def SingleStat(
title,
id,
targets,
colorValue=False,
gauge=Gauge(show=True),
valueFontSize='80%',
thresholds=None,
valueName='avg',
valueMaps=None,
rangeMaps=None,
mappingTypes=None,
mappingType=None,
postfix=None,
sparkline=SparkLine(),
prefixFontSize='50%',
colors=[
(50, 172, 45, 0.97),
(237, 129, 40, 0.89),
(245, 54, 54, 0.9),
],
span=None,
format='none',
transparent=None,
):
def merge_target(target):
return {
**{
'intervalFactor': 2,
'refId': 'A',
'step': 600,
},
**target
}
targets = [merge_target(t) for t in targets]
return core.SingleStat(
title=title,
id=id,
colorValue=colorValue,
dataSource='${DS_PROMETHEUS}',
gauge=gauge,
valueFontSize=valueFontSize,
thresholds=thresholds,
valueName=valueName,
valueMaps=valueMaps,
rangeMaps=rangeMaps,
mappingTypes=mappingTypes,
targets=targets,
mappingType=mappingType,
format=format,
colors=colors,
span=span,
postfix=postfix,
sparkline=sparkline,
prefixFontSize=prefixFontSize,
hideTimeOverride=None,
transparent=transparent,
)
def test_single_stat():
data_source = 'dummy data source'
targets = ['dummy_prom_query']
title = 'dummy title'
single_stat = G.SingleStat(data_source, targets, title)
data = single_stat.to_json_data()
assert data['targets'] == targets
assert data['datasource'] == data_source
assert data['title'] == title
def make(prefix, title):
def target(expr, **kw):
return G.Target(expr=expr.format(prefix), **kw)
return G.Dashboard(
title=title,
rows=[
G.Row(panels=[
G.SingleStat(
title='Pods up (web)',
dataSource='prometheus',
valueName='current',
sparkline=G.SparkLine(show=True),
targets=[
target(
expr=
'count by(service) (up{{service="{}-isaacranks-web"}} == 1)'
)
]),
G.SingleStat(
title='Pods up (rebuild)',
dataSource='prometheus',
valueName='current',
sparkline=G.SparkLine(show=True),
targets=[
target(
expr=
'count by(service) (up{{service="{}-isaacranks-rebuild"}} == 1)'
)
]),
]),
G.Row(panels=[
G.Graph(
title='HTTP RPS',
dataSource='prometheus',
targets=[
target(
expr=
'service_status:http_request_duration_seconds_count:irate{{service="{}-isaacranks-web",status_code=~"1.."}}',
legendFormat='1xx',
refId='A'),
target(
expr=
'service_status:http_request_duration_seconds_count:irate{{service="{}-isaacranks-web",status_code=~"2.."}}',
legendFormat='2xx',
refId='B'),
target(
expr=
'service_status:http_request_duration_seconds_count:irate{{service="{}-isaacranks-web",status_code=~"3.."}}',
legendFormat='3xx',
refId='C'),
target(
expr=
'service_status:http_request_duration_seconds_count:irate{{service="{}-isaacranks-web",status_code=~"4.."}}',
legendFormat='4xx',
refId='D'),
target(
expr=
'service_status:http_request_duration_seconds_count:irate{{service="{}-isaacranks-web",status_code=~"5.."}}',
legendFormat='5xx',
refId='E'),
],
aliasColors=ALIAS_COLORS,
yAxes=[
G.YAxis(format=G.OPS_FORMAT),
G.YAxis(format=G.SHORT_FORMAT, show=False)
],
nullPointMode=G.NULL_AS_ZERO,
stack=True,
lineWidth=0,
fill=10,
tooltip=G.Tooltip(valueType=G.INDIVIDUAL)),
G.Graph(
title='HTTP latency',
dataSource='prometheus',
targets=[
target(
expr=
'service:http_request_duration_seconds:50p{{service="{}-isaacranks-web"}} * 1000',
legendFormat='0.5q',
refId='A'),
target(
expr=
'service:http_request_duration_seconds:90p{{service="{}-isaacranks-web"}} * 1000',
legendFormat='0.9q',
refId='B'),
target(
expr=
'service:http_request_duration_seconds:99p{{service="{}-isaacranks-web"}} * 1000',
legendFormat='0.99q',
refId='C'),
],
aliasColors=ALIAS_COLORS,
yAxes=[
G.YAxis(format=G.MILLISECONDS_FORMAT),
G.YAxis(format=G.SHORT_FORMAT, show=False)
]),
]),
G.Row(panels=[
G.Graph(
title='Ballots',
dataSource='prometheus',
targets=[
target(
expr=
'service_version:isaacranks_ballot_generation_seconds_count:irate{{service="{}-isaacranks-web"}}',
legendFormat='{{version}}',
refId='A')
],
yAxes=[
G.YAxis(format=G.OPS_FORMAT),
G.YAxis(format=G.SHORT_FORMAT, show=False),
],
nullPointMode=G.NULL_AS_ZERO,
stack=True,
lineWidth=0,
fill=10,
tooltip=G.Tooltip(valueType=G.INDIVIDUAL)),
G.Graph(
title='Ballot latency',
dataSource='prometheus',
targets=[
target(
expr=
'service:isaacranks_ballot_generation_seconds:50p{{service="{}-isaacranks-web"}} * 1000',
legendFormat='0.5q',
refId='A'),
target(
expr=
'service:isaacranks_ballot_generation_seconds:90p{{service="{}-isaacranks-web"}} * 1000',
legendFormat='0.9q',
refId='B'),
target(
expr=
'service:isaacranks_ballot_generation_seconds:99p{{service="{}-isaacranks-web"}} * 1000',
legendFormat='0.99q',
refId='C'),
],
yAxes=[
G.YAxis(format=G.MILLISECONDS_FORMAT),
G.YAxis(format=G.SHORT_FORMAT, show=False)
]),
]),
G.Row(panels=[
G.Graph(
title='Votes',
dataSource='prometheus',
targets=[
target(
expr=
'service_version:isaacranks_vote_casting_seconds_count:irate{{service="{}-isaacranks-web"}}',
legendFormat='{{version}}',
refId='A')
],
yAxes=[
G.YAxis(format=G.OPS_FORMAT),
G.YAxis(format=G.SHORT_FORMAT, show=False)
],
nullPointMode=G.NULL_AS_ZERO,
stack=True,
lineWidth=0,
fill=10,
tooltip=G.Tooltip(valueType=G.INDIVIDUAL)),
G.Graph(
title='Vote latency',
dataSource='prometheus',
targets=[
target(
expr=
'service:isaacranks_vote_casting_seconds:50p{{service="{}-isaacranks-web"}} * 1000',
legendFormat='0.5q',
refId='A'),
target(
expr=
'service:isaacranks_vote_casting_seconds:90p{{service="{}-isaacranks-web"}} * 1000',
legendFormat='0.9q',
refId='B'),
target(
expr=
'service:isaacranks_vote_casting_seconds:99p{{service="{}-isaacranks-web"}} * 1000',
legendFormat='0.99q',
refId='C'),
],
yAxes=[
G.YAxis(format=G.MILLISECONDS_FORMAT),
G.YAxis(format=G.SHORT_FORMAT, show=False)
]),
]),
G.Row(panels=[
G.Graph(
title='Time since last rebuild',
dataSource='prometheus',
targets=[
target(
expr=
'time() - (isaacranks_last_rebuild_timestamp{{service="{}-isaacranks-rebuild"}} != 0)',
legendFormat='Age')
],
legend=G.Legend(current=True),
yAxes=[
G.YAxis(format=G.SECONDS_FORMAT),
G.YAxis(format=G.SHORT_FORMAT, show=False)
]),
G.Graph(
title='Rebuild duration',
dataSource='prometheus',
targets=[
target(
expr=
'isaacranks_last_rebuild_duration_seconds{{service="{}-isaacranks-rebuild"}} != 0',
legendFormat='Duration')
],
legend=G.Legend(current=True),
yAxes=[
G.YAxis(format=G.SECONDS_FORMAT),
G.YAxis(format=G.SHORT_FORMAT, show=False)
]),
])
]).auto_panel_ids()
def dashboard():
PROMETHEUS = "prometheus"
return G.Dashboard(
title="S4",
rows=[
G.Row(panels=[
G.Graph(
title="Signups",
dataSource=PROMETHEUS,
xAxis=X_TIME,
yAxes=[
G.YAxis(
format="none",
label="Count",
),
G.YAxis(
format="none",
label="Count",
),
],
targets=[
G.Target(
expr='wormhole_signup_started{pod=~"s4-signup.*"}',
legendFormat="Wormhole Signups Started",
refId="A",
),
G.Target(
expr='wormhole_signup_success{pod=~"s4-signup.*"}',
legendFormat="Wormhole Signups Completed",
refId="B",
),
G.Target(
expr='wormhole_signup_failure{pod=~"s4-signup.*"}',
legendFormat="Wormhole Signups Failed",
refId="C",
),
],
),
G.Graph(
title="Usage",
dataSource=PROMETHEUS,
# Stack the connection graphs on each other, revealing
# both a total and a distribution across different grid
# router instances.
stack=True,
tooltip=G.Tooltip(
valueType=G.INDIVIDUAL,
),
xAxis=X_TIME,
yAxes=[
G.YAxis(
format="none",
label="Count",
),
G.YAxis(
format="none",
label="Count",
),
],
targets=[
G.Target(
expr="grid_router_connections",
legendFormat="Tahoe-LAFS Connections",
refId="D",
),
],
),
]),
G.Row(
title="Cluster",
panels=[
cpu_usage(PROMETHEUS, ["1m", "5m", "10m"]),
memory_usage(PROMETHEUS),
network_usage(PROMETHEUS),
filesystem_usage(PROMETHEUS),
],
),
G.Row(panels=[
G.SingleStat(
title='Current Customer Deployments',
dataSource='prometheus',
valueName='current',
sparkline=G.SparkLine(show=True),
targets=[
G.Target(
expr='s4_deployment_gauge',
refId="E",
),
],
),
G.SingleStat(
title='Unhandled Errors',
dataSource='prometheus',
valueName='current',
sparkline=G.SparkLine(show=True),
targets=[
G.Target(
expr='s4_unhandled_error_counter',
refId="F",
),
],
),
]),
],
).auto_panel_ids()
G.SingleStat(
editable=True,
colorBackground=True,
colorValue=False,
colors=[G.ORANGE, G.GREEN, G.RED],
dataSource="Prometheus",
gauge=G.Gauge(
maxValue=100,
minValue=0,
show=False,
thresholdLabels=False,
thresholdMarkers=True,
),
title=app["name"],
targets=[
G.Target(
expr='up{{instance="{hostname}:{port}"}}'.format(
hostname=app["hostname"], port=app["port"]),
intervalFactor=2,
legendFormat="",
step=4,
)
],
span=2,
thresholds="0.1",
transparent=True,
valueMaps=[
G.ValueMap(op="=", text="N/A", value="null"),
G.ValueMap(op="=", text="Down", value="0"),
G.ValueMap(op="=", text="Up", value="1"),
],
sparkline=G.SparkLine(
fillColor=G.RGBA(31, 118, 109, 0.18),
full=False,
lineColor=G.RGB(31, 120, 193),
show=False,
),
rangeMaps=[G.RangeMap(start=None, text="N/A", end=None)],
valueName="current",
) for app in APPS
"3xx": BLUE,
"4xx": ORANGE,
"5xx": RED,
"success": GREEN,
"error": RED,
}
dashboard = G.Dashboard(
title='UCDAPI',
rows=[
G.Row(panels=[
G.SingleStat(
title='Pods Up',
dataSource='prometheus',
valueName='current',
sparkline=G.SparkLine(show=True),
targets=[
G.Target(
expr='count by(service) (up{service="ucdapi"} == 1)', )
]),
]),
G.Row(panels=[
G.Graph(
title='HTTP RPS',
dataSource='prometheus',
targets=[
G.Target(
expr=
'service_status:http_request_duration_seconds_count:irate{service="ucdapi",status_code=~"1.."}',
legendFormat='1xx',
refId='A'),
'kubelet_volume_stats_used_bytes{namespace="cortex"} / kubelet_volume_stats_capacity_bytes{namespace="cortex"}'
),
],
yAxes=common.PercentageAxes(),
),
],
),
G.Row(
title="Last runs",
panels=[
G.SingleStat(
dataSource=common.PROMETHEUS,
title="Last Successful Compactor Run",
targets=[
G.Target(
'(time()-cortex_compactor_last_successful_run_timestamp_seconds) / 60',
refId='A',
),
],
format='m', # TODO: Add 'MINUTES_FORMAT' to grafanalib
),
G.SingleStat(
dataSource=common.PROMETHEUS,
title="Last Successful Bucket Index Update",
targets=[
G.Target(
'(time()-max(cortex_bucket_index_last_successful_update_timestamp_seconds)) / 60',
refId='A',
),
],
format='m', # TODO: Add 'MINUTES_FORMAT' to grafanalib