Repository navigation
Release v19.116.0
Summary
Observer reliability and recovery, client failover, Workbench improvements, projection and PII correctness fixes, and a large documentation pass.
Added
- Observers can be given their own retry policy for failed partitions through host configuration (#494)
- Observers report when their running state last changed, exposed through the gRPC contract, the client and the Workbench (#3732)
- Reactors registered explicitly can take their source event store from configuration (#3951)
- Event sequence reads can be cancelled, including tail reads
EventScenariocan redact events (#4749)- Event writes use the injected clock for occurrence and receipt times (#4930)
- The Workbench can ignore and restore recommendations, replay a single event source, show quarantine and failed-partition columns, and inspect the exact events of failed partition attempts (#5310) (part of #776) (part of #735)
- Model-bound projections support stable identities (part of #4757)
- Read models can be folded through an existing test store in tests (part of #5464)
- Explicit event reads can opt in to erased-property metadata, so erased numeric and boolean values can be told apart from real zero and false values (part of #4620)
- Constraint index repair retries, time budgets, cadence, poison threshold and memory caps can be configured in
chronicle.jsonor environment variables (#4906) - MVC actions and Razor Page handlers can report failed unit-of-work commits before writing results by enabling
ReportFailuresBeforeResponseStarts; minimal APIs opt in withWithUnitOfWork()(#5339)
Changed
- Reading from the System event store now requires administrator access; API registration against it is refused for every caller
- MongoDB indexes can be pre-provisioned and verified before deployment with schema commands; runtime index creation is unchanged (part of #5284)
- Documentation describes the existing retention of replay revert containers and that orphaned replay collections are not cleaned up automatically (part of #5268)
- MongoDB bulk redaction no longer overwrites already-redacted events, and revisions are no longer appended to events that are being redacted concurrently (part of #3920)
- Observer hot paths allocate less per batch (#5471)
- Observer list streams cache running-state timestamps per subscription instead of re-reading them on every emission (#5479)
- Kernel tracing no longer records health checks and keep-alives (#4614)
- Test scenarios discover reducers once per store and code analyzers bind once per invocation (#5464) (#5467)
- Code generation runs only when its inputs change (#5466)
- Wire-compatibility checks no longer compare against releases newer than the release being cut, or the newest release on a maintenance branch (#4312)
- Local Chronicle images are tagged with local-only names so they no longer shadow the published image (#3795)
- Workbench adopts the shared checkbox list filter (#5441)
- Documentation consolidates event migration guidance, adds regulatory control mappings and an Operations section (#4020) (#3801) (#4021)
- Documentation updates the TypeScript FromAll guidance (#4518)
- Documentation clarifies read-model rebuilds after subject erasure (#4681)
- Documentation corrects tagging and replay retention claims (part of #5170)
Fixed
- Improve observer stream replacement during rolling restarts on SQL Server (part of #5350)
- The SQL live observer list now sees running-state timestamp writes that follow the state save (#5483)
- Stalled observer jobs are resumed while the observer is running, and failed resumes are retried (#4991) (#4665)
- Replay step retries wait until the replay resume succeeds (#5401)
- A replay whose job steps cannot be prepared, or a job step that throws, no longer leaves the observer or job stuck (#5408) (#5423)
- A catch-up marker is rolled back when its catch-up job is refused before it starts; markers are kept when the outcome is unclear (part of #4496)
- Projections stay connected after a silo restart (#4324)
- A failed constraint rebuild no longer blocks governed appends until restart (#5391)
- Constraint index catch-up no longer stalls while another constraint awaits repair (#4939)
- Incompatible MongoDB index keys and options are detected (part of #5284)
- Pinned event generations are honoured by strongly consistent projection reads (#4976)
- Capture source authorization is now submitted and unknown authorization types are rejected (part of #4591)
- Containers forward SIGTERM so the kernel shuts down gracefully (#5361)
- Connection-lifetime loops and reactor observers no longer inherit the first caller's execution context (#4369)
- Client logs transient startup connection retries below error level (#5426)
UseCratisChronicleno longer fails whenWithTelemetryis not called (#5492)- Projections no longer corrupt arrays of byte-array values (#4595)
- Projections no longer store a single-member
valueobject as a bare value (#5250) - Child
From<T>updates are applied when the creating event maps a differently keyed root (#4501) - MongoDB sinks treat decimal composite key parts that differ only in scale (for example 1.0 and 1.00) as the same key when caching (#5427)
- A reactor returning an empty concurrency-scope result no longer fails its partition (#5432)
- Cleared projection futures no longer share mutable state (#4604)
- Model-bound
[SetFromContext]accepts derived event-context functions (#5373) - Effect receipt lookups read with majority read concern (#4969)
- Formatted Expando conversion honours the provider's serializer options (part of #1298)
- CHR0026 keeps an explicit
[Subject]on another source's identifier (#4661) - Registering an event type with SQL storage no longer drops earlier generations (#4717)
- Saving a projection with an inferred read model in the Workbench works for children, counters and every-mappings (#4802)
- Disabled Workbench projection actions explain why (#4804)
- Projection editor and recommendations no longer suggest replay for event types that never occurred (#777)
- Failed-partition observation waits are diagnosed and snapshots survive poll failures (part of #4309)
- The NuGet package landing page renders a compatible readme (#3874)
- The .NET client now fails over to another kernel node within seconds when its node restarts, instead of waiting out the connect timeout (#5421)
- Reducers and reactors now log an error when they fail to re-register after their observation stream ends, instead of losing the failure silently (#5420)
- Reducers keep observing after a kernel silo restart instead of staying Disconnected until the application restarts (#4326)
- PII protection through
$ref-referenced schemas no longer allocates more per event than before (#5472) - Concurrent SQL Server kernel start-up no longer times out while enabling read-committed snapshot (#5475)
- Renaming an observer's stored state on MongoDB no longer fails on the immutable document id (#5493)
Security
- Protected event properties in lower-generation copies are protected with the higher-generation classification, fixing corrupted cross-store reads of
[PII]properties (#4645)