Following are known issues, limitations, and changes to default behavior in HEAVY.AI.
Because HeavyDB supports Web Mercator projection only, applications that use coordinates other than Web Mercator may not render data accurately on their maps.
Whenever the result of a query is going to be used for another query (for example, CREATE TABLE AS SELECT, any multi-step query, and so on), HEAVY.AI performs columnar conversion to change the intermediate results into proper columnar format that all HEAVY.AI queries expect as input. Variable length types, including all geometry targets, are not supported when performing columnar conversion.
BLOSC_* environment variablesHEAVY.AI uses a compression library called BLOSC that reads operating system environment variables and changes behavior according to those variables. Do not set any of the environment variables listed below.
OPTIMIZE TABLE with VACUUM increases metadataThe with (vacuum = 'true') option has the suboptimal effect of increasing the size of your metadata in HEAVY.AI Core version 4.5.0. Do not use the vacuum option.
ALTER TABLE ADD COLUMN does not work with geo column typeALTER TABLE ADD COLUMN for a geo column type in MapD 4.1 only partially adds the column. Any queries on that column will result in a system failure. If you encounter this issue, contact HEAVY.AI support at support@heavyai.com.
This issue was fixed in MapD version 4.1.1.
To prevent return of negative counts, set bigint-count = true in heavyai.conf.
HEAVY.AI does not currently support UPDATE from a subquery. For example, the following will not work:
UPDATE is not currently supported on variable-length data types.
On NVIDIA DGX systems, you might get the following error:
2021-05-14T15:56:16.571249 E 40446 0 DBHandler.cpp:403 Unable to instantiate CudaMgr, falling back to CPU-only mode. CUDA Error (999): unknown error
This error occurs if no fabric manager is installed on the system. To resolve the issue, install the fabric manager on the system.
In OmniSci 5.1.2, a deadlock can result if a render_vega request is executed at the same time as a table modification request (DROP/TRUNCATE/RENAME/APPEND TABLE) and the same table is referenced in both requests.
This is scheduled to be fixed in a future release. Until that time, HEAVY.AI recommends that you avoid executing a table modification request at the same time as a render_vega request against the same table.
When evaluating the new convert_meters_to_pixel_width and convert_meters_to_pixel_height extension functions for accuracy against circular polygons created with ST_Buffer in other packages, some errors are introduced by the extension functions somewhere at large zoom levels.
The resulting point/symbol sized by meters is just an approximate. It does not represent the exact area on the globe. There is more error in the approximate as you get closer to the poles in a mercator-projected view: a circle defined in meters should become egg-shaped, whereas the current symbol remains elliptical.
Workaround: If your clients are going to use these extension functions, HEAVY.AI recommends you use the legacysymbol vega mark type if the size in meters is large and zooming in close is useful for your analysis.
In Release 5.6.0, parameters do not work with geo joins.
You can import higher-precision timestamps (3, 6, 9, milli, micro, nano, rather than the default of seconds) via the data manager, but you cannot use them as a part of the actual queries or filters for a chart (as opposed to displaying them as results). For example, you cannot use a high-precision timestamp as the time dimension for a combo chart.
In MapD 4.0, dashboards can be shared only as 'read-only'. Users with whom a dashboard has been shared cannot currently make edits to the dashboard.
For security reasons, dashboard sharing does not automatically provide permissions on underlying tables/views. For now, this requires a one-time setup by a superuser/administrative to configure a group of users or a role with permissions on the underlying objects.
Dashboard sharing does not currently work in HEAVY.AI Cloud because each HEAVY.AI Cloud user currently has a dedicated HEAVY.AI instance. This limitation will be addressed in a future release.
If your OmniSci instance is set up to autoload specific dashboards on login by specifying the name in servers.json, you need to update the entry to use the dashboard ID instead. If you do not, your dashboards will not autoload. Find the dashboard ID by running `\dash` from omnisql, and then update the servers.json entry accordingly:
MapD Immerse versions 3.4.0 and higher work only with MapD Core versions 3.4.0 and higher.
Any update on Line or Histogram charts also starts an update query for range chart that is not required.
Binned columns and date extracts appear as JSON strings when exported to CSV format.
Old share links (for example, https://www.mapd.com/demos/ships/#/link/mapd/228ae04e\ no longer load and render all charts immediately. You must resize the browser or otherwise cause the page to re-render to see all charts.