fix bug -- durable stuff commit must happen before workflow is started

Author: maxdml

dbos/dbos.go

@@ -154,7 +154,6 @@ func WithoutCancel(ctx DBOSContext) DBOSContext {
 		return nil
 	}
 	if dbosCtx, ok := ctx.(*dbosContext); ok {
-		// Spawn a new child context without the cancel function
 		return &dbosContext{
 			ctx:                context.WithoutCancel(dbosCtx.ctx),
 			logger:             dbosCtx.logger,
@@ -168,15 +167,13 @@ func WithoutCancel(ctx DBOSContext) DBOSContext {
 		}
 	}
 	return nil
-
 }
 
 func WithTimeout(ctx DBOSContext, timeout time.Duration) (DBOSContext, context.CancelFunc) {
 	if ctx == nil {
 		return nil, func() {}
 	}
 	if dbosCtx, ok := ctx.(*dbosContext); ok {
-		// Spawn a new child context with a deadline and a cancel function
 		newCtx, cancelFunc := context.WithTimeoutCause(dbosCtx.ctx, timeout, errors.New("DBOS context timeout"))
 		return &dbosContext{
 			ctx:                newCtx,
@@ -319,6 +316,7 @@ func (c *dbosContext) Launch() error {
 }
 
 // We might consider renaming this to "Cancel" to me more idiomatic
+// TODO: shutdown should really have a timeout and return an error if it wasn't able to shutdown everything
 func (c *dbosContext) Shutdown() {
 	c.logger.Info("Shutting down DBOS context")
 

dbos/workflow.go

@@ -572,6 +572,7 @@ func (c *dbosContext) RunAsWorkflow(_ DBOSContext, fn WorkflowFunc, input any, o
 	}
 	insertStatusResult, err := c.systemDB.InsertWorkflowStatus(uncancellableCtx, insertInput)
 	if err != nil {
+		c.logger.Error("failed to insert workflow status", "error", err, "workflow_id", workflowID)
 		return nil, err
 	}
 
@@ -597,6 +598,7 @@ func (c *dbosContext) RunAsWorkflow(_ DBOSContext, fn WorkflowFunc, input any, o
 		}
 		err = c.systemDB.RecordChildWorkflow(uncancellableCtx, childInput)
 		if err != nil {
+			c.logger.Error("failed to record child workflow", "error", err, "parent_workflow_id", parentWorkflowState.workflowID, "child_workflow_id", workflowID)
 			return nil, newWorkflowExecutionError(parentWorkflowState.workflowID, fmt.Sprintf("recording child workflow: %v", err))
 		}
 	}
@@ -631,6 +633,11 @@ func (c *dbosContext) RunAsWorkflow(_ DBOSContext, fn WorkflowFunc, input any, o
 		stopFunc = context.AfterFunc(workflowCtx, dbosCancelFunction)
 	}
 
+	// Commit the transaction. This must happen before we start the goroutine to ensure the workflow is found by steps in the database
+	if err := tx.Commit(uncancellableCtx); err != nil {
+		return nil, newWorkflowExecutionError(workflowID, fmt.Sprintf("failed to commit transaction: %v", err))
+	}
+
 	// Run the function in a goroutine
 	c.workflowsWg.Add(1)
 	go func() {
@@ -668,11 +675,6 @@ func (c *dbosContext) RunAsWorkflow(_ DBOSContext, fn WorkflowFunc, input any, o
 		close(outcomeChan)
 	}()
 
-	// Commit the transaction
-	if err := tx.Commit(uncancellableCtx); err != nil {
-		return nil, newWorkflowExecutionError(workflowID, fmt.Sprintf("failed to commit transaction: %v", err))
-	}
-
 	return &workflowHandle[any]{workflowID: workflowID, outcomeChan: outcomeChan, dbosContext: uncancellableCtx}, nil
 }