Service-specific recommendations for watched folders

The Generate PDF service can convert only one file at a time for these file types: Microsoft Word, Microsoft Excel, Microsoft PowerPoint, Microsoft Project, AutoCAD, Adobe Photoshop®, Adobe FrameMaker®, and Adobe PageMaker®. These are long running jobs; therefore, make sure you keep the batch size to a low setting. Also increase the repeat interval if there are more nodes in the cluster.

For PostScript (PS), Encapsulated PostScript (EPS), and image file types, the Generate PDF service can process several files in parallel. You should carefully tune the session bean pool size (which governs the number of conversions that will be done in parallel) depending on the capacity of your server and the number of nodes in the cluster. Then increase the batch size to a number that is equal to the session bean pool size for the file types you are trying to convert. The polling frequency should be dictated by the number of nodes in the cluster; however, because the Generate PDF service processes these kinds of jobs quite fast, you could configure the repeat interval to a low value such as 5 or 10.

Even though the Generate PDF service can convert only one OpenOffice file at a time, the conversion is quite fast. The above logic for PS, EPS, and image conversions also applies to OpenOffice conversions.

To enable uniform load distribution in the cluster, keep the batch size low and increase the repeat interval.

For best performance when processing barcoded forms (small files), enter 10 for Batch Size and 2 for Repeat Interval.

When many files are placed in the input folder, errors with hidden files called thumbs.db may occur. It is therefore recommended that you set the Include File Pattern for the include files to the same value specified for the input Variable (for example, *.tiff). This prevents Watched Folder from processing the DB files.

A Batch Size value of 5 and Repeat Interval of 2 is normally sufficient because the Barcoded Forms service usually takes about .5 seconds to process one barcode.

Watched Folder does not wait for the Process Engine to finish the job before it picks up new files or folders. It keeps scanning the watched folder and invoking the target service. This behavior can overload the engine, causing resourcing issues and time-outs. Ensure that you use repeat interval and batch size to throttle the Watched Folder input. You can increase the repeat interval and reduce the batch size if more watched folders exist or enable throttling on the endpoint. For information about throttling, see About throttling.

Watched Folder impersonates the user specified in the user name and domain name. Watched Folder invokes the service as this user if invoked directly or if the process is short-lived. For a long-lived process, the process is invoked with the System context. Administrators can set operating system policies for Watched Folder to determine which user to allow or deny access to.

Use file patterns to organize result, failure, and preserve folders. (See About file patterns.)

Watched Folder relies on the Quartz scheduler for scanning the watched folders. The Quartz scheduler has a thread pool to scan them. If the repeat interval for the watched folder is very low (< 5 seconds) and the batch size is high (> 2), a race condition can occur. When this condition occurs, one file is picked up by two Quartz threads:

• One of the threads successfully finds the file and invokes the target service with the file.

• The second thread sees the file but fails when it tries to find out if the file is valid (read or write file), which causes false failures that indicate that the file cannot be processed because it is read-only. This happens only with a low repeat interval and a high batch size.