During background downsampling, rate(vm_deduplicated_samples_total{type="merge"}) could be much bigger than
rate(vm_rows_added_to_storage_total) and it could last quite some time,
which causes negative values of Storage full ETA and confuses users, see playground.
Instead of trying to get more accurate results during downsampling, I think it's ok to ignore
vm_deduplicated_samples_total at all, it's more reasonable to see Storage full ETA increase after downsampling.
---------
Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>
This is follow-up after
75196d7234
It updates some of the alerting rules to remove unnecessary aggregations.
It keeps aggregations for expressions which are using multiple time series
filters to make sure their label will match.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
(cherry picked from commit 8fb68152e6)
The `ConcurrentFlushesHitTheLimit` could be related to components like
vminsert, vmstorage, vm-single-node and vmagent. Moving this alert
to the `health` section of alerts will be benefitial for all components
and will remove the duplicates from single/cluster alerts.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Previously the -maxConcurrentInserts was limiting the number of established client connections,
which write data to VictoriaMetrics. Some of these connections could be idle.
Such connections do not consume big amounts of CPU and RAM, so there is a little sense in limiting
the number of such connections. So now the -maxConcurrentInserts command-line option
limits the number of concurrently executed insert requests, not including idle connections.
It is recommended removing -maxConcurrentInserts command-line option, since the default value
for this option should work good for most cases.
Alert `RequestErrorsToAPI` could be permanently triggered due to
mistakes in clients configuration. However, such requests are unlikely
to cause VM health state change. So there is no need in displaying
this alert because there will be no correlation caused by it.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Signed-off-by: hagen1778 <roman@victoriametrics.com>
The change list is the following:
* bump Grafana version to 9.2.6;
* replace old "Graph" panel with "TimeSeries" panel;
* show % usage of Mem and CPU additionally to of absolute values;
* `Caches` row was removed. All needed info for caches is now part of `Troubleshooting`;
* add Annotations for Alert triggers. Not all alerts are supposed to be displayed
on the dashboard, but only those with label `show_at: dashboard`.
See `alerts.yml` change.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* deployment/docker: move cluster compose env to master branch
The change supposed to simplify the process of maintaining for
single/cluster docker-compose envs, alerts, dashboards. It also
supposes to reduce confusion for users when looking for cluster
related alerts/configs.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* deployment/docker: move cluster compose env to master branch
Review updates.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Signed-off-by: hagen1778 <roman@victoriametrics.com>
The new metric `vmagent_remotewrite_queues` exports a static value of
number of configured remote write queus. This metric is useful to
calculate total saturation per each configured URL with given number
of queues. See corresponding changes to vmagent alerts and dashboard.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
The negative value for ETA can happen when deduplication is enabled
and `rate` over `vm_deduplicated_samples_total` becomes bigger
than actual ingestion rate.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
alerts: filter out non error log messages for `TooManyLogs`
Info and Warn error levels aren't always a result of malfunctioning
or faulty state. So we filter them out.
* dashboards: add `CPU percentage` panel for cluster dashboards
The new panel `CPU percentage` was added instead if adding a limit
to the existing `CPU` panel because dasbhoard may display big number
of components each with own limits. The separate panel should provide
a clear display of CPU load.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* dashboards: sync vmagent and vmalert changes from single version
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* docker: remove unsupported param from vmagent config
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* alerts: add `TooHighCPUUsage` alert for all VM components
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* dashboards/vmagent: shuffle panels for better visibility
More important error/dropped panels were moved higher on the main row.
Network usage panel moved to Resource usage row.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* dashboards/vmagent: add Troubleshooting row to show top 5 instances/jobs by churn rate
New panels are supposed to show top 5 jobs or targets which generate the most
of the churn rate. They were placed into a new row "Troubleshooting".
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* dashboards/vmagent: add panels for showing persistent queue saturation
New panels were added to Torubleshooting row to show the persistent queue
saturation. The corresponding alerts were added and linked to these
panels as well.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* dashboards/vmagent: add alert "RejectedRemoteWriteDataBlocksAreDropped"
New alert suppose to send a notification when vmagent starts to drop
data blocks rejected by configured remote write destiantion.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* dasbhoard: replace `null` datasources
null datasource value may confuse Grafana and make it drop panel query in some
versions.
* docker: bump grafana image version
* dashboards: add URL variable selector to vmagent dashboard
* dashboards: add new panel `Remote write connection saturation` to vmagent dashboard
* alerts: add new alert for `Remote write connection saturation` panel of vmagent dashboard
* dashboards: add "Logging rate" panel to vmagent dashboard
Original query can't be executed via PromQL which results in error
if expression is evaluated by Prometheus. The new expression is
compatible with both engines.
