k8s.io/perf-tests/clusterloader2@v0.0.0-20240304094227-64bdb12da87e/pkg/prometheus/manifests/dashboards/master_panels.py

#!/usr/bin/env python3

# Copyright 2022 The Kubernetes Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from grafanalib import core as g
import defaults as d


def api_call_latency_panel(expression):
    def api_call_latency(title, verb, scope, threshold):
        return d.Graph(
            title=title,
            targets=[
                d.Target(expr=str(threshold), legendFormat="threshold"),
                d.Target(
                    expr=d.one_line(expression % {"verb": verb, "scope": scope}
                                    ),
                    # TODO(github.com/grafana/grafana/issues/19410): uncomment once fixed
                    # legendFormat="{{verb}} {{scope}}/{{resource}}",
                ),
            ],
            yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
        )

    return [
        api_call_latency(
            title="GET resource latency (percentaile=99, scope=resource, threshold=1s)",
            verb="GET",
            scope="resource",
            threshold=1,
        ),
        api_call_latency(
            title="LIST namespace latency (percentaile=99, scope=namespace, threshold=5s)",
            verb="LIST",
            scope="namespace",
            threshold=5,
        ),
        api_call_latency(
            title="LIST cluster latency (percentaile=99, scope=cluster, threshold=30s)",
            verb="LIST",
            scope="cluster",
            threshold=30,
        ),
        api_call_latency(
            title="Mutating API call latency (threshold=1s)",
            verb=d.any_of("CREATE", "DELETE", "PATCH", "POST", "PUT"),
            scope=d.any_of("namespace", "cluster", "resource"),
            threshold=1,
        ),
    ]

API_CALL_LATENCY_PANELS = api_call_latency_panel("""
apiserver:apiserver_request_latency_1m:histogram_quantile{
  quantile="0.99",
  verb=~"%(verb)s",
  scope=~"%(scope)s",
  resource=~"${resource:regex}s*",
  subresource!~"exec|proxy",
}""")

QUANTILE_API_CALL_LATENCY_PANELS = api_call_latency_panel("""
quantile_over_time(0.99,
apiserver:apiserver_request_latency_1m:histogram_quantile{
  quantile="0.99",
  verb=~"%(verb)s",
  scope=~"%(scope)s",
  resource=~"${resource:regex}s*",
  subresource!~"exec|proxy",
}[5d])""")

APF_PANELS = [
    d.simple_graph(
        "Requests waiting time",
        "histogram_quantile(0.99, sum(rate(apiserver_flowcontrol_request_wait_duration_seconds_bucket{instance=~\".*(${instance:pipe})\"}[1m]))  by (le, instance, priority_level))",
        legend="{{instance}} {{priority_level}}",
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
    ),
    d.simple_graph(
        "Execution time",
        "histogram_quantile(0.99, sum(rate(apiserver_flowcontrol_request_execution_seconds_bucket{instance=~\".*(${instance:pipe})\"}[1m]))  by (le, instance, priority_level))",
        legend="{{instance}} {{priority_level}}",
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
    ),
    d.simple_graph(
        "Total execution time per second",
        "sum(irate(apiserver_flowcontrol_request_execution_seconds_sum{instance=~\".*(${instance:pipe})\"}[1m]))  by (instance, priority_level)",
        legend="{{instance}} {{priority_level}}",
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
    ),
    d.simple_graph(
        "Requests rate by priority level",
        "sum(irate(apiserver_flowcontrol_dispatched_requests_total{instance=~\".*(${instance:pipe})\"}[1m])) by (instance, priority_level)",
        legend="{{instance}} {{priority_level}}",
        yAxes=g.single_y_axis(format=g.OPS_FORMAT),
    ),
    d.simple_graph(
        "Concurrency in use",
        "sum(apiserver_flowcontrol_request_concurrency_in_use{instance=~\".*(${instance:pipe})\"}) by (instance, priority_level)",
        legend="{{instance}} {{priority_level}}",
        yAxes=g.single_y_axis(format=g.OPS_FORMAT),
    ),
    d.simple_graph(
        "Current executing requests",
        "sum(apiserver_flowcontrol_current_executing_requests{instance=~\".*(${instance:pipe})\"}) by (instance, priority_level)",
        legend="{{instance}} {{priority_level}}",
        yAxes=g.single_y_axis(format=g.OPS_FORMAT),
    ),
    d.simple_graph(
        "Inqueue requests",
        "sum(apiserver_flowcontrol_current_inqueue_requests{instance=~\".*(${instance:pipe})\"}) by (instance, priority_level)",
        legend="{{instance}} {{priority_level}}",
        yAxes=g.single_y_axis(format=g.OPS_FORMAT),
    ),
    d.simple_graph(
        "Nominal number of execution seats",
        "avg(apiserver_flowcontrol_nominal_limit_seats{instance=~\".*(${instance:pipe})\"}) by (priority_level)",
        legend="{{priority_level}}",
    ),
    d.simple_graph(
        "Lower bound on number of execution seats",
        "avg(apiserver_flowcontrol_lower_limit_seats{instance=~\".*(${instance:pipe})\"}) by (priority_level)",
        legend="{{priority_level}}",
    ),
    d.simple_graph(
        "Upper bound on number of execution seats",
        "avg(apiserver_flowcontrol_upper_limit_seats{instance=~\".*(${instance:pipe})\"}) by (priority_level)",
        legend="{{priority_level}}",
    ),
    d.simple_graph(
        "Number of seats Priority Level could use divided by nominal seats (50th percentile)",
        "histogram_quantile(0.5, rate(apiserver_flowcontrol_demand_seats_bucket{instance=~\".*(${instance:pipe})\"}[10s]))",
        legend="{{instance}} {{priority_level}}",
    ),
    d.simple_graph(
        "High watermark of demand seats over last adjustment period",
        "apiserver_flowcontrol_demand_seats_high_watermark{instance=~\".*(${instance:pipe})\"}",
        legend="{{instance}} {{priority_level}}",
    ),
    d.simple_graph(
        "Smoothed seat demands",
        "apiserver_flowcontrol_demand_seats_smoothed{instance=~\".*(${instance:pipe})\"}",
        legend="{{instance}} {{priority_level}}",
    ),
    d.simple_graph(
        "Current seat limit for each Priority Level",
        "apiserver_flowcontrol_current_limit_seats{instance=~\".*(${instance:pipe})\"}",
        legend="{{instance}} {{priority_level}}",
    ),
]

HEALTH_PANELS = [
    d.simple_graph(
        "Unhealthy nodes",
        "sum(node_collector_unhealthy_nodes_in_zone) by (zone)",
        legend="{{zone}}",
    ),
    d.simple_graph(
        "Pod creations",
        'sum(irate(apiserver_request_total{verb="POST", resource="pods", subresource=""}[1m]))',
        yAxes=g.single_y_axis(format=g.OPS_FORMAT),
    ),
    d.simple_graph(
        "Pod bindings",
        'sum(irate(apiserver_request_total{verb="POST", resource="pods", subresource="binding"}[1m]))',
        yAxes=g.single_y_axis(format=g.OPS_FORMAT),
    ),
    # It's not clear which "Component restarts" shows more accurate results.
    d.simple_graph(
        "Component restarts",
        "sum(rate(process_start_time_seconds[1m]) > bool 0) by (job, endpoint)",
    ),
    d.simple_graph(
        "Component restarts 2",
        'sum(min_over_time(container_start_time_seconds{container!="",container!="POD"}[2m])) by (container)',
    ),
    d.simple_graph(
        "Active component", "sum(leader_election_master_status) by (name, instance)"
    ),
]

ETCD_PANELS = [
    d.simple_graph("etcd leader", "etcd_server_is_leader", legend="{{instance}}"),
    d.simple_graph(
        "etcd bytes sent",
        "rate(etcd_network_client_grpc_sent_bytes_total[1m])",
        yAxes=g.single_y_axis(format=g.BYTES_PER_SEC_FORMAT),
        legend="{{instance}}",
    ),
    d.simple_graph(
        "etcd operations rate",
        d.one_line(
            """
sum(
  rate(
    etcd_request_duration_seconds_count{
      operation=~"${etcd_operation:regex}",
      type=~".*(${etcd_type:pipe})"
    }[1m]
  )
) by (operation, type)
"""
        ),
        yAxes=g.single_y_axis(format=g.OPS_FORMAT),
        legend="{{operation}} {{type}}",
    ),
    d.simple_graph(
        "etcd get latency by type (99th percentile)",
        d.one_line(
            """
histogram_quantile(
  0.99,
  sum(
    rate(
      etcd_request_duration_seconds_bucket{
        operation=~"${etcd_operation:regex}",
        type=~".*(${etcd_type:pipe})"
      }[1m]
    )
  ) by (le, operation, type, instance)
)
"""
        ),
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
        legend="{{operation}} {{type}} on {{instance}}",
    ),
    d.simple_graph(
        "etcd get latency by type (50th percentile)",
        d.one_line(
            """
histogram_quantile(
  0.50,
  sum(
    rate(
      etcd_request_duration_seconds_bucket{
        operation=~"${etcd_operation:regex}",
        type=~".*(${etcd_type:pipe})"
      }[1m]
    )
  ) by (le, operation, type, instance)
)
"""
        ),
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
    ),
    d.simple_graph("etcd instance id", "sum(etcd_server_id) by (instance, server_id)"),
    d.simple_graph(
        "etcd network latency (99th percentile)",
        "histogram_quantile(0.99, sum(rate(etcd_network_peer_round_trip_time_seconds_bucket[1m])) by (le, instance, To))",
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
    ),
    d.simple_graph(
        "etcd compaction keys",
        "delta(etcd_debugging_mvcc_db_compaction_keys_total[1m])",
    ),
    d.simple_graph(
        "etcd compaction pause sum duration",
        "delta(etcd_debugging_mvcc_db_compaction_pause_duration_milliseconds_sum[1m])",
        yAxes=g.single_y_axis(format=g.MILLISECONDS_FORMAT),
    ),
    d.simple_graph(
        "etcd compaction pause num chunks",
        "delta(etcd_debugging_mvcc_db_compaction_pause_duration_milliseconds_count[1m])",
    ),
    d.simple_graph(
        "etcd_disk_backend_commit_duration_seconds",
        "histogram_quantile(0.99, sum(rate(etcd_disk_backend_commit_duration_seconds_bucket[1m])) by (le, instance))",
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
    ),
    d.simple_graph(
        "etcd wal fsync duration",
        "histogram_quantile(1.0, sum(rate(etcd_disk_wal_fsync_duration_seconds_bucket[1m])) by (le, endpoint))",
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
    ),
    d.Graph(
        title="etcd compaction max pause",
        points=True,
        lines=False,
        targets=[
            d.Target(
                expr="histogram_quantile(1.0, sum(rate(etcd_debugging_mvcc_db_compaction_pause_duration_milliseconds_bucket[1m])) by (le, instance))"
            )
        ],
        yAxes=g.single_y_axis(format=g.MILLISECONDS_FORMAT),
    ),
    d.simple_graph(
        "etcd objects",
        "sum(etcd_object_counts) by (resource, instance)",
        legend="{{instance}}: {{resource}}",
    ),
    d.simple_graph(
        "etcd db size",
        [
            "etcd_mvcc_db_total_size_in_bytes",
            "etcd_mvcc_db_total_size_in_use_in_bytes",
            "etcd_server_quota_backend_bytes",
        ],
        yAxes=g.single_y_axis(format=g.BYTES_FORMAT),
    ),
]

APISERVER_PANELS = [
    d.simple_graph(
        "goroutines",
        'go_goroutines{job="master", endpoint="apiserver"}',
        legend="{{instance}}",
    ),
    d.simple_graph(
        "gc rate",
        'rate(go_gc_duration_seconds_count{job="master", endpoint="apiserver"}[1m])',
        legend="{{instance}}",
    ),
    d.simple_graph(
        "alloc rate",
        'rate(go_memstats_alloc_bytes_total{job="master", endpoint="apiserver"}[1m])',
        yAxes=g.single_y_axis(format=g.BYTES_PER_SEC_FORMAT),
        legend="{{instance}}",
    ),
    d.simple_graph(
        "Number of active watches",
        'sum(apiserver_registered_watchers{kind=~"(?i:(${resource:regex}))s*"}) by (instance, group, version, kind)',
        legend="{{instance}}: {{version}}.{{group}}.{{kind}}",
    ),
    d.simple_graph(
        "Watch events rate",
        d.one_line(
            """
sum(
  irate(
    apiserver_watch_events_total{
      kind=~"(?i:(${resource:regex}))s*"
    }[1m]
  )
) by (instance, group, version, kind)"""
        ),
        legend="{{instance}}: {{version}}.{{group}}.{{kind}}",
    ),
    d.simple_graph(
        "Watch events traffic",
        d.one_line(
            """
sum(
  irate(
    apiserver_watch_events_sizes_sum{
      kind=~"(?i:(${resource:regex}))s*"
   }[1m]
  )
) by (instance, group, version, kind)"""
        ),
        yAxes=g.single_y_axis(format=g.BYTES_PER_SEC_FORMAT),
        legend="{{instance}}: {{version}}.{{group}}.{{kind}}",
    ),
    d.simple_graph(
        "Watch event avg size",
        d.one_line(
            """
sum(
  rate(
    apiserver_watch_events_sizes_sum{
      kind=~"(?i:(${resource:regex}))s*"
    }[1m]
  )
  /
  rate(
    apiserver_watch_events_sizes_count{
      kind=~"(?i:(${resource:regex}))s*"
    }[1m]
  )
) by (instance, group, version, kind)"""
        ),
        legend="{{instance}}: {{version}}.{{group}}.{{kind}}",
    ),
    d.simple_graph(
        "Watch terminated total",
        "sum(rate(apiserver_terminated_watchers_total{}[1m])) by (resource, instance)",
        legend="{{instance}}: {{resource}}",
    ),
    d.simple_graph(
        "Inflight requests",
        "sum(apiserver_current_inflight_requests) by (requestKind, instance)",
        legend="{{instance}}: {{requestKind}}",
    ),
    d.simple_graph(
        "Request rate",
        d.one_line(
            """
sum(
  rate(
    apiserver_request_total{
      verb=~"${verb:regex}",
      resource=~"${resource:regex}s*"
    }[1m]
  )
) by (verb, resource, subresource, instance)"""
        ),
        # TODO(github.com/grafana/grafana/issues/19410): uncomment once fixed
        # legend="{{instance}}: {{verb}} {{resource}}",
    ),
    d.simple_graph(
        "Request rate by code",
        "sum(rate(apiserver_request_total[1m])) by (code, instance)",
        legend="{{instance}}: {{code}}",
    ),
    d.simple_graph(
        "Request latency (50th percentile) (excl. WATCH)",
        d.one_line(
            """
apiserver:apiserver_request_latency:histogram_quantile{
  quantile="0.50",
  verb!="WATCH",
  verb=~"${verb:regex}",
  resource=~"${resource:regex}s*"
}"""
        ),
        # TODO(github.com/grafana/grafana/issues/19410): uncomment once fixed
        # legend="{{verb}} {{scope}}/{{resource}}",
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
    ),
    d.simple_graph(
        "Request latency (99th percentile) (excl. WATCH)",
        d.one_line(
            """
apiserver:apiserver_request_latency:histogram_quantile{
  quantile="0.99",
  verb!="WATCH",
  verb=~"${verb:regex}",
   resource=~"${resource:regex}s*"
}"""
        ),
        # TODO(github.com/grafana/grafana/issues/19410): uncomment once fixed
        # legend="{{verb}} {{scope}}/{{resource}}",
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
    ),
    d.simple_graph(
        "Traffic (excl. WATCH)",
        d.one_line(
            """
sum(
  rate(
    apiserver_response_sizes_sum{
      verb!="WATCH",
      verb=~"${verb:regex}",
      resource=~"${resource:regex}s*"
    }[1m]
  )
) by (verb, version, resource, subresource, scope, instance)"""
        ),
        yAxes=g.single_y_axis(format=g.BYTES_PER_SEC_FORMAT),
    ),
    d.simple_graph(
        "Webhook admission duration (99th percentile)",
        "histogram_quantile(0.99, sum(rate(apiserver_admission_webhook_admission_duration_seconds_bucket[1m])) by (le, type, name))",
        legend="{{type}}: {{name}}",
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
    ),
    d.simple_graph(
        "Request filter latency for each filter type (99th percentile)",
        "histogram_quantile(0.99, sum(rate(apiserver_request_filter_duration_seconds_bucket[1m])) by (le, filter))",
        legend="{{filter}}",
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
    ),
    d.simple_graph(
        "Failed external requests",
        'sum(rate(rest_client_requests_total{endpoint="apiserver", code!="200", host!="[::1]:443"}[1m])) by (code, instance, method)',
        legend="{{instance}}: {{code}} {{method}}",
    ),
    d.simple_graph(
        "Extrernal requests latency (99th percentile)",
        'histogram_quantile(0.99, sum(rate(rest_client_request_duration_seconds_bucket{endpoint="apiserver", host!="[::1]:443"}[1m])) by (verb, host, instance, le))',
        legend="{{instance}}: {{verb}} {{host}}",
        yAxes=g.single_y_axis(format=g.SECONDS_FORMAT),
    )
]

CONTROLLER_MANAGER_PANELS = [
  d.simple_graph(
    "Workqueue depths",
    'workqueue_depth{endpoint="kube-controller-manager"}',
    legend="{{name}}",
  )
]

VM_PANELS = [
    d.simple_graph(
        "fs bytes reads by container",
        "sum(rate(container_fs_reads_bytes_total[1m])) by (container, instance)",
        legend="{{instance}}: {{container}}",
        yAxes=g.single_y_axis(format=g.BYTES_FORMAT),
    ),
    d.simple_graph(
        "fs reads by container",
        "sum(rate(container_fs_reads_total[1m])) by (container, instance)",
        legend="{{instance}}: {{container}}",
    ),
    d.simple_graph(
        "fs bytes writes by container",
        "sum(rate(container_fs_writes_bytes_total[1m])) by (container, instance)",
        legend="{{instance}}: {{container}}",
        yAxes=g.single_y_axis(format=g.BYTES_FORMAT),
    ),
    d.simple_graph(
        "fs writes by container",
        "sum(rate(container_fs_writes_total[1m])) by (container, instance)",
        legend="{{instance}}: {{container}}",
    ),
    d.Graph(
        title="CPU usage by container",
        targets=[
            d.TargetWithInterval(
                expr='sum(rate(container_cpu_usage_seconds_total{container!=""}[1m])) by (container, instance)',
                legendFormat="{{instance}}: {{container}}",
            ),
            d.TargetWithInterval(expr="machine_cpu_cores", legendFormat="limit"),
        ],
    ),
    d.Graph(
        title="memory usage by container",
        targets=[
            d.TargetWithInterval(
                expr='sum(container_memory_usage_bytes{container!=""}) by (container, instance)',
                legendFormat="{{instance}}: {{container}}",
            ),
            d.TargetWithInterval(expr="machine_memory_bytes", legendFormat="limit"),
        ],
        yAxes=g.single_y_axis(format=g.BYTES_FORMAT),
    ),
    d.Graph(
        title="memory working set by container",
        targets=[
            d.TargetWithInterval(
                expr='sum(container_memory_working_set_bytes{container!=""}) by (container, instance)',
                legendFormat="{{instance}}: {{container}}",
            ),
            d.TargetWithInterval(expr="machine_memory_bytes", legendFormat="limit"),
        ],
        yAxes=g.single_y_axis(format=g.BYTES_FORMAT),
    ),
    d.Graph(
        title="Network usage (bytes)",
        targets=[
            d.Target(
                expr='rate(container_network_transmit_bytes_total{id="/"}[1m])',
                legendFormat="{{instance}} transmit",
            ),
            d.Target(
                expr='rate(container_network_receive_bytes_total{id="/"}[1m])',
                legendFormat="{{instance}} receive",
            ),
        ],
        yAxes=g.single_y_axis(format=g.BYTES_PER_SEC_FORMAT),
    ),
    d.Graph(
        title="Network usage (packets)",
        targets=[
            d.Target(
                expr='rate(container_network_transmit_packets_total{id="/"}[1m])',
                legendFormat="{{instance}} transmit",
            ),
            d.Target(
                expr='rate(container_network_receive_packets_total{id="/"}[1m])',
                legendFormat="{{instance}} receive",
            ),
        ],
    ),
    d.Graph(
        title="Network usage (avg packet size)",
        targets=[
            d.Target(
                expr='rate(container_network_transmit_bytes_total{id="/"}[1m]) / rate(container_network_transmit_packets_total{id="/"}[1m])',
                legendFormat="{{instance}} transmit",
            ),
            d.Target(
                expr='rate(container_network_receive_bytes_total{id="/"}[1m]) / rate(container_network_receive_packets_total{id="/"}[1m])',
                legendFormat="{{instance}} receive",
            ),
        ],
        yAxes=g.single_y_axis(format=g.BYTES_FORMAT),
    ),
    d.Graph(
        title="Network tcp segments",
        targets=[
            d.Target(
                expr="sum(rate(node_netstat_Tcp_InSegs[1m])) by (instance)",
                legendFormat="InSegs {{instance}}",
            ),
            d.Target(
                expr="sum(rate(node_netstat_Tcp_OutSegs[1m])) by (instance)",
                legendFormat="OutSegs {{instance}}",
            ),
            d.Target(
                expr="sum(rate(node_netstat_Tcp_RetransSegs[1m])) by (instance)",
                legendFormat="RetransSegs {{instance}}",
            ),
        ],
        yAxes=g.single_y_axis(format=g.SHORT_FORMAT, logBase=10),
    ),
]