New in InterSystems IRIS 2025.3

The internal layout of the Approximate Nearest Neighbour (ANN) index used by vector search has been modified to improve duplicate entry management and reduce query-time overhead. Benchmarks show throughput and latency improvements of up to 10% depending on workload characteristics.

The updated format is applied automatically when the index is rebuilt. Rebuilding is not required for continued operation: the runtime is fully backwards compatible and can read and write indices in both the legacy and updated formats.

Adaptive Analytics has published AtScale 2025.4.1. This release contains bug fixes and performance enhancements, and introduces support for inbound query CROSSJOIN operations that return inbound cells.

In past releases, it was necessary to configure pool sizes of one in order to force sequential message processing in order to maintain first-in-first-out (FIFO) queuing behavior for messages in a message feed. This release introduces customer-calculated message group identifiers, which allows for pool sizes of greater than one while maintaining FIFO. These identifiers also enable parallel processing of messages that had no direct dependency, increasing throughput of a production.

There are two settings provided for Business Processes and Business Operations under the header “FIFO Message Grouping” - Group Calculation (FMGCalculation) and Group Completion Hosts (FMGCompletionHosts) - that allow developers to configure the group identifier and the host that releases the group handle for a given group identifier, respectively. Group identifiers are configured by a Data Transformation (DTL) and is typically set to a value that identifies a FIFO dependency, such as a patient or device id.

Journal Archiving allows InterSystems IRIS to push closed journal files—after compression—directly to a cloud-based archive location, reducing the use of high-IOPS local storage. In addition to Amazon S3, Azure Blob Storage is now supported as an archive target.

This release extends Foreign Table pushdown from the individual table level to the server level, enabling pushdown of entire JOIN operations across two or more tables originating from the same foreign server. Previously, such joins required pulling all qualifying rows from each table into InterSystems IRIS for local execution, which could be costly when querying large tables without direct filtering. This change represents the final major milestone in maximizing query computation pushed down to eligible foreign sources.

When executing complex queries, InterSystems IRIS SQL may create temporary data structures (tempfiles) for operations such as sorting or aggregation. If table statistics indicate these structures will be small, the optimizer may use in-memory arrays, which are significantly faster than process-private globals (PPGs). However, if the statistics are inaccurate, the optimizer may underestimate the size of a tempfile, potentially exhausting process memory at runtime and resulting in a <STORE> error.

To address this, InterSystems IRIS SQL now monitors memory usage during query execution. If the system detects that memory is running low, it will automatically fall back to using PPGs in cases where inaccurate statistics led to an underestimation. While representative table statistics remain strongly recommended, this safeguard reduces the risk of runtime failures due to misestimates.

This release improves the storage of lightweight runtime statistics for each SQL statement in the Statement Index, enhancing the performance of common queries against INFORMATION_SCHEMA.STATEMENTS and related tables.

Additional enhancements have been made to SQL Query Plan output, including new information messages.

Performance has been improved for queries that aggregate data across low-cardinality field on columnar tables.

The 2025.3 release includes substantial metrics additions to the metrics available through OpenTelemetry and the metrics API. These additions include:

• An entry for each iris_process, listing its pid, state, internal wait state, job type, and client name.

• Counters by process for global references, ObjectScript commands, block allocations, physical reads, blocks queued for writing, and journal entries.

• A gauge for PPG block usage in megabytes by process.

• The iris_ecp_servers metric, which reports a series for each ECP connection, listing the server it is connected to and the connection state.

• The iris_ecp_clients metric, which is similar to iris_ecp_servers, but lists the information for clients.

• iris_wd_* metrics, which provide insights into the state of the Write Daemon.

When enabled, your instance will collect daily system performance reports. This feature is configured through the SystemPerformanceDailyReportsOn CPF parameter.

This release now exposes the parent process ID and start time in %SYS.ProcessQuery. Both metrics are exposed in the Management Portal (on the Processes and Process Details pages) and in ^JOBEXAM (under Examine details).

The new security wallet lets you securely store and access “secrets” in InterSystems IRIS without exposing those secrets to application code. In this context, a “secret” refers to any piece of sensitive information required by an application at runtime, including API keys, encryption keys, usernames, and passwords. Secrets are grouped together in collections for convenient access management and access to these secrets is audited.

Secrets are stored in the IRISSECURITY database, which must be encrypted to protect sensitive data.

• When editing host categories, viewing connections are permissible

• A view icon on all rule sets open the specific rule set in the Rule Editor split panel

• Updated icons to open and close a panel

• When selecting the test button, the button is highlighted to indicate that the test panel is open

• Added a legend for Host Statuses in the information icon on a host

• Added drop down icons for all drop-down enabled settings fields

• Host Properties are now searchable, with added “expand all” sections

• IP Address and Port Number are now searchable in the Production Configuration Filter Bar

• Host Connections appear as they are registered by the front end, rather than appearing all at once when connections are fetched

• DTL Segment Path copies with both field text and ordinal numbers

• The “hover over” segment icons for copying or adding a mapping are more accessible

• When adding a mapping, hovering over the particular segment now only shows the copy icon

• The DTL Editor split screen header and toolbar both remain in place when scrolling