* Alerting: Attempt to retry retryable errors
Retrying has been broken for a good while now (at least since version 9.4) - this change attempts to re-introduce them in their simplest and safest form possible.
I first introduced #79095 to make sure we don't disrupt or put additional load on our customer's data sources with this change in a patch release. Paired with this change, retries can now work as expected.
There's two small differences between how retries work now and how they used to work in legacy alerting.
Retries only occur for valid alert definitions - if we suspect that that error comes from a malformed alert definition we skip retrying.
We have added a constant backoff of 1s in between retries.
---------
Signed-off-by: gotjosh <josue.abreu@gmail.com>
* Alerting: Attempt to retry retryable errors
Currently in a draft state, but this was the minimal diff I could put together to exemplify how could achieve this.
Signed-off-by: gotjosh <josue.abreu@gmail.com>
---------
Signed-off-by: gotjosh <josue.abreu@gmail.com>
* Alerting: post alerts to the remote Alertmanager and fetch them
* fix broken tests
* Alerting: Add Mimir Backend image to devenv (blocks)
* add alerting as code owner for mimir_backend block
* Alerting: Use Mimir image to run integration tests for the remote Alertmanager
* skip integration test when running all tests
* skipping integration test when no Alertmanager URL is provided
* fix bad host for mimir_backend
* remove basic auth testing until we have an nginx image in our CI
* add integration tests for alerts
* fix tests
* change SendCtx -> Send, add context.Context to Send, fix CI
* add reover() for functions from the Prometheus Alertmanager HTTP client that could panic
* add TODO to implement PutAlerts in a way that mimicks what Prometheus does
* fix log format
* add metrics and tracing to state manager
* propagate tracer to state manager
* add scheduler metrics
* fix backtesting
* add test for state metrics
* remove StateUpdateCount
* update docs
* metrics can be null
* add tracer to new tests
This commit adds support for concurrent queries when saving alert
instances to the database. This is an experimental feature in
response to some customers experiencing delays between rule evaluation
and sending alerts to Alertmanager, resulting in flapping. It is
disabled by default.
* Alerting: Repurpose rule testing endpoint to return potential alerts
This feature replaces the existing no-longer in-use grafana ruler testing API endpoint /api/v1/rule/test/grafana. The new endpoint returns a list of potential alerts created by the given alert rule, including built-in + interpolated labels and annotations.
The key priority of this endpoint is that it is intended to be as true as possible to what would be generated by the ruler except that the resulting alerts are not filtered to only Resolved / Firing and ready to be sent.
This means that the endpoint will, among other things:
- Attach static annotations and labels from the rule configuration to the alert instances.
- Attach dynamic annotations from the datasource to the alert instances.
- Attach built-in labels and annotations created by the Grafana Ruler (such as alertname and grafana_folder) to the alert instances.
- Interpolate templated annotations / labels and accept allowed template functions.
Takes a specific code path for data that identifies itself as dataplane instead of "guessing" what the data is.
The data must identify itself by being in the dataplane by having both the following frame metadata properties:
- TypeVersion property that is greater than 0.0
- 'Type' property
The flag is disableSSEDataplane and disables this functionality and uses the old code for all queries regardless.
See https://github.com/grafana/grafana-plugin-sdk-go/blob/main/data/contract_docs/contract.md for dataplane details.
* stop using the scheduler's Update and Delete methods all communication must be via the database
* update scheduler's registry to calculate diff before re-setting the cache
* update fetcher to return the diff generated by registry
* update processTick to update rule eval routine if the rule was updated and it is not going to be evaluated at this tick.
* remove references to the scheduler from api package
* remove unused methods in the scheduler
The `rule_groups_rules` metric is currently defined and computed by `State`.
It makes more sense for this metric to be computed off of the configured rule
set, not based on the rule evaluation state. There could be an edge condition
where a rule does not have a state yet, and so is uncounted.
Additionally, we would like this metric (and others), to have a `rule_group`
label, and this is much easier to achieve if the metric is produced from the
`Scheduler` package.
* Use suggested value for uid
* update the snapshot
* use __expr__
* replace all -100 with __expr__
* update snapshot
* more changes
* revert redundant change
* Use expr.DatasourceUID where it's possible
* generate files
* Add field in alert_rule model, add state to alert_instance model, and state to eval
* Remove paused state from eval package
* Skip paused alert rules in scheduler
* Add migration to add is_paused field to alert_rule table
* Convert to postable alerts only if not normal, pernding, or paused
* Handle paused eval results in state manager
* Add Paused state to eval package
* Add paused alerts logic in scheduler
* Skip alert on scheduler
* Remove paused status from eval package
* Apply suggestions from code review
Co-authored-by: George Robinson <george.robinson@grafana.com>
* Remove state
* Rethink schedule and manager for paused alerts
* Change return to continue
* Remove unused var
* Rethink alert pausing
* Paused alerts storing annotations
* Only add one state transition
* Revert boolean method renaming refactor
* Revert take image refactor
* Make registry errors public
* Revert method extraction for getting a folder title
* Revert variable renaming refactor
* Undo unnecessary changes
* Revert changes in test
* Remove IsPause check in PatchPartiLAlertRule function
* Use SetNormal to set state
* Fix text by returning to old behaviour on alert rule deletion
* Add test in schedule_unit_test.go to test ticks with paused alerts
* Add coment to clarify usage of context.Background()
* Add comment to clarify resetStateByRuleUID method usage
* Move rule get to a more limited scope
* Update pkg/services/ngalert/schedule/schedule.go
Co-authored-by: George Robinson <george.robinson@grafana.com>
* rum gofmt on pkg/services/ngalert/schedule/schedule.go
* Remove defer cancel for context
* Update pkg/services/ngalert/models/instance_test.go
Co-authored-by: Santiago <santiagohernandez.1997@gmail.com>
* Update pkg/services/ngalert/models/testing.go
Co-authored-by: Santiago <santiagohernandez.1997@gmail.com>
* Update pkg/services/ngalert/schedule/schedule_unit_test.go
Co-authored-by: Santiago <santiagohernandez.1997@gmail.com>
* Update pkg/services/ngalert/schedule/schedule_unit_test.go
Co-authored-by: Santiago <santiagohernandez.1997@gmail.com>
* Update pkg/services/ngalert/models/instance_test.go
Co-authored-by: Santiago <santiagohernandez.1997@gmail.com>
* skip scheduler rule state clean up on paused alert rule
* Update pkg/services/ngalert/schedule/schedule.go
Co-authored-by: Santiago <santiagohernandez.1997@gmail.com>
* Fix mock in test
* Add (hopefully) final suggestions
* Use error channel from recordAnnotationsSync to cancel context
* Run make gen-cue
* Place pause alert check in channel update after version check
* Reduce branching un update channel select
* Add if for error and move code inside if in state manager ResetStateByRuleUID
* Add reason to logs
* Update pkg/services/ngalert/schedule/schedule.go
Co-authored-by: George Robinson <george.robinson@grafana.com>
* Do not delete alert rule routine, just exit on eval if is paused
* Reduce branching and create-close a channel to avoid deadlocks
* Separate state deletion and state reset (includes history saving)
* Add current pause state in rule route in scheduler
* Split clearState and bring errCh closer to RecordStatesAsync call
* Change rule to ruleMeta in RecordStatesAsync
* copy state to be able to modify it
* Add timeout to context creation
* Shorten the timeout
* Use resetState is rule is paused and deleteState if rule is not paused
* Remove Empty state reason
* Save every rule change in historian
* Add tests for DeleteStateByRuleUID and ResetStateByRuleUID
* Remove useless line
* Remove outdated comment
Co-authored-by: George Robinson <george.robinson@grafana.com>
Co-authored-by: Santiago <santiagohernandez.1997@gmail.com>
Co-authored-by: Armand Grillet <2117580+armandgrillet@users.noreply.github.com>
* Refactor state and manager to not depend directly on image interface
* Move generic errors to models package
* Move NotAvailableImageService to state as its only references are in state tests
* Move NoopImageService to state package
* Move mock to state package
* Fix linter error
* Fix comment styling
* Fix a couple added references introduced by rebase
* Empty commit to kick build
* extract method processTick
* make processTick return scheduled rules
* move state manager tests to state manager
* update test
* move all tests into one file
* remove unused fields
* delete all stale states in one lock
* do not use touched states to detect stale rely only on LastEvaluationTime maintained correctly
* fix tests to use correct eval time
* delete unused method
* create contextual log context provider
* use contextual provider in scheduler
* init logger in the package
* use context for log context
* use context in state manager
* Touch up log statements, fix casing, add and normalize contexts
* Dedicated logger for dashboard resolver
* Avoid injecting logger to historian
* More minor log touch-ups
* Dedicated logger for state manager
* Use rule context in annotation creator
* Rename base logger and avoid redundant contextual loggers
Prior to this change, all alert instance writes and deletes happened
individually, in their own database transaction. This change batches up
writes or deletes for a given rule's evaluation loop into a single
transaction before applying it.
These new transactions are off by default, guarded by the feature toggle "alertingBigTransactions"
Before:
```
goos: darwin
goarch: arm64
pkg: github.com/grafana/grafana/pkg/services/ngalert/store
BenchmarkAlertInstanceOperations-8 398 2991381 ns/op 1133537 B/op 27703 allocs/op
--- BENCH: BenchmarkAlertInstanceOperations-8
util.go:127: alert definition: {orgID: 1, UID: FovKXiRVzm} with title: "an alert definition FTvFXmRVkz" interval: 60 created
util.go:127: alert definition: {orgID: 1, UID: foDFXmRVkm} with title: "an alert definition fovFXmRVkz" interval: 60 created
util.go:127: alert definition: {orgID: 1, UID: VQvFuigVkm} with title: "an alert definition VwDKXmR4kz" interval: 60 created
PASS
ok github.com/grafana/grafana/pkg/services/ngalert/store 1.619s
```
After:
```
goos: darwin
goarch: arm64
pkg: github.com/grafana/grafana/pkg/services/ngalert/store
BenchmarkAlertInstanceOperations-8 1440 816484 ns/op 352297 B/op 6529 allocs/op
--- BENCH: BenchmarkAlertInstanceOperations-8
util.go:127: alert definition: {orgID: 1, UID: 302r_igVzm} with title: "an alert definition q0h9lmR4zz" interval: 60 created
util.go:127: alert definition: {orgID: 1, UID: 71hrlmR4km} with title: "an alert definition nJ29_mR4zz" interval: 60 created
util.go:127: alert definition: {orgID: 1, UID: Cahr_mR4zm} with title: "an alert definition ja2rlmg4zz" interval: 60 created
PASS
ok github.com/grafana/grafana/pkg/services/ngalert/store 1.383s
```
So we cut time by about 75% and memory allocations by about 60% when
storing and deleting 100 instances.
* Move annotation functionality behind a history persistence interface
* Rename to RecordState
* Fix lint error in import aliasing
* One more import linter error
* Add consumer-side store interface to state manager
* Remove dead dependency
* Delete dead dependency in API struct
* Delete store-layer InstanceStore interface
* Move fake for state's InstanceStore interface to state package
Prior to this change, all alert instance writes and deletes happened
individually, in their own database transaction. This change batches up
writes or deletes for a given rule's evaluation loop into a single
transaction before applying it.
Before:
```
goos: darwin
goarch: arm64
pkg: github.com/grafana/grafana/pkg/services/ngalert/store
BenchmarkAlertInstanceOperations-8 398 2991381 ns/op 1133537 B/op 27703 allocs/op
--- BENCH: BenchmarkAlertInstanceOperations-8
util.go:127: alert definition: {orgID: 1, UID: FovKXiRVzm} with title: "an alert definition FTvFXmRVkz" interval: 60 created
util.go:127: alert definition: {orgID: 1, UID: foDFXmRVkm} with title: "an alert definition fovFXmRVkz" interval: 60 created
util.go:127: alert definition: {orgID: 1, UID: VQvFuigVkm} with title: "an alert definition VwDKXmR4kz" interval: 60 created
PASS
ok github.com/grafana/grafana/pkg/services/ngalert/store 1.619s
```
After:
```
goos: darwin
goarch: arm64
pkg: github.com/grafana/grafana/pkg/services/ngalert/store
BenchmarkAlertInstanceOperations-8 1440 816484 ns/op 352297 B/op 6529 allocs/op
--- BENCH: BenchmarkAlertInstanceOperations-8
util.go:127: alert definition: {orgID: 1, UID: 302r_igVzm} with title: "an alert definition q0h9lmR4zz" interval: 60 created
util.go:127: alert definition: {orgID: 1, UID: 71hrlmR4km} with title: "an alert definition nJ29_mR4zz" interval: 60 created
util.go:127: alert definition: {orgID: 1, UID: Cahr_mR4zm} with title: "an alert definition ja2rlmg4zz" interval: 60 created
PASS
ok github.com/grafana/grafana/pkg/services/ngalert/store 1.383s
```
So we cut time by about 75% and memory allocations by about 60% when
storing and deleting 100 instances.
This change also updates some of our tests so that they run successfully against postgreSQL - we were using random Int64s, but postgres integers, which our tables use, max out at 2^31-1
* Update GetAlertRulesForScheduling to query for folders (if needed)
* Update scheduler's alertRulesRegistry to cache folder titles along with rules
* Update rule eval loop to take folder title from the
* Extract interface RuleStore
* Pre-fetch the rule keys with the version to detect changes, and query the full table only if there are changes.
Removes various custom headers logic sprinkled around in the backend.
It should automatically be applied to outgoing HTTP requests via the
CustomHeadersMiddleware.
This also removes decryption of SecureJSONData to populate custom
headers in ngalert which seemed to have caused a ton of CPU usage.
* move saving the state to state manager when scheduler stops
* move saving state to ProcessEvalResults
* add GetRuleKey to State
* add LogContext to AlertRuleKey
* remove support for bus from scheduler
* rename event to FolderTitleUpdated and fire only if title has changed
* add method to increase version of all rules that belong to a folder
* update ngalert service to subscribe to folder title change event call data store and update scheduler
* add tests
