v0.29.0

@kitcn/resend

Published @kitcn/resend@0.29.0.

kitcn

Minor Changes

• #421 26dd9b5 Thanks @MikeyZhang75! - ## Breaking changes

  • Register migrationStatus and aggregateBackfillStatus as internal
    queries instead of internal mutations. Polling migration or aggregate
    status no longer opens a write transaction, so a status monitor stops
    competing for OCC write slots on the very tables it is reporting on. Both can
    now back a live subscription, and neither can be scheduled any more.

  // Before — status read had to run from a mutation context
  export const getStatus = authMutation.mutation(async ({ ctx }) => {
    const server = createServerCaller(ctx);
    return await server.migrationStatus({});
  });
  
  // After — status reads from a query context
  export const getStatus = authQuery.query(async ({ ctx }) => {
    const server = createServerCaller(ctx);
    return await server.migrationStatus({});
  });

  • Accept an internalQuery builder in createOrm(...) alongside
    internalMutation. Apps generated by kitcn codegen pass it automatically;
    hand-written createOrm calls that use orm.api() should pass it too so the
    status procedures are built with the app's own Convex builder.

  // Before
  const orm = createOrm({ schema, ormFunctions, internalMutation });
  
  // After
  const orm = createOrm({
    schema,
    ormFunctions,
    internalMutation,
    internalQuery,
  });

  Patches

  • Fix migrationStatus reading the entire migration_run history to return the
    most recent runs. The listing now walks a new by_started_at index in reverse
    and stops at limit, so the cost of a status call no longer grows with the
    number of migrations ever run. Runs that share a startedAt millisecond now
    order newest-first.
  • Bound the limit argument of migrationStatus at MAX_STATUS_RUN_LIMIT
    (100, default 25), exported from kitcn/orm/migrations, so a caller
    cannot reintroduce an unbounded read through the args.
  • Resolve migrationStatus's runId and activeRun through their existing
    indexes instead of scanning the run history. activeRun now agrees with the
    run that migrate cancel targets.