Release Notes
Release notes for currently supported releases
Last updated
Release notes for currently supported releases
Last updated
| | | | | | | | | | | | |
For release notes for releases that are no longer supported, as well as links to documentation for those releases, see .
As with any software upgrade, it is important to back up your data before you upgrade HEAVY.AI. In addition, we recommend testing new releases before deploying in a production environment.
For assistance during the upgrade process, contact HEAVY.AI Support by logging a request through the .
A configuration parameter for setting the maximum number of GPUs per concurrent query execution.
Support for Parquet import and HeavyConnect of POINT geometry columns from separate longitude and latitude columns.
Add zip compression support for shapefile export.
Export date columns to shapefiles as OFTDate instead of string.
A configuration parameter that allows exporting timestamps in non-ISO format.
Support for accelerated bidirectional CPU/GPU data transfers using pinned memory jump buffers.
Optimize queries that project variable (non-fixed) length column expressions.
Add H3_CellToPoint SQL function and H3_CellToBoundary_POLYGON SQL function.
A crash that could occur when tables are reordered for queries with ST function join predicates and multiple column projections.
Issue where H3_CellToBoundary_WKT SQL functions did not return closed polygon rings.
A crash that could occur when other ST SQL function calls are nested within ST_NPoints or ST_NRings function calls.
New default chart type - Box and Whisker/Violin Plot
Update Text and Text2 chart types to sanitize input and prevent javascript execution with custom code
Add binary multi-byte units to the “Number Formatting” options list
Add Shapefile and FlatGeoBuf export options for map charts
Enable ENABLE_GLOBAL_CUSTOM_SQL feature flag by default
Crash that can occur with certain combo chart configurations where # Records aggregate was set for both size and color measure
IMPORTANT - In HeavyDB Release 8.x.x, the system catalog is automatically migrated to support the new Column Level Security feature. This migration is necessary regardless if you intend to use this feature or not. Once the system catalog has been migrated in this manner, it is not backwards-compatible with earlier versions of HeavyDB. If you revert to an earlier version in this state, the system will be in an unstable state and manual intervention will be required. We recommend backing up your data before you upgrade HEAVY.AI.
8.x.x introduces a new licensing version that feature two new types of licenses: Node Locked Licenses and Floating Licenses. Enterprise customers that are upgrading to the 8.x release of HEAVY.AI are required to reach out to the HEAVY.AI Customer Success team for a new license before attempting to upgrade.
Fixes a race condition that could occur when concurrently accessing the memory_summary and memory_details system tables.
Fixes an intermittent crash that could occur for join queries with row-level security enabled.
Fixes a crash that could occur for certain query patterns involving the IN operator and subqueries with text projections.
Fixes a crash that could occur when the APPROX_PERCENTILE function is called with invalid arguments.
Improved logging around renderer device initialization and clean up
Fix for intermittent server crash when restarting the renderer after a CPU out-of-memory event
Combo chart with group by dimension displaying incorrect values in legend
Fix chromadb index creation bug and added support for importing faiss module on master process
Crossfiltering a chart with a non-joined data source did not refresh a chart with a joined datasource
Adjusting chart-level filters via the notch menu on a dashboard caused the chart to crash
Add ruler control to raster map charts and Esc key to close ruler control
Color all values - Colors an unlimited number of categorical values in raster charts using a hash function. Defaults to on, but can be turned off to access legacy coloring mode (coloring by top k)
Column names as export headers - Updates export logic to use column names as headers in the export file instead of names like measure0, dimension1; for the following charts: table, combo, pointmap, scatterchart, bubble chart, pie chart, heat
Scatter chart export - Implements data export for scatter plots and uses column names as export headers
Log scale axis - Implements log scale option for combo and scatter chart Y-axis
Infinite scrolling legend - Only active when “Color All Values” is on, allows the user to continually scroll the legend, automatically loading more values at the end of the scroll.
Parameter change (in custom SQL filter) not updating table chart
Improve logged statistics whenever rendering errors occur
ARM support (including NVIDIA Grace Hopper) - Docker install only
RHEL/Rocky 8 support - bare metal install only
Adds new Uber H3 functions and removes existing deprecated functions
Change how the GEOs library path is specified for use by some geospatial functions
Update CPU join hash tables to use the "cpu-buffer-mem-bytes" configured CPU memory buffer pool for memory allocations
Fixes a crash that could occur when a subquery within an IN clause references a non-existent column
Fixes a race condition that could occur when disk level caching is enabled for non-foreign tables
RTL Text support in mapbox basemap labels
Fix for shows the palette selector in combo chart when “# Records” is selected as a color measure
Fixes a crash when loading older dashboards, or instances and dashboards with customized/removed color palettes with new color consistency feature.
Fix for render error on multilayer raster charts that use new color mapping features.
Fix for render error when switching chart types while using new color mapping features.
Fix for bubble chart crash when using numeric dimension.
Fix for legends not showing in combo chart for continuous measures with no color measure selected.
Optimizes concurrent access and caching of table data during query execution.
Optimizes string dictionary memory allocations
Adds a validation to ensure that raster import/HeavyConnect with floating point coordinate types are either world or file space transformed.
Fixes a crash that could occur when attempting to import or HeavyConnect to Apache Spark generated empty Parquet files.
Fixes a crash that could occur when attempting to import or HeavyConnect to Parquet files containing Null columns.
Fixes a potential race condition that could occur when the Executor Resource Manager is enabled.
Fixes an exception that could occur for log based system tables when the log directory contains unexpected files.
Support for customizable color mappings across charts, including Combo, Pointmap, Linemap, Choropleth, Scatterplot, Pie, and Bubble, enabling consistent and tailored color schemes for enhanced visual control. Users can create, import, manage, and apply mappings, with options to save changes, delete mappings, or reset to default hash coloring as needed.
Uses Hash Coloring to ensure consistent color assignment for the same column values across all charts and dashboards.
Unified, default Categorical Colors palette is now used across all supported charts, ensuring out-of-the-box consistency in visualizations.
The previous "Color Set 2" palette has been integrated into the Categorical Colors palette set to streamline options and reduce confusion.
ARM support
Security updates and fixes
Fixes a crash that could occur when executing queries against ODBC backed foreign tables with negative decimal column values.
Updated web server dependencies to improve security
Improves memory utilization for queries with APPROX_COUNT_DISTINCT function calls
Significantly faster columnarization of "lazy fetched" result set data for multi-step queries
Significantly faster string operations, particularly for high-cardinality string inputs (3X+ speedups in some cases)
Fixes an error where users without column level privileges could not access views
Fixes a crash that could occur for queries that order by none encoded text expressions
Fixes an issue where wrong results could be returned for repeated queries with joins on encoded text column expressions
Fixes an error that could occur when large integer values are used in row level security policies
Fixes a CUDA 700 crash that could occur for left join queries with geospatial function predicates
Fix side panel overlay issue for time-picking filter.
Allow support for Guidance without an embedding server.
Support FAISS as an alternative to ChromaDB for embeddings.
Various minor improvements and enhancements
Improves memory management for import requests
Improves performance of group by queries through expanded use of shared memory
Improves performance of low cardinality group by queries
Improves performance of queries with sort on encoded string column expressions
Improves performance of result set reductions
Optimizes memory utilization and performance for sort queries
Adds more instrumentation around memory utilization
Adds support for ST_Centroid function calls with MULTILINESTRING column type arguments
Fixes an issue where S3 imports would not roll back correctly in certain error cases
Fixes an error where repetition of certain geo join queries could result in excessive memory utilization
Make Dashboard configuration panel resizeable.
Add UI customization options for Cloud Edition.
Fix scrolling bug with Guidance Snippet list.
Improve handling for loading a database with no tables.
Fix map points display issue on first render.
Add support for a vLLM Embedding server.
Fix ChromaDB configuration issues.
Adds support for bounding box clipping when importing geospatial files
Retries estimation queries on CPU if the initial query execution fails on GPU due to out of memory errors
Improves the error message that is logged when table data file reads result in errors
Fixes a crash that could occur when Parquet backed foreign tables reference incompatible files
Fixes an issue where spatial transforms were not applied correctly for some raster files containing sub-datasets
Fixes an issue where some string functions would result in error responses when used in predicates and update and delete queries
Significantly reduced GPU render memory usage and improved performance when rendering complex polygon data
Internal restructuring laying groundwork for improved concurrency between HeavyDB and Heavy Render
Fixed an issue where some GPUs would fail to identify a useful GPU memory heap reporting 0 available memory
Fixed a crash that could occur if a table involved in a query is removed between a render and a later hit-test query (mouse rollover in Immerse)
Fixed a crash that could occur during renderer shutdown and restart if an out of GPU memory error occurs during certain buffer allocation operations
Manage Table and Column Comments in Data Manager
Immerse theming control panel
Add alternative geocoder support using custom location list from a CSV
Add support for custom AWS Regions
Fix layer ordering label change after layer switching
Fixed the addition of the last categorical color not being used in color by measure
Make SQL Notebook line charts use GMT instead of local timezone
Added Column Preview summarizing for columns in SQL Notebook source list.
New Log File viewing for IQ Startup and IQ Build log tiles
Improvements to SQL Notebook UI
BETA: Added the new Guidance feature in the HeavyIQ SQL Notebook, powered by HeavyLM's advanced LLM service, provides contextual recommendations and real-time insights for optimized query generation and analysis.
Improved the SQL tool with optimized top-k calculations for string columns, adjusted default settings for cardinality thresholds, and added support for multiline comments in schema caching.
Improve runtime support for Python versions > 3.10.
Various performance improvements and fixes.
Relaxes Coordinate Reference System validations that occur on geospatial data import.
Fixes an issue where cross database queries could result in use of incorrect string dictionaries.
Fixes an issue where Host IDs could not be generated on OCI instances.
Fixes an issue where database switching resulted in errors on certain server configurations.
Fixes a CUDA 700 crash that could occur due to incorrect internal caching.
Fixes a parsing error that could occur when projecting array literals with the same element data types.
Fixes an issue where updates via subqueries with non unique rows could result in the wrong error message.
Fixes a crash that could occur when using the CARDINALITY function in certain cases.
Fixes a crash that could occur for certain query patterns involving casts to a REAL data type.
Fixes a crash that could occur for queries with certain case expression patterns involving DATE data types.
Fixes a crash that could occur in certain cases where resultset caching is enabled.
Fixes a crash that could occur for queries that specify LIMIT clauses with floating point values.
Fixes an issue where sql_validate requests with queries containing WIDTH_BUCKET function calls with subquery arguments can result in an error.
Fixes a crash that could occur for certain query patterns with subqueries that include LIMIT and/or OFFSET clauses.
Fix to return null hit-test result if not found.
Fixed an 8.0.0 issue with render gpu selection that would trigger a crash when using the start-gpu program option
Fixed a rare 8.0.0 issue with free GPU memory detection on systems with resizable BAR enabled
Enable HeavyIQ SQL Notebook for Free Edition.
Fix to show “Column not found” error text in a raster chart pop-up instead of crashing chart when hit-test column is not found.
Fix for crossfiltering of Number chart when filtering on a Combo chart with a custom SQL base dimension.
Change feature flag from “ui/hide_charts_headers” to “ui/hide_text_chart_headers” and only hide headers on Text charts.
Fix to set priority colors in Pointmap charts.
Update categorical legend domain on map chart zoom.
Fixes regression with prioritized color layers.
Enable HeavyIQ for Free Edition.
Improved caching and processing of table and column metadata.
8.0.0 introduces a new licensing version that feature two new types of licenses: Node Locked Licenses and Floating Licenses. Enterprise customers that are upgrading to the 8.0.0 release of HEAVY.AI are required to reach out to the HEAVY.AI Customer Success team for a new license before attempting to upgrade.
SELECT privileges are now required in order to execute UPDATE or DELETE commands.
Adds a new "columns" system table that contains information about all table columns across all databases.
Adds support for foreign tables that are backed by raster files.
Adds a new raster import mechanism that stores raster data in a tiled format. This can be enabled by setting the "enable-legacy-raster-import" HeavyDB server configuration parameter to false.
Adds support for column level SELECT privileges.
Adds support for comments on tables.
Adds support for comments on columns.
Adds support for null value filtering on raster import.
Adds MULTIPOLYGON to MULTIPOLYGON ST_Contains function support.
Improves performance for certain query patterns with BETWEEN predicate clauses.
Enables the "use-cpu-mem-pool-for-output-buffers" HeavyDB server configuration parameter by default.
Updates the default value for the "ndv-group-estimator-multiplier" HeavyDB server configuration parameter.
Fixes a crash that could occur in the load_table_binary_columnar Thrift API when request payloads are malformed.
Fixes a crash that could occur in the detect_column_types Thrift API when an uneven number of levels across columns are read for Parquet files.
Fixes a crash that could occur in certain cases where Parquet files are imported from S3 with debug timers enabled.
Fixes a race condition that could occur with foreign table scheduled refreshes.
Fixes a crash that could occur when the ST_NPoints function is called with literal arguments.
Fixes a hang that could occur when a "clear CPU" or "clear GPU" request is made after the Executor Resource Manager returns an error response for a previous query.
Fixes an issue where certain sorted join query patterns could return wrong results.
Fixes an intermittent "Ran out of slots in the query output buffer" error that could occur due to incorrect reuse of the cardinality cache for similar queries.
Fixes a crash that could occur in certain cases where the null value results from the ST_Centroid function are passed in as arguments to other ST functions.
Fixes use cases where the Executor Resource Manager did not account for result set buffers that are allocated through the CPU buffer pool.
Fixes an issue where the use of the approx_count_distinct function could cause a slowdown for certain query patterns in distributed configurations.
Fixes an issue where wrong results could be returned due to an overflow when dividing two decimals.
Fixes a crash that could occur for certain query patterns with Common Table Expressions that project point columns.
Fixes a crash that could occur for queries that specify LIMIT clauses with floating point values.
Fixes a crash that could occur for certain cases where the output of an ST function is passed in as an argument to the ST_Contains function.
Fixes an issue where the ST_Distance function can return wrong results for cases where a polygon encloses another polygon.
Shader compiler optimizations.
Increased default res-gpu-mem value from 385Mb to 768Mb.
Fixed shader compiler error when combining symbol rotation and accumulation.
Catch failed render result buffer allocations and throw an out of GPU memory error.
Fixed an issue that could cause a Cuda Error 700.
Fixes a crash that could occur for render requests containing certain sort query patterns.
Introducing the SQL Notebook in Immerse, which in addition to supporting a notebook view of query history, allows inline visualization of query results using bar charts, line charts, scatterplots, heatmaps, choropleths, and pointmaps. The SQL notebook also integrates directly with the HeavyIQ conversational analytics module, allowing you to ask questions of your data in natural language, returning visualizations, natural language answers, and tabular results along with the generated SQL. The new SQL notebook is off by default, but it can be enabled by setting "dev/enable_notebook_ui_sql_editor": true in the servers.json file.
Viridis, Turbo, and Plasma added as new color palettes.
Fixes dropping of filter clauses from SQL on exiting dash panel edit mode.
Fixes cross-filtering on Number chart when filtering on a Vega Combo chart with a custom SQL base dimension.
Fixed regression for enabling feature flag “ui/enable_chart_filter_view".
Fixed dashboard loading error when switching tabs with a HeavyDB error present.
Fixed joined table referenced in a size measure on the Line Map chart.
Fixed issue with non-integer operands in modulo operation error for Contour chart.
Fixed issue with color range display of filtered values for the Contour chart.
Fixed code generation for order of evaluation in map charts using SAMPLE_RATIO.
Fixed issue with Pointmap rendering of filtered data.
Fixed various minor display and styling issue.
HeavyIQ Conversational Analytics allows users to ask questions of their data using natural language by leveraging a custom Large Language Model (LLM) that has been fine-tuned on over 60,000+ training pairs to provide state-of-the-art performance on text-to-sql and other data analytics tasks. HeavyIQ requires no external API calls and guarantees data privacy by virtue of being a fully offline model.
The primary interface point for the new HeavyIQ conversational analytics capabilities are via the new SQL editor (see above), but it can also be utilized via API if desired.
Added operator, allowing access to large language model inference within SQL
Adds support for a function with array parameters, which can be used to compute the similarity between arrays (vectors)