* fix: use dsIndexProvider cache on migrations
* chore: use same comment as before
* feat: org-aware TTL cache for schemaversion migration and warmup for single tenant
* chore: use LRU cache
* chore: change DefaultCacheTTL to 1 minute
* chore: address copilot reviews
* chore: use expirable cache
* chore: remove unused import
* fix(dashboard): prevent memory leak in CUE validation by using fresh contexts
Fixes#114344
The CUE validation was reusing a single cue.Context across all
validations, which caused unbounded memory growth due to CUE's internal
caching of intermediate computation results (disjunctions, unifications,
etc.).
Root Cause:
- A single cue.Context was created and reused via getValidator()
- Each validation added entries to the context's internal caches
- These caches grew unboundedly over time
- Memory could not be garbage collected because the context held references
Solution:
- Store the schema source string instead of a compiled cue.Value
- Create a fresh cuecontext.New() for each validation
- This allows the context and its caches to be garbage collected after
each validation completes
Performance Impact:
- ~2x slower due to schema recompilation per validation
- Acceptable trade-off to prevent memory leaks
- Memory usage stays bounded instead of growing unboundedly
* fix(dashboard): use periodic context recreation to prevent CUE memory leaks
Replace fresh context creation with periodic context recreation approach.
The context is reused for up to 100 validations, then recreated to allow
garbage collection of cached values while maintaining good performance.
This balances performance (only 19% slower than leaky approach) with memory
safety (stable at ~5 MB vs ~2 GB leak).
See https://github.com/grafana/grafana/issues/114344#issuecomment-3605562491
* refactor(cuevalidator): simplify to use mutex instead of atomic counter
Since CUE is not thread-safe, we need the mutex for the entire validation
operation anyway. Using a regular int counter protected by the mutex is
simpler and cleaner than using atomic operations.
* Reapply "v1 to v2 conversion: Fix issue when Grafana datasource wouldn't be resolved correctly (#114555)"
This reverts commit db9cff8e2d.
* Update input and fix the failing test
* Update docs
* Remove 406 response since now it is converted
* fix linter
---------
Co-authored-by: Stephanie Hingtgen <stephanie.hingtgen@grafana.com>
* fix: add missing transformation for scenes -> save model v2
* fix: link placement transformation on the backend between schemas
* fix: update the openapi spec in the tests
* tes: add tests for `transformSceneToSaveModelSchemaV2`
* tests: extend conversion_test.go to cover link placements
* add viridis and others, allow passing interpolator directly to FieldColorSchemeMode
* fix bug
* be more defensive when getting the fieldcolor
* backend changes
* add d3-scale-chromatic to list of esModules
* Add library panel repeat options to v2 schema during conversion
* use any instead of interface{}
* change to common.Unstructured instead of byte[] for model field
* Fix the tests and let the library panel behavior fetch repeat options in public and scripted dashboards
* fix library panel differences between backend and frontend conversion
* feat: add a placement property to annotations model v2
* chore: update scenes to `v6.42.1`
* chore: run `make gen-apps`
* fix: cater for cases when there is no data layer
* chore: swagger clean
* chore: update api clients
* fix: correct type guard
* fix: display control labels in the default renderer as well for DashboardDataLayerSet
* StateTimeline: E2E tests for basic rendering, no data, and tooltip interactions
* Add tests for StateTimelineTooltip sub-component
* Remove unused import from state-timeline.spec.ts
* updates from PR review
* NewGauge: Fix segmented gauge with min value != 0
* update gdev dashboard to exercise case
* update gdev dashboard with panel to exercise
* update
---------
Co-authored-by: Paul Marbach <paul.marbach@grafana.com>
Dashboard Migration: Fix v38 table panel overrides migration when defaults.custom is missing
The v38 migration was incorrectly skipping field config overrides processing
when fieldConfig.defaults.custom didn't exist. This caused custom.displayMode
properties in overrides to not be migrated to the new custom.cellOptions format.
The fix ensures migrateOverrides is always called for table panels, regardless
of whether defaults.custom exists.
Added comprehensive unit test covering this edge case to prevent regression.
Issue discovered through comprehensive migration testing infrastructure.
Fix v16 grid position calculation for fractional spans
Match frontend span-to-width conversion logic by flooring span first,
then multiplying by width factor, instead of multiplying first then flooring.
This fixes dashboard layout inconsistencies where panels with fractional
spans (e.g. 5.929860088365242) would have different widths between
backend and frontend migration paths.
- Backend old: Math.floor(5.93 * 2) = Math.floor(11.86) = 11
- Backend new: Math.floor(5.93) * 2 = 5 * 2 = 10 (matches frontend)
Includes comprehensive unit test to prevent regression.
