* (WIP) Alerting: Use the forked Alertmanager for remote secondary mode
* fall back to using internal AM in case of error
* remove TODOs, clean up .ini file, add orgId as part of remote AM config struct
* log warnings and errors, fall back to remoteSecondary, fall back to internal AM only
* extract logic to decide remote Alertmanager mode to a separate function, switch on mode
* tests
* make linter happy
* remove func to decide remote Alertmanager mode
* refactor factory function and options
* add default case to switch statement
* remove ineffectual assignment
This PR replaces the vendored models in the migration with their equivalent ngalert models. It also replaces the raw SQL selects and inserts with service calls.
It also fills in some gaps in the testing suite around:
- Migration of alert rules: verifying that the actual data model (queries, conditions) are correct 9a7cfa9
- Secure settings migration: verifying that secure fields remain encrypted for all available notifiers and certain fields migrate from plain text to encrypted secure settings correctly e7d3993
Replacing the checks for custom dashboard ACLs will be replaced in a separate targeted PR as it will be complex enough alone.
This change marks tests in the `sender` package that use an external
process as integration tests instead of unit tests. This speeds up the
package's unit tests by about 20 seconds.
This change also reduces the number of alert instances in the `store`
package's bulk write integration test from 20_000 to 10_000. This is
still enough to exercise the bulk-write code but speeds up the package
tests from about 250s to 130s.
Put together, integration tests go to about 160s while also speeding up
unit tests by 20s.
* chore: add alias for InitTestDB and Session
Adds an alias for the sqlstore InitTestDB and Session, and updates tests using these to reduce dependencies on the sqlstore.Store.
* next pass of removing sqlstore imports
* last little bit
* remove mockstore where possible
* (WIP) switch to fork AM, first implementation of the API, generate spec
* get receivers avoiding race conditions
* use latest version of our forked AM, tests
* make linter happy, delete TODO comment
* update number of expected paths to += 2
* delete unused endpoint code, code review comments, tests
* Update pkg/services/ngalert/notifier/alertmanager.go
Co-authored-by: Matthew Jacobson <matthew.jacobson@grafana.com>
* remove call to fmt.Println
* clear naming for fields
* shorter variable names in GetReceivers
Co-authored-by: Matthew Jacobson <matthew.jacobson@grafana.com>
* chore: replace artisnal FakeDashboardService with generated mock
Maintaining a handcrafted FakeDashboardService is not sustainable now that we are in the process of moving the dashboard-related functions out of sqlstore.
* sqlstore: finish removing Find and SearchDashboards
Find and SearchDashboards were previously copied into the dashboard service. This commit completes that work, removing Find and SearchDashboards from the sqlstore and updating callers to use the dashboard service.
* dashboards: remove SearchDashboards from Store interface
SearchDashboards is a wrapper around FindDashboard that transforms the results, so it's been moved out of the Store entirely and the functionality moved into the Dashboard Service's search implementation.
The database tests depended heavily on the transformation, so I added testSearchDashboards, a copy of search dashboards, instead of (heavily) refactoring all the tests.
This test of silence cleanup was flaky because of its use of real wall
time. In CI environments with slow execution, delays could cause the
test to fail. This change mitigates the problem by increasing the end time of
silences in the test.
After Prometheus merges this PR: https://github.com/prometheus/alertmanager/pull/2867
we can make the test fully deterministic by using a fake clock.
Fixes#47470
Signed-off-by: Joe Blubaugh <joe.blubaugh@grafana.com>
Expired silences older than the retention period were not being cleaned up. The root problem was that notifier.Alertmanager overrides the Prometheus alert manager's silence maintenance function and was not calling Silences.GC() in the overriden function.
The directory created by `T.TempDir` is automatically removed when the
test and all its subtests complete.
Reference: https://pkg.go.dev/testing#T.TempDir
Signed-off-by: Eng Zer Jun <engzerjun@gmail.com>
* pass notification service down to the notifiers
* add ns to all notifiers
* remove bus from ngalert notifiers
* use smaller interfaces for notificationservice
* attempt to fix the tests
* remove unused struct field
* simplify notification service mock
* trying to resolve issues in the tests
* make linter happy
* make linter even happier
* linter, you are annoying
* Create API test for overwriting invalid alertmanager config
* Avoid requiring alertmanager readiness for config changes
* AlertmanagerSrv depends on functionality rather than concrete types
* Add test for non-ready alertmanagers
* Additional cleanup and polish
* Back out previous integration test changes
* Refactor of tests incorrectly caused a test to become redundant
* Use pre-existing fake secret service
* Drop unused interface
* Test against concrete MultiOrgAlertmanager re-using fake infra from other tests
* Fix linter error
* Empty commit to rerun checks
* Use secrets service in pluginproxy
* Use secrets service in pluginxontext
* Use secrets service in pluginsettings
* Use secrets service in provisioning
* Use secrets service in authinfoservice
* Use secrets service in api
* Use secrets service in sqlstore
* Use secrets service in dashboardshapshots
* Use secrets service in tsdb
* Use secrets service in datasources
* Use secrets service in alerting
* Use secrets service in ngalert
* Break cyclic dependancy
* Refactor service
* Break cyclic dependancy
* Add FakeSecretsStore
* Setup Secrets Service in sqlstore
* Fix
* Continue secrets service refactoring
* Fix cyclic dependancy in sqlstore tests
* Fix secrets service references
* Fix linter errors
* Add fake secrets service for tests
* Refactor SetupTestSecretsService
* Update setting up secret service in tests
* Fix missing secrets service in multiorg_alertmanager_test
* Use fake db in tests and sort imports
* Use fake db in datasources tests
* Fix more tests
* Fix linter issues
* Attempt to fix plugin proxy tests
* Pass secrets service to getPluginProxiedRequest in pluginproxy tests
* Fix pluginproxy tests
* Revert using secrets service in alerting and provisioning
* Update decryptFn in alerting migration
* Rename defaultProvider to currentProvider
* Use fake secrets service in alert channels tests
* Refactor secrets service test helper
* Update setting up secrets service in tests
* Revert alerting changes in api
* Add comments
* Remove secrets service from background services
* Convert global encryption functions into vars
* Revert "Convert global encryption functions into vars"
This reverts commit 498eb19859.
* Add feature toggle for envelope encryption
* Rename toggle
Co-authored-by: Emil Tullstedt <emil.tullstedt@grafana.com>
Co-authored-by: Joan López de la Franca Beltran <joanjan14@gmail.com>
* Encryption: Add support to encrypt/decrypt sjd
* Add datasources.Service as a proxy to datasources db operations
* Encrypt ds.SecureJsonData before calling SQLStore
* Move ds cache code into ds service
* Fix tlsmanager tests
* Fix pluginproxy tests
* Remove some securejsondata.GetEncryptedJsonData usages
* Add pluginsettings.Service as a proxy for plugin settings db operations
* Add AlertNotificationService as a proxy for alert notification db operations
* Remove some securejsondata.GetEncryptedJsonData usages
* Remove more securejsondata.GetEncryptedJsonData usages
* Fix lint errors
* Minor fixes
* Remove encryption global functions usages from ngalert
* Fix lint errors
* Minor fixes
* Minor fixes
* Remove securejsondata.DecryptedValue usage
* Refactor the refactor
* Remove securejsondata.DecryptedValue usage
* Move securejsondata to migrations package
* Move securejsondata to migrations package
* Minor fix
* Fix integration test
* Fix integration tests
* Undo undesired changes
* Fix tests
* Add context.Context into encryption methods
* Fix tests
* Fix tests
* Fix tests
* Trigger CI
* Fix test
* Add names to params of encryption service interface
* Remove bus from CacheServiceImpl
* Add logging
* Add keys to logger
Co-authored-by: Emil Tullstedt <emil.tullstedt@grafana.com>
* Add missing key to logger
Co-authored-by: Emil Tullstedt <emil.tullstedt@grafana.com>
* Undo changes in markdown files
* Fix formatting
* Add context to secrets service
* Rename decryptSecureJsonData to decryptSecureJsonDataFn
* Name args in GetDecryptedValueFn
* Add template back to NewAlertmanagerNotifier
* Copy GetDecryptedValueFn to ngalert
* Add logging to pluginsettings
* Fix pluginsettings test
Co-authored-by: Tania B <yalyna.ts@gmail.com>
Co-authored-by: Emil Tullstedt <emil.tullstedt@grafana.com>
Remove validation for labels to be accepted in the Alertmanager, This helps with datasources that produce non-compatible labels.
Adds an "object_matchers" to alert manager routers so we can support labels names with extended characters beyond prometheus/openmetrics. It only does this for the internal Grafana managed Alert Manager.
This requires a change to alert manager, so for now we use grafana/alertmanager which is a slight fork, with the intention of going back to upstream.
The frontend handles the migration of "matchers" -> "object_matchers" when the route is edited and saved. Once this is done, downgrades will not work old versions will not recognize the "object_matchers".
Co-authored-by: Kyle Brandt <kyle@grafana.com>
Co-authored-by: Nathan Rodman <nathanrodman@gmail.com>
* Add method GetAllLatestAlertmanagerConfiguration to DBStore
* add method ApplyConfig to AlertManager
* update multiorg alert manager to load all alertmanager configs at once
* Alerting: Refactor & fix unified alerting metrics structure
Fixes and refactors the metrics structure we have for the ngalert service. Now, each component has its own metric struct that includes the JUST the metrics it uses. Additionally, I have fixed the configuration metrics and added new metrics to determine if we have discovered and started all the necessary configurations of an instance.
This allows us to alert on `grafana_alerting_discovered_configurations - grafana_alerting_active_configurations != 0` to know whether an alertmanager instance did not start successfully.
* Alerting: Persist notification log and silences to the database
This removes the dependency of having persistent disk to run grafana alerting. Instead of regularly flushing the notification log and silences to disk we now flush the binary content of those files to the database encoded as a base64 string.
Introduces org-level isolation for the Alertmanager and its components.
Silences, Alerts and Contact points are not separated by org and are not shared between them.
Co-authored with @davidmparrott and @papagian
* Alerting: modify table and accessors to limit org access appropriately
* Update migration to create multiple Alertmanager configs
* Apply suggestions from code review
Co-authored-by: gotjosh <josue@grafana.com>
* replace mg.ClearMigrationEntry()
mg.ClearMigrationEntry() would create a new session.
This commit introduces a new migration for clearing an entry from migration log for replacing mg.ClearMigrationEntry() so that all dashboard alert migration operations will run inside the same transaction.
It adds also `SkipMigrationLog()` in Migrator interface for skipping adding an entry in the migration_log.
Co-authored-by: gotjosh <josue@grafana.com>
* [Alerting]: forbid viewers for updating rules if viewers can edit
check for CanSave instead of CanEdit
* Clear ngalert tables when deleting the folder
* Apply suggestions from code review
* Log failure to check save permission
Co-authored-by: gotjosh <josue@grafana.com>
* Alerting: Allow the notifier to log
The notifier upstream code uses go-kit as its logging library. The
grafana specific logger is not compatible with this API. In this PR, I
have created a wrapper that implements io.Writer to make them
compatible.
* Alerting: Use a default configuration and periodically poll for new ones
Use a default configuration to make sure we always start the grafana
instance. Then, regularly poll for new ones.
I've also made sure that failures to apply configuration do not stop the
Grafana server but instead keep polling until it is a success.
* Alerting: Fetch configuration from the database and run a notification
instance
Co-Authored-By: Ganesh Vernekar <15064823+codesome@users.noreply.github.com>
