Fluens.Messaging.SqlServer 0.9.1

Fluens.Messaging.SqlServer

SQL Server provider for Fluens.Messaging — nvarchar(max) attempt_history, dbo schema, SqlException transient-failure classification, and a startup READ_COMMITTED_SNAPSHOT guard.

Installation

dotnet add package Fluens.Messaging.SqlServer

Usage

Opt in from the AddMessaging callback — there is no separate top-level registration method:

fluensBuilder.AddMessaging(o => o.AddSqlServer());

AddSqlServer() registers SqlServerMessagingConventions as the singleton IMessagingProviderConventions (TryAddSingleton — calling it twice, or alongside AddPostgres(), keeps whichever ran first) plus a ReadCommittedSnapshotValidator hosted service (idempotent — repeated calls do not add a second one).

Startup schema verification

This package supplies the catalog reader (SqlServerSchemaCatalogReader, reading INFORMATION_SCHEMA) that the Fluens.Messaging startup schema verification needs: with AddSqlServer() registered, a host whose messaging tables do not match the installed Fluens version is refused at start with a SchemaVerificationException that names the upgrade helper to call. The verification runs before any hosted service starts, so on a database that is both stale and without snapshot isolation the schema error is reported first. Without a provider package there is no catalog reader and the check is skipped. Fluens:Messaging:VerifySchemaOnStartup set to false switches it off. See the Fluens.Messaging README.

READ_COMMITTED_SNAPSHOT startup guard

SQL Server's default READ COMMITTED isolation takes shared locks on read, which does not reproduce PostgreSQL's MVCC read semantics the outbox/inbox fetch paths rely on. At host start, ReadCommittedSnapshotValidator resolves every distinct module database (deduplicated by connection string) and queries sys.databases.is_read_committed_snapshot_on. When it is false for any database, host start aborts with OptionsException<MessagingOptions> carrying the remedy:

ALTER DATABASE [YourDatabase] SET READ_COMMITTED_SNAPSHOT ON; -- requires no other active connections

Conventions

The transient classification feeds the stage-1 in-memory retry gate on LaneWorker: a transient failure walks the full two-stage retry policy, a non-transient failure (e.g. a constraint violation) skips straight to stage-2 scheduling.

License

This project is licensed under the MIT License.