Resolved Issues

Additional DSS added to handle Apache client-level timeout

n Connector rule, the system timed out a connection after 1 hour even after a connection time timeout was configured to be 30 seconds. Apache's refactoring of their HTTPClient in version 4.3 specifies request timeouts in two ways: on the client-level and on the request-level. Client-level timeouts are enforced prior to and during the handshake, whereas request-level timeouts are enforced after the handshake has been made. Previously, RESTConnector.java and ComponentsHTTPClient.java were configured to only set the user-specified timeout on the request, and not on the client. This caused the client-level timeout to default to "0", or an infinite amount of time. In this reported issue, the remote host closed the connection before the handshake had been completed so the connection remained open for several hours. To resolve this, the DSS "ClientLevelHTTPTimeout" has been added to allow specifying a client-level socket timeout, and ComponentsHTTPClient.java and RESTConnector.java have been amended to assign this value to the client. This DSS is set to 0 by default to preserve backwards compatibility.

Check added to minimize Obj-Open-By-Handle error logging

When using a Data Type with the "Automatically generate a unique id " option, calling the Save-DataPage method by using the savable data page of the data type finished correctly but showed Obj-Open-By-Handle errors on PegaRules.log. Investigation showed the exception was thrown when running the save plan from DataPageSaverImpl: while attempting to run the save plan, the system does not know whether a parameter (pyGUID in this case) will be required to run the save plan or not, meaning that it can't detect any possible error in DataPageSaverImpl. The implementation instead makes a call to db.open to check if an instance exists and hence logs are thrown. To resolve the error logging, a check has been added: if the save-to class has an autogen key and the savable data page instance doesn't have the autogen key in it, the system will directly call pxCreateRecord. This will avoid a call to db.open to check if instance exists and hence no failed logs will be thrown. This partial change will work only for classes having an autogen key and in cases where the page is trying to create a record by intentionally not passing the key.

Handling added for BIX count mismatch in manifest and extracted CSV data

A mismatch in the counts in the manifest file when compared to the data extracted in the CSV files was traced to a corrupted or wrong pzinskey in the source database. If the BLOB is corrupted while extracting to output type and manifest type to CSV, an invalidstream exception was generated and resulted in the manifest file and extracted CSV file total instance count mismatch. This has been resolved by adding handling to balance the counts.

Landing page refresh modified to avoid frequent reloads

Approximately every two weeks, clients were able to login but it was not possible to work as the landing page refreshed constantly until all of the nodes were rebooted and the issue was cleared. Investigation indicated the frequent reloads were related to the cacheing of the operator details, and this has been resolved by updating the datapage reload strategy based on 'when' so frequent reloads will be avoided.

Handling added to avoid primary key exception in pxExtractDataWithArgs

While running the activity pxExtractDataWithArgs with -c, the "Error while getting the sequenceNumberERROR: duplicate key value violates unique constraint pr_log_pk" occurred if both Parent and Child classes were pointing to different table and extraction started from the child class. In this case, because the Parent and Child classes were pointing to different tables while extracting with the -c option, the class list order caused a primary key constraint violation exception while generating the process sequence number. To resolve this, while adding the classes to the instance map, the order will be changed to parent followed by child classes.

Resolved BIX Extract Summary Status Count mismatch

A mismatch was seen between the summary of the Extract Summary Status Count in the Execution History stats versus the logs, for example the execution history saying 98.74% completed while the log says 100% or 100.45% completed. Investigation showed there was mismatch in the query based on the filter value given in the extract rule and the where clause in the query that resulted in an error in calculating percentage in case of any exception. This has been resolved.

Landing page refresh modified to avoid frequent reloads

Approximately every two weeks, clients were able to login but it was not possible to work as the landing page refreshed constantly until all of the nodes were rebooted and the issue was cleared. Investigation indicated the frequent reloads were related to the cacheing of the operator details, and this has been resolved by updating the datapage reload strategy based on 'when' so frequent reloads will be avoided.

Enhancements added to S3 repository implementation

The following changes have been made to avoid redundant/unnecessary S3 SDK calls to look-up metadata for files/folders related to integrated S3 Repository API implementation:

 - Provided storage/AWSSDK/enableSDKMetrics DASS to enable AWS SDK metrics for S3 repositories system-wide to troubleshoot performance issues
 - Added timers for Pega platform repository APIs. To see these timers in Pega application log, set log level to WARN for the log category pxIntegration.Repository.S3 and set log level to DEBUG for the logger com.amazonaws.latency
 - Optimized FileS3Adapter.getMetadata() by avoiding object-metadata lookup for directories and caching rootpath directory metadata for 5 mins
 - Handled exceptions while uploading file to S3 bucket and convey appropriate error message to RepositoryManager.createFile() API consumer
 - Refined DEBUG logging and introduced new loggers to emit stacktraces
 - Enhanced RepositoryManager.initCache() and  FileStorageManager.registerStorageClasses() to maintain non-overridable repositories (that are initialized based on DASS/DNode settings on node start-up) and to prohibit overriding the same.

Oracle temporary space explicitly freed after writing LOBs

Oracle temporary space used to write the LOB was not freed up after writing records to Log-Service-File when using OJDBC8 or OJDBC10 versions and the table involved had inline BLOB storage such as small blob tables like Log-service-file (2-4kB blobs). This did not occur when using OJDBC7, but investigation with Oracle could not determine why the older version did not have this issue. This issue has been resolved by explicitly freeing the LOB space with a call to free() / freeTemporary() , a utility for LOB objects that was introduced in JDBC 4.

Resolved exception from expanding deserialized Data Page

When using a data page backed by a report definition where the data page data type was Rule- and the virtual page data type was Rule-Obj-Property, serializing and deserializing the data page and then expanding the deserialized data page generated the exception "the reference .pzInsKey is not valid. Reason: .pzInsKey is mapped to class Rule-Obj-Property, which is not the primary class of the list: Rule-; must specify a group name in mapping". The issue occurred only if "Run on Data Page" on the data page was checked and "Report on descendant class instances" on the report definition was unchecked, and has been resolved by updating the runLoadActivity() of DeclarativePageDirectoryImpl to skip setting pyObjClass for the report definition if it already present, i.e. if already set by the loader activity.