RepoMirrors
/
prometheus
mirror of https://github.com/prometheus/prometheus
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

add cluster variable to Overview dashboard 

Signed-off-by: Erik Sommer <ersotech@posteo.de>
This commit is contained in:
Erik Sommer 2023-11-23 17:04:57 +01:00
parent 0102425af1
commit 0e585bf5c0
1 changed files with 24 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

46
documentation/prometheus-mixin/dashboards.libsonnet
View File

@ -13,16 +13,18 @@ local template = grafana.template;
g.dashboard(
'%(prefix)sOverview' % $._config.grafanaPrometheus
)
.addMultiTemplate('job', 'prometheus_build_info{%(prometheusSelector)s}' % $._config, 'job')
.addMultiTemplate('instance', 'prometheus_build_info{job=~"$job"}', 'instance')
.addMultiTemplate('cluster', 'prometheus_build_info{%(prometheusSelector)s}' % $._config, 'cluster')
.addMultiTemplate('job', 'prometheus_build_info{cluster=~"$cluster"}', 'job')
.addMultiTemplate('instance', 'prometheus_build_info{cluster=~"$cluster", job=~"$job"}', 'instance')
.addRow(
g.row('Prometheus Stats')
.addPanel(
g.panel('Prometheus Stats') +
g.tablePanel([
'count by (job, instance, version) (prometheus_build_info{job=~"$job", instance=~"$instance"})',
'max by (job, instance) (time() - process_start_time_seconds{job=~"$job", instance=~"$instance"})',
'count by (cluster, job, instance, version) (prometheus_build_info{cluster=~"$cluster", job=~"$job", instance=~"$instance"})',
'max by (cluster, job, instance) (time() - process_start_time_seconds{cluster=~"$cluster", job=~"$job", instance=~"$instance"})',
], {
cluster: { alias: 'Cluster' },
job: { alias: 'Job' },
instance: { alias: 'Instance' },
version: { alias: 'Version' },
@ -35,12 +37,12 @@ local template = grafana.template;
g.row('Discovery')
.addPanel(
g.panel('Target Sync') +
g.queryPanel('sum(rate(prometheus_target_sync_length_seconds_sum{job=~"$job",instance=~"$instance"}[5m])) by (scrape_job) * 1e3', '{{scrape_job}}') +
g.queryPanel('sum(rate(prometheus_target_sync_length_seconds_sum{cluster=~"$cluster",job=~"$job",instance=~"$instance"}[5m])) by (cluster, job, scrape_job, instance) * 1e3', '{{cluster}}:{{job}}:{{scrape_job}}') +
{ yaxes: g.yaxes('ms') }
)
.addPanel(
g.panel('Targets') +
g.queryPanel('sum(prometheus_sd_discovered_targets{job=~"$job",instance=~"$instance"})', 'Targets') +
g.queryPanel('sum by (cluster, job, instance) (prometheus_sd_discovered_targets{cluster=~"$cluster", job=~"$job",instance=~"$instance"})', '{{cluster}}:{{job}}') +
g.stack
)
)
@ -48,29 +50,29 @@ local template = grafana.template;
g.row('Retrieval')
.addPanel(
g.panel('Average Scrape Interval Duration') +
g.queryPanel('rate(prometheus_target_interval_length_seconds_sum{job=~"$job",instance=~"$instance"}[5m]) / rate(prometheus_target_interval_length_seconds_count{job=~"$job",instance=~"$instance"}[5m]) * 1e3', '{{interval}} configured') +
g.queryPanel('rate(prometheus_target_interval_length_seconds_sum{cluster=~"$cluster", job=~"$job",instance=~"$instance"}[5m]) / rate(prometheus_target_interval_length_seconds_count{cluster=~"$cluster", job=~"$job",instance=~"$instance"}[5m]) * 1e3', '{{cluster}}:{{job}} {{interval}} configured') +
{ yaxes: g.yaxes('ms') }
)
.addPanel(
g.panel('Scrape failures') +
g.queryPanel([
'sum by (job) (rate(prometheus_target_scrapes_exceeded_body_size_limit_total[1m]))',
'sum by (job) (rate(prometheus_target_scrapes_exceeded_sample_limit_total[1m]))',
'sum by (job) (rate(prometheus_target_scrapes_sample_duplicate_timestamp_total[1m]))',
'sum by (job) (rate(prometheus_target_scrapes_sample_out_of_bounds_total[1m]))',
'sum by (job) (rate(prometheus_target_scrapes_sample_out_of_order_total[1m]))',
'sum by (cluster, job, instance) (rate(prometheus_target_scrapes_exceeded_body_size_limit_total{cluster=~"$cluster",job=~"$job",instance=~"$instance"}[1m]))',
'sum by (cluster, job, instance) (rate(prometheus_target_scrapes_exceeded_sample_limit_total{cluster=~"$cluster",job=~"$job",instance=~"$instance"}[1m]))',
'sum by (cluster, job, instance) (rate(prometheus_target_scrapes_sample_duplicate_timestamp_total{cluster=~"$cluster",job=~"$job",instance=~"$instance"}[1m]))',
'sum by (cluster, job, instance) (rate(prometheus_target_scrapes_sample_out_of_bounds_total{cluster=~"$cluster",job=~"$job",instance=~"$instance"}[1m]))',
'sum by (cluster, job, instance) (rate(prometheus_target_scrapes_sample_out_of_order_total{cluster=~"$cluster",job=~"$job",instance=~"$instance"}[1m]))',
], [
'exceeded body size limit: {{job}}',
'exceeded sample limit: {{job}}',
'duplicate timestamp: {{job}}',
'out of bounds: {{job}}',
'out of order: {{job}}',
'exceeded body size limit: {{cluster}} {{job}}',
'exceeded sample limit: {{cluster}} {{job}}',
'duplicate timestamp: {{cluster}} {{job}}',
'out of bounds: {{cluster}} {{job}}',
'out of order: {{cluster}} {{job}}',
]) +
g.stack
)
.addPanel(
g.panel('Appended Samples') +
g.queryPanel('rate(prometheus_tsdb_head_samples_appended_total{job=~"$job",instance=~"$instance"}[5m])', '{{job}} {{instance}}') +
g.queryPanel('rate(prometheus_tsdb_head_samples_appended_total{cluster=~"$cluster", job=~"$job",instance=~"$instance"}[5m])', '{{cluster}} {{job}} {{instance}}') +
g.stack
)
)
@ -78,12 +80,12 @@ local template = grafana.template;
g.row('Storage')
.addPanel(
g.panel('Head Series') +
g.queryPanel('prometheus_tsdb_head_series{job=~"$job",instance=~"$instance"}', '{{job}} {{instance}} head series') +
g.queryPanel('prometheus_tsdb_head_series{cluster=~"$cluster",job=~"$job",instance=~"$instance"}', '{{cluster}} {{job}} {{instance}} head series') +
g.stack
)
.addPanel(
g.panel('Head Chunks') +
g.queryPanel('prometheus_tsdb_head_chunks{job=~"$job",instance=~"$instance"}', '{{job}} {{instance}} head chunks') +
g.queryPanel('prometheus_tsdb_head_chunks{cluster=~"$cluster",job=~"$job",instance=~"$instance"}', '{{cluster}} {{job}} {{instance}} head chunks') +
g.stack
)
)
@ -91,12 +93,12 @@ local template = grafana.template;
g.row('Query')
.addPanel(
g.panel('Query Rate') +
g.queryPanel('rate(prometheus_engine_query_duration_seconds_count{job=~"$job",instance=~"$instance",slice="inner_eval"}[5m])', '{{job}} {{instance}}') +
g.queryPanel('rate(prometheus_engine_query_duration_seconds_count{cluster=~"$cluster",job=~"$job",instance=~"$instance",slice="inner_eval"}[5m])', '{{cluster}} {{job}} {{instance}}') +
g.stack,
)
.addPanel(
g.panel('Stage Duration') +
g.queryPanel('max by (slice) (prometheus_engine_query_duration_seconds{quantile="0.9",job=~"$job",instance=~"$instance"}) * 1e3', '{{slice}}') +
g.queryPanel('max by (slice) (prometheus_engine_query_duration_seconds{quantile="0.9",cluster=~"$cluster", job=~"$job",instance=~"$instance"}) * 1e3', '{{slice}}') +
{ yaxes: g.yaxes('ms') } +
g.stack,
)