It could happen for low evaluation intervals and irregular
delays during execution that evaluation time would get
a negative offset. This could result into cumulative
discrepancy between the actual time and evaluation time for rules.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Each group in vmalert starts with an artifical delay to avoid
thundering herd problem. For some groups with high evaluation
intervals, the delay could be significant.
If during this delay user will remove the group from the config
and hot-reload it - vmalert will have to wait until the delay
ends. This results into slow config reloading and UI hang.
The change moves the start-delay logic back to the group's
`start` method. Now, group can immediately exit from the
delay when `group.close()` method is called.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
When group's update() or close() method is called, the group
still need to wait for its current evaluation to finish.
Sometimes, evaluation could take a significant amount of time
which slows configuration update or vmalert's graceful shutdown.
The change interrupts current evaluation in order to speed up
the graceful shutdown or config update procedures.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: speed up state restore procedure on start
Alerts state restore procedure has been changed to become asynchronous.
It doesn't block groups start anymore which significantly improves vmalert's startup time.
Instead, state restore is called by each group in their goroutines after the first rules
evaluation.
While previously state restore attempt was made for all loaded alerting rules,
now it is called only for alerts which became active after the first evaluation.
This reduces the amount of API calls to the configured remote read URL.
This also means that `remoteRead.ignoreRestoreErrors` command-line flag becomes deprecated now
and will have no effect if configured.
See relevant issue https://github.com/VictoriaMetrics/VictoriaMetrics/issues/2608
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* make lint happy
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* Apply suggestions from code review
---------
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>
* vmalert: correctly return error for RW failures
By mistake, in 0989649ad0 the error
for remote write failures weren't return to user.
This change fixes it.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: remove head of line blocking for sending alerts
This change makes sending alerts to notifiers concurrent instead
of sequential. This eliminates head of line blocking, where first
faulty notifier address prevents the rest of notifiers from
receiving notifications.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: make default timeout for sending alerts 10s
Previous value of 1m was too high and was inconsistent
with default timeout defined for notifiers via
configuration file.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: linter checks fix
Signed-off-by: hagen1778 <roman@victoriametrics.com>
vmalert: support `limit` param in groups definition
`limit` param limits number of time series samples produced by a single rule
during execution.
On reaching the limit rule will return an err.
Signed-off-by: lihaowei <haoweili35@gmail.com>
The new metric shows the configured evaluation interval per group.
Metric updates its value when group's interval is changed during
hot reload.
The new metric can be used to estimate how close group
is to start missing evaluation rounds. The following query
will show the % of used time by the group to evaluate all rules
before the next round:
```
(max(vmalert_iteration_duration_seconds{quantile="0.99"}) / vmalert_iteration_interval_seconds) * 100
```
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/2618
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Rules executor within group tracks series sent to remote write
in order to mark them as stale if they had disappeared in next
evaluation round.
The executor uses rules ID as a key to identifies series which belong to rule.
On config reload, executor remains active but the set of rules could change.
Hence, we need to properly cleanup the tracker for rules which has been disappeared
on config reload.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: calculate time for firing alert based on the given timestamp
Previously, current time was used for checking the `firing` threshold.
This is not correct, since alerts are evaluated at specific timestamps.
Hence, this specific timestamp supposed to be used in the calculation.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: properly calculate evaluation timestamp for rules
Timestamp for rules evaluation should be calculated after
the artifical delay for groups start. Otherwise, evaluation
timestamp can fall back too far in time.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Executor recently gain field for storing previously sent series.
Since the same executor object can be used in multiple goroutines,
the access to this field should be serialized.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: split alert's `Start` field into `ActiveAt` and `Start`
The `ActiveAt` field identifies when alert becomes active for rules
with `for > 0`. Previously, this value was stored in field `Start`.
The field `Start` now identifies the moment alert became `FIRING`.
The split is needed in order to distinguish these two moments
in the API responses for alerts.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: support specific moment of time for rules evaluation
The Querier interface was extended to accept a new argument
used as a timestamp at which evaluation should be made.
It is needed to align rules execution time within the group.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: mark disappeared series as stale
Series generated by alerting rules, which were sent to remote write
now will be marked as stale if they will disappear on the next
evaluation. This would make ALERTS and ALERTS_FOR_TIME series
more precise.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* wip
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: evaluate rules at fixed timestamp
Before, time at which rules were evaluated was calculated
right before rule execution. The change makes sure
that timestamp is calculated only once per evalution round
and all rules are using the same timestamp.
It also updates the logic of resending of already resolved
alert notification.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: allow overridin `alertname` label value if it is present in response
Previously, `alertname` was always equal to the Alerting Rule name. Now,
its value can be overriden if series in response containt the different value
for this label.
The change is needed for improving compatibility with Prometheus.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: align rules evaluation in time
Now, evaluation timestamp for rules evaluates as if
there was no delay in rules evaluation. It means, that
rules will be evaluated at fixed timestamps+group_interval.
This way provides more consistent evaluation results and
improves compatibility with Prometheus,
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: add metric for missed iterations
New metric `vmalert_iteration_missed_total` will show
whether rules evaluation round was missed.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: reduce delay before the initial rule evaluation in group
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: rollback alertname override
According to the spec:
```
The alert name from the alerting rule (HighRequestLatency from the example above) MUST be added to the labels of the alert with the label name as alertname. It MUST override any existing alertname label.
```
https://github.com/prometheus/compliance/blob/main/alert_generator/specification.md#step-3
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: throw err immediately on dedup detection
```
The execution of an alerting rule MUST error out immediately and MUST NOT send any alerts
or add samples to samples receiver if there is more than one alert with the same labels
```
https://github.com/prometheus/compliance/blob/main/alert_generator/specification.md#step-4
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: cleanup
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: use strings builder to reduce allocs
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: add support of `resendDelay` flag for alerts
Co-authored-by: dmitryk-dk <dmitry.kozlov@brightlocal.com>
Co-authored-by: hagen1778 <roman@victoriametrics.com>
* vmalert: add support for `sortByLabel` template function
* vmalert: update API according to Prometheus conformance program
The changes to the API, field names and URL path has been made
according to the Prometheus specification for `alert_generator`
https://github.com/prometheus/compliance/blob/main/alert_generator/specification.md
* vmalert: fix the timestamp of the evaluated rules
The timestamp used for alert's `EndsAt` was calculated
before sending the notification. While the correct way
is to use the timestamp taken right before rules evaluation.
* vmalert: add `-datasource.queryTimeAlignment` flag
The flag is supposed to provide ability to disable `time`
param alignment when executing rules. By default, this flag
is enabled, so it remains backward compatible.
The flag was introduced to achieve better compatibility
with Prometheus behaviour according to https://github.com/prometheus/compliance/blob/main/alert_generator/specification.md
Signed-off-by: hagen1778 <roman@victoriametrics.com>
vmalert: support configuration file for notifiers
* vmalert notifiers now can be configured via file
see https://docs.victoriametrics.com/vmalert.html#notifier-configuration-file
* add support of Consul service discovery for notifiers config
see https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1947
* add UI section for currently loaded/discovered notifiers
* deprecate `-rule.configCheckInterval` in favour of `-configCheckInterval`
* add ability to suppress logs for duplicated targets for notifiers discovery
* change behaviour of `vmalert_alerts_send_errors_total` - it now accounts
for failed alerts, not HTTP calls.
* vmalert: introduce additional HTTP URL params per-group configuration
The new group field `params` allows to configure custom HTTP URL params
per each group. These params will be applied to every request before
executing rule's expression. Hot config reload is also supported.
Field `extra_filter_labels` was deprecated in favour of `params` field.
vmalert will print deprecation log message if config file contains
the deprecated field.
`params` fields are supported by both Prometheus and Graphite datasource types.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: provide more examples for `params` field
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: set higher priority for `params` setting
If there would be a conflict between URL params set in `datasource.url` flag
and params in group definition the latter will have higher priority.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Previously only the lower part of 64-bit hash was used for calculating the offset.
This may give uneven distribution in some cases. So let's use all the available 64 bits from the hash
for calculating the offset.
Prometheus allows to have groups with no rules, so we should support
it in vmalert as well for compatibility reasons.
It is also allowed to hot-reload empty groups by adding or removing rules.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Commit fixes potential race condition when group update
and generating of ID() happens simultaneously.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
`omitempty` tag resulted into skipping this param on marshaling,
which was used as a checksum for groups configuration. Since on
config reload checksums are compared before applying changes,
any change to `interval` only didn't trigger config reload.
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1641
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* vmalert: add flag to limit the max value for auto-resovle duration for alerts
The new flag `rule.maxResolveDuration` suppose to limit max value for
alert.End param, which is used by notifiers like Alertmanager for alerts auto resolve.
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1586
* vmalert: remove `vmalert_execution_duration_seconds` metric
The summary for `vmalert_execution_duration_seconds` metric gives no additional
value comparing to `vmalert_iteration_duration_seconds` metric.
* vmalert: update config reload success metric properly
Previously, if there was unsuccessfull attempt to reload config and then
rollback to previous version - the metric remained set to 0.
* vmalert: add Grafana dashboard to overview application metrics
* docker: include vmalert target into list for scraping
* vmalert: extend notifier metrics with addr label
The change adds an `addr` label to metrics for alerts_sent and alerts_send_errors
to identify which exact address is having issues.
The according change was made to vmalert dashboard.
* vmalert: update documentation and docker environment for vmalert's dashboard
Mention Grafana's dashboard in vmalert's README in a new section #Monitoring.
Update docker-compose env to automatically add vmalert's dashboard.
Update docker-compose README with additional info about services.
* vmalert: support rules backfilling (aka `replay`)
vmalert can `replay` configured rules in the past
and backfill results via remote write protocol.
It supports MetricsQL/PromQL storage as data source,
and can backfill data to remote write compatible
storage.
Supports recording and alerting rules `replay`. See more
details in README.
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/836
* vmalert: review fixes
* vmalert: readme fixes
The new setting `extra_filter_labels` may be assigned to group.
If it is, then all rules within a group will automatically filter
for configured labels. The feature is well-described here
https://docs.victoriametrics.com#prometheus-querying-api-enhancements
New setting is compatible only with VM datasource.
* changes vmalert Querier with per rule querier
it allows to changes some parametrs based on rule setting
for instance - alert type, tenant for cluster version or event endpoint url.
* init implementation for graphite alerts
* adds graphite support for vmalert
* small fix
* changes vmalert graphite api with type
* updates tests
* small fix
* fixes graphite parse
* Fixes graphite from time
On config reload event `vmalert` reloads configuration for every group. While
it works for simple configurations, the more complex and heavy installations may
suffer from frequent config reloads.
The change introduces the `checksum` field for every group and is set to md5 hash
of yaml configuration. The checksum will change if on any change to group
definition like rules order or annotation change. Comparing the `checksum` field
on config reload event helps to detect if group should be updated.
The groups update is now done concurrently, so reload duration will be limited by
the slowest group now.
Partially solves #691 by improving config reload speed.
* feat: spread load of rule evaluation by group when starting new groups
* review: reduce the resulting diff.
* Update app/vmalert/group.go
Co-authored-by: Roman Khavronenko <hagen1778@gmail.com>
Co-authored-by: Aliaksandr Valialkin <valyala@gmail.com>
Co-authored-by: Roman Khavronenko <hagen1778@gmail.com>
* app/vmalert: extend metrics set exported by `vmalert` #573
New metrics were added to improve observability:
+ vmalert_alerts_pending{alertname, group} - number of pending alerts per group
per alert;
+ vmalert_alerts_acitve{alertname, group} - number of active alerts per group
per alert;
+ vmalert_alerts_error{alertname, group} - is 1 if alertname ended up with error
during prev execution, is 0 if no errors happened;
+ vmalert_recording_rules_error{recording, group} - is 1 if recording rule
ended up with error during prev execution, is 0 if no errors happened;
* vmalert_iteration_total{group, file} - now contains group and file name labels.
This should improve control over specific groups;
* vmalert_iteration_duration_seconds{group, file} - now contains group and file name labels. This should improve control over specific groups;
Some collisions for alerts and recording rules are possible, because neither
group name nor alert/recording rule name are unique for compatibility reasons.
Commit contains list of TODOs for Unregistering metrics since groups and rules
are ephemeral and could be removed without application restart. In order to
unlock Unregistering feature corresponding PR was filed - https://github.com/VictoriaMetrics/metrics/pull/13
* app/vmalert: extend metrics set exported by `vmalert` #573
The changes are following:
* add an ID label to rules metrics, since `name` collisions within one group is
a common case - see the k8s example alerts;
* supports metrics unregistering on rule updates. Consider the case when one rule
was added or removed from the group, or the whole group was added or removed.
The change depends on https://github.com/VictoriaMetrics/metrics/pull/16
where race condition for Unregister method was fixed.
`external.label` flag supposed to help to distinguish alert or recording rules
source in situations when more than one `vmalert` runs for the same datasource
or AlertManager.
* app/vmalert: add retries to remotewrite
Remotewrite pkg now does limited number of retries if write request failed.
This suppose to make vmalert state persisting more reliable.
New metrics were added to remotewrite in order to track rows/bytes sent/dropped.
defaultFlushInterval was increased from 1s to 5s for sanity reasons.
* fix
* wip
* wip
* wip
* fix bits alignment bug for 32-bit systems
* fix mistakenly dropped field
* app/vmalert: support multiple notifier urls (#584)
User now can set multiple notifier URLs in the same fashion
as for other vmutils (e.g. vmagent). The same is correct for
TLS setting for every configured URL. Alerts sending is done
in sequential way for respecting the specified URLs order.
* app/vmalert: add basicAuth support for notifier client (#585)
The change adds possibility to set basicAuth creds for notifier
client in the same fasion as for remote write/read and datasource.
The feature allows to speed up group rules execution by
executing them concurrently.
Change also contains README changes to reflect configuration
details.
* vmalert: Add recording rules support.
Recording rules support required additional service refactoring since
it wasn't planned to support them from the very beginning. The list
of changes is following:
* new entity RecordingRule was added for writing results of MetricsQL
expressions into remote storage;
* interface Rule now unites both recording and alerting rules;
* configuration parser was moved to separate package and now performs
more strict validation;
* new endpoint for listing all groups and rules in json format was added;
* evaluation interval may be set to every particular group;
* vmalert: uncomment tests
* vmalert: rm outdated TODO
* vmalert: fix typos in README
Before the change we were sending notifications to notifier
if following conditions are met:
* alert is in Fire state
* alert is in Inactive state
We were sending Inactive notifications to resolve alert ASAP.
Unfortunately, we were sending resolves for Pending alerts that become
Inactive, which is wrong.
In this change we delete alert from the active list if
it was Pending and become Inactive. In this way we now
have Inactive alerts only if they were in state Fire before.
See test change for example.
During group's update rules deletion was causing slice
mutations while slice index was assumed to be unchanged.
This caused "slice bounds out of range" errors when multiple
rules were deleted sequentially.
The change introduces new entity `manager` which replaces
`watchdog`, decouples requestHandler and groups. Manager
supposed to control life cycle of groups, rules and
config reloads.
Groups export an ID method which returns a hash
from filename and group name. ID supposed to be unique
identifier across all loaded groups.
Some tests were added to improve coverage.
Bug with wrong annotation value if $value is used in
templates after metrics being restored fixed.
Notifier interface was extended to accept context.
New set of metrics was introduced for config reload.
