INC-142216 · Issue 596668
VBD performance improvements
Resolved in Pega Version 8.5.2
Performance issues were seen when running a decision funnel simulation. As part of the simulation, a VBD collection was created on the fly to store the metrics and linked to the simulation. Before each insert, which happened in batches of 500 records, there was a check if such collection exists. This check was required in earlier versions as the process relied on the VBD service being running, and the run would fail otherwise. Recently, VBD handling became more resilient and was updated to allow inserts even if the VBD server is down. If the VBD service is stopped before the simulation is completed, the simulation will continue writing records. When the VBD service starts again, it reads the backlog, and the simulation displays the final results. To adapt to this resiliency and improve performance, the system has been updated so that the VBD collection is not created and checked before the insert call, but only when the simulation is run or scheduled.
INC-143441 · Issue 608899
Strategy Simulation Runs made more resilient
Resolved in Pega Version 8.5.2
Intermittently, a decision funnel simulation would fail with particular strategy logic or a particular StrategyTracker would not contain the results for all components so the system would not be able to calculate input SRs. Handling has now been added to make the simulations more resilient so they will not fail with a null pointer error due to incomplete strategy inputs.
INC-144864 · Issue 599146
Unneeded AUC error percentage calculation removed for better system performance
Resolved in Pega Version 8.5.2
When generating snapshots of the ADM models, an inefficiency in the metric calculation for the AUC error percentage caused an excessive amount of memory to be allocated which could cause slowdown or require a system reboot. This has been resolved by removing the problematic AUC error percentage calculation as it is not used in the UI.
INC-148006 · Issue 605581
Enhancements added for very heavy use of strategy decision logic
Resolved in Pega Version 8.5.2
Recent updates were made to improve memory performance for scenarios where a single request primary page had tens of thousands of pages under an embedded pagelist property and the decision logic involved a strategy running on all the pages in that page list. To further improve performance, the array iterator has been modified and a DSS has been added that makes the threshold (50) in SSAExecutionContextIteratorFactory tunable.
INC-150854 · Issue 608368
Logic optimization disabled in offers
Resolved in Pega Version 8.5.2
Offers that should not be eligible to customers were passing through even though the offer was narrowed down to using “OR” conditions in the Engagement policies. This was traced to the logic optimization ignoring precedence, and has been resolved by disabling the optimization.
INC-150395 · Issue 625068
Tokenizer updated to handle commas
Resolved in Pega Version 8.4.5
The Text Analyzer was not working as expected in cases where the number was combined with a comma (,) with it but was working when a space was used between the number and the comma. This was traced to the tokenizer not correctly processing and splitting the input text when there was a special character before or after the token. This has been resolved by updating the tokenizer logic.
INC-150873 · Issue 612896
Performance improvement for saving ADM model rule
Resolved in Pega Version 8.4.5
Saving an ADM model rule generated a heap dump. The stack trace from the heap dump showed a thread consuming the maximum memory (4.7 GB of memory). Configurations on all factories are updated when a model rule is saved, but at the time of development it was not expected that there would be a lot of factories in a Dev environment so the system was loading all existing factories simultaneously into memory and updating configurations on them. To improve performance, an update has been made which will now sequentially load factories and update the configuration.
INC-156818 · Issue 628465
Materialization uses time limit boundary for query
Resolved in Pega Version 8.4.5
After turning on Materialization for pyIHSummary and OfferOutcomesForPast45Days datasets, an SQL query was taking an excessive amount of time and causing multiple alerts in the logs. Investigation traced the issue to database partitioning, specifically that running a query where the pyOutcomeTime range spanned multiple partitions was causing the indexes for all partitions in the range to be opened. To resolve this, the query has been updated with a DSS to support a partition size of min(pxOutcomeTime) to limit the time range to querying day by day, or hour by hour, or any other chronology unit specified. If there are no records for the current limit, then it will look at the next partition. This should prevent the query from needing to open more than 1 or 2 partitions.
INC-157357 · Issue 636711
Hazelcast remote execution not called from synchronized context
Resolved in Pega Version 8.4.5
After navigating to the Admin Studio portal to view the nodes, the portal was temporarily freezing. Investigation of the thread dump revealed this was caused by a DDS pulse sending a remote execution call to all nodes to update logger settings even though the site was not using DDS. This has been resolved by updating the system to avoid calling Hazelcast remote execution from a synchronized context.
INC-157629 · Issue 626632
Duplicate key exception resolved for adaptive model
Resolved in Pega Version 8.4.5
During the model snapshot update, a DuplicateKeyException was generated while trying to insert a record in to the predictor table. This did not affect the model's learning, but did appear ion the model monitoring report. This was traced to a local scenario of having the same outcome values defined on the model with different cases (Accept and accept). All predictors used in an Adaptive model are inserted into the model monitoring tables as a part of the monitoring job: because the monitoring tables are not case sensitive, this lead to a unique constraint exception since there were multiple IH predictors with the same name. To resolve this, validation has been added which will skip adding duplicates from new responses.