Description:
Dear customers,

A new patch, GridServer-5.1_SP1_P20110202, has been uploaded to customer.datasynaspe.com and available to download under /GridServer 5.1/SP1/Patches (http://customer.datasynapse.com/aidix/BrowseControl.do?state=browse&context=GridServer&path=.%2FGridServer+5.1%2FSP1%2FPatches%2FGridServer-5.1_SP1_P20110202-update.jar). This is a cumulative patch which includes fixes from previous patches. See release notes below for more details, and as always, feel free to contact us for any questions.

Global DataSynapse TIBCO support
support@tibco.com

***

GridServer 5.1 SP1 Patch 20110202 Release Notes

This patch applies updates to the following modules: Manager, Java

Stability

GS-10230    A rare race condition resulted in an internal corruption that ended up blocking the scheduler and component login and logoff, thus rendering the broker unusable.ther similar but far more unlikey race conditions were also fixed.    5.1 SP1

This patch also includes the following fixes from earlier patches that includes updates for the included modules.

Java

GS-10235    Locking optimizations have been added to the Java driver which should result in noticeably better collection performance. In order to enable these, use system property com.livecluster.driver.JobSpaceRunner.submissionLockOptimization=true on your client application.     5.1 SP1
GS-10238    By default, if a Driver is using DDT and it runs out of disk space, it will switch to non-DDT and send its input to the Broker. While this behavior might be desirable for some cases and increases fault tolerance, it can cause instability on the Broker if that also runs out of disk space. The following will now occur on a DDT file write failure:
On the Driver
•    New log messages now indicate what will happen in the event the driver runs out of space (e.g., "Task will fail" or "Data will be sent to the Manager")
•    - If the Java system property 'com.livecluster.driver.DriverDataTransferPlugin.failTaskOnWriteFailure' is set to 'true' on the Java driver, it will no longer switch to non-DDT, and the task will eventually fail due to the engine not finding the input file.
On the Broker
•    The task will not be kept in memory and eventually fail for the same reason. This prevents the Broker from running out of memory.    5.1 SP1

Admin Interface

GS-9579    When a task was retried due to timeout, and then cancelled due to too many retries, the amount of completed tasks was higher then the total amount.    5.1 SP1

Stability

GS-10230    Under rare circumstances, it was possible for the broker to use 100% of the CPU even though the broker was not particularly busy.     5.1 SP1

Database

GS-9380    When a task was rescheduled, the end time for the task completion event would be incorrectly reported in the DB if the original Engine completed the task.     5.0 FCS

Routing and Balancing

GS-9547    A rare race condition on task retry and reassignment resulted in the number of waiting tasks to be incorrectly incremented. This affected Engine balancing when the Service indicated waiting tasks when there were actually none.    5.1 SP1 20100416
GS-9170    When an Engine restarted to load a Grid Library, it was counted as a busy Engine even when not logged in. This could lead to negative idle engine counts, and the Max Engines setting on a Broker being violated. This has been fixed.     5.1 SP1
GS-9046    A bug in the Engine Sharing algorithm resulted in idle Engines not being assigned to Brokers when needed, under certain situations.    5.1 SP1

Configuration and Repository

GS-9126    A new user entry was created when a user logged in using a different case, when login was configured case-insensitive.    5.0 Service Pack 3

Services

GS-9489    When running a Java Driver on an Engine, Calling ServiceInvocationException.getLastCause() would fail with a ClassNotFound exception if the cause was a user-defined exception. This method now uses the thread's context classloader to load the class, so it will be found as long as the exception class is in the Service's classpath and the method is called in the task execution thread.    5.1 SP1
GS-9140    DSMaxConcurrentJobs was not respected on the driver causing it to go over the limit of submitting services. Fixed previously for C++ Drivers, now fixed for Java Driver. .NET Driver did not have this issue.    5.1 SP1
GS-9095    The DSMaxConcurrentJobs was not being respected on the Driver, allowing for more concurrently submitting services to run than desired. This issue is fixed for C++ Windows Drivers only in this patch.    5.1 SP1

Optimizations

GS-9073    The Client Wait Time (when no results were available for collection on the previous request) is based on the total number of clients connected. However, this is not optimal for Brokers which have many clients connected of which many have no running services and this not collecting. This wait time can now be set to only consider clients with running services. This is done by setting the Java system property on the Broker: com.livecluster.broker.DriverServicePlugin.driverWaitCountActiveOnly=true     

Security

GS-8951    All potential Manager HTTP 500 errors are now caught and dealt with appropriately (except for those required by Web Services.) All Manager input is now validated to eliminate cross-site scripting vulnerabilities.    5.1 SP1

new patch release: GridServer-5.1_SP1_P20110202

Product: TIBCO DataSynapse GridServer Version: OS: --------------------