

cluster_namespace_controller_pod_container:spec_cpu_shares =
  sum by (cluster,namespace,controller,pod_name,container_name) (
    label_replace(
      container_spec_cpu_shares{container_name!=""},
      "controller", "$1",
      "pod_name", "^(.*)-[a-z0-9]+"
    )
  )

cluster_namespace_controller_pod_container:cpu_usage:rate =
  sum by (cluster,namespace,controller,pod_name,container_name) (
    label_replace(
      irate(
        container_cpu_usage_seconds_total{container_name!=""}[5m]
      ),
      "controller", "$1",
      "pod_name", "^(.*)-[a-z0-9]+"
    )
  )

cluster_namespace_controller_pod_container:memory_usage:bytes =
  sum by (cluster,namespace,controller,pod_name,container_name) (
    label_replace(
      container_memory_usage_bytes{container_name!=""},
      "controller", "$1",
      "pod_name", "^(.*)-[a-z0-9]+"
    )
  )

cluster_namespace_controller_pod_container:memory_working_set:bytes =
  sum by (cluster,namespace,controller,pod_name,container_name) (
    label_replace(
      container_memory_working_set_bytes{container_name!=""},
      "controller", "$1",
      "pod_name", "^(.*)-[a-z0-9]+"
    )
  )

cluster_namespace_controller_pod_container:memory_rss:bytes =
  sum by (cluster,namespace,controller,pod_name,container_name) (
    label_replace(
      container_memory_rss{container_name!=""},
      "controller", "$1",
      "pod_name", "^(.*)-[a-z0-9]+"
    )
  )

cluster_namespace_controller_pod_container:memory_cache:bytes =
  sum by (cluster,namespace,controller,pod_name,container_name) (
    label_replace(
      container_memory_cache{container_name!=""},
      "controller", "$1",
      "pod_name", "^(.*)-[a-z0-9]+"
    )
  )

cluster_namespace_controller_pod_container:disk_usage:bytes =
  sum by (cluster,namespace,controller,pod_name,container_name) (
    label_replace(
      container_disk_usage_bytes{container_name!=""},
      "controller", "$1",
      "pod_name", "^(.*)-[a-z0-9]+"
    )
  )

cluster_namespace_controller_pod_container:memory_pagefaults:rate =
  sum by (cluster,namespace,controller,pod_name,container_name,scope,type) (
    label_replace(
      irate(
        container_memory_failures_total{container_name!=""}[5m]
      ),
      "controller", "$1",
      "pod_name", "^(.*)-[a-z0-9]+"
    )
  )

cluster_namespace_controller_pod_container:memory_oom:rate =
  sum by (cluster,namespace,controller,pod_name,container_name,scope,type) (
    label_replace(
      irate(
        container_memory_failcnt{container_name!=""}[5m]
      ),
      "controller", "$1",
      "pod_name", "^(.*)-[a-z0-9]+"
    )
  )

### Cluster resources ###

cluster:memory_allocation:percent =
  100 * sum by (cluster) (
    container_spec_memory_limit_bytes{pod_name!=""}
  ) / sum by (cluster) (
    machine_memory_bytes
  )

cluster:memory_used:percent =
  100 * sum by (cluster) (
    container_memory_usage_bytes{pod_name!=""}
  ) / sum by (cluster) (
    machine_memory_bytes
  )

cluster:cpu_allocation:percent =
  100 * sum by (cluster) (
    container_spec_cpu_shares{pod_name!=""}
  ) / sum by (cluster) (
    container_spec_cpu_shares{id="/"} * on(cluster,instance) machine_cpu_cores
  )

cluster:node_cpu_use:percent =
  100 * sum by (cluster) (
    rate(node_cpu{mode!="idle"}[5m])
  ) / sum by (cluster) (
    machine_cpu_cores
  )

### API latency ###

# Raw metrics are in microseconds. Convert to seconds.
cluster_resource_verb:apiserver_latency:quantile_seconds{quantile="0.99"} =
  histogram_quantile(
    0.99,
    sum by(le,cluster,job,resource,verb) (apiserver_request_latencies_bucket)
  ) / 1e6
cluster_resource_verb:apiserver_latency:quantile_seconds{quantile="0.9"} =
  histogram_quantile(
    0.9,
    sum by(le,cluster,job,resource,verb) (apiserver_request_latencies_bucket)
  ) / 1e6
cluster_resource_verb:apiserver_latency:quantile_seconds{quantile="0.5"} =
  histogram_quantile(
    0.5,
    sum by(le,cluster,job,resource,verb) (apiserver_request_latencies_bucket)
  ) / 1e6

### Scheduling latency ###

cluster:scheduler_e2e_scheduling_latency:quantile_seconds{quantile="0.99"} =
  histogram_quantile(0.99,sum by (le,cluster) (scheduler_e2e_scheduling_latency_microseconds_bucket)) / 1e6
cluster:scheduler_e2e_scheduling_latency:quantile_seconds{quantile="0.9"} =
  histogram_quantile(0.9,sum by (le,cluster) (scheduler_e2e_scheduling_latency_microseconds_bucket)) / 1e6
cluster:scheduler_e2e_scheduling_latency:quantile_seconds{quantile="0.5"} =
  histogram_quantile(0.5,sum by (le,cluster) (scheduler_e2e_scheduling_latency_microseconds_bucket)) / 1e6

cluster:scheduler_scheduling_algorithm_latency:quantile_seconds{quantile="0.99"} =
  histogram_quantile(0.99,sum by (le,cluster) (scheduler_scheduling_algorithm_latency_microseconds_bucket)) / 1e6
cluster:scheduler_scheduling_algorithm_latency:quantile_seconds{quantile="0.9"} =
  histogram_quantile(0.9,sum by (le,cluster) (scheduler_scheduling_algorithm_latency_microseconds_bucket)) / 1e6
cluster:scheduler_scheduling_algorithm_latency:quantile_seconds{quantile="0.5"} =
  histogram_quantile(0.5,sum by (le,cluster) (scheduler_scheduling_algorithm_latency_microseconds_bucket)) / 1e6

cluster:scheduler_binding_latency:quantile_seconds{quantile="0.99"} =
  histogram_quantile(0.99,sum by (le,cluster) (scheduler_binding_latency_microseconds_bucket)) / 1e6
cluster:scheduler_binding_latency:quantile_seconds{quantile="0.9"} =
  histogram_quantile(0.9,sum by (le,cluster) (scheduler_binding_latency_microseconds_bucket)) / 1e6
cluster:scheduler_binding_latency:quantile_seconds{quantile="0.5"} =
  histogram_quantile(0.5,sum by (le,cluster) (scheduler_binding_latency_microseconds_bucket)) / 1e6