Managing data cubes

This topic explains how to create and configure a data cube.

About Pivot 2.0 and Pivot SQL data cubes

A data cube can be one of two types: Pivot 2.0 or Pivot Classic. The manner in which the types are enabled and accessed differs slightly between new Imply deployments that are new as of 2022.01 and deployments that have been upgraded to 2022.01 from a previous version. The differences are as follows:

  • For new installations, Pivot 2.0 and Pivot SQL are enabled by default. An administrator can still disable these features using a feature flag. Disabling them results in new data cubes using Pivot Classic by default.
  • For existing installations that have been upgraded from a previous version, there are no changes to existing data cubes or dashboards. Administrators can enable Pivot 2.0 and Pivot SQL as feature flags. If enabled, users have the ability to choose whether to use Pivot 2.0 or Pivot Classic when creating data cubes.

Note that Plywood cubes and Pivot Classic are deprecated and will be no longer supported in a future release of the product.

Create a data cube

To create a new data cube, from the home view click the Create new data cube link from the Imply home page. The Create new data cube modal allows you to define the source data for a data cube and initial data cube settings:

Add cube modal

When creating data cubes, depending on how your administrator has configured Pivot, you may have the ability to choose whether to make the data cube a SQL data cube. This enables Pivot SQL features in the data cube, such as the ability to specify dimensions and measures using Pivot SQL rather than the Plywood expression language.

Checking the Auto fill dimensions and measures checkbox enables automatic detection of the data schema, and your data cube is populated with default dimensions and measures. Alternatively, you can create them manually.

Another way to create a new data cube is to duplicate an existing data cube and edit its properties. To do so, click the Duplicate button from the Edit data cube page of the source data cube.

After creating the data cube, the edit screen appears, where you can adjust configuration settings, if desired.

Convert data cubes to Pivot SQL

In certain circumstances, you can convert an existing, non-SQL data cube into a SQL data cube.

Conversion to SQL is not possible for data cubes that use advanced options like altDataSource or data cubes that have individual dimensions or measures that can’t be converted to SQL for any reason. If you attempt to convert data cubes in such cases, an error dialog appears that indicates why the data cube can't be converted, including the expression or option that prevents conversion.

Note that converting a data cube to SQL can’t be reversed and may have unexpected consequences. As a backup measure, you should duplicate the data cube you intend to convert before attempting the operation on the original or copied data cube.

To convert a data cube to SQL, follow these steps:

  1. Open the data cube settings by clicking the information icon in the data cube view and then clicking Edit.

  2. Click the Advanced tab to open advanced settings for the data cube.

  3. Click the Convert to SQL button at the bottom of the page.

  4. Confirm the operation by clicking Convert:

    Convert data cube

You can now use SQL expressions to compose dimensions and formulas, as described next.

Compose SQL expressions

Once a data cube is enabled for Pivot SQL, you can use SQL expressions to define custom dimension and measure formulas, as shown in the Formula field in the following figure:

SQL expression

The syntax for querying columns follows the t.”columnName” format. Inline help in the data cube UI provides troubleshooting information in the event of syntax errors.

For more information about custom dimensions and measures in data cubes, including examples, see Measures and Managing data cubes. Also see Druid SQL for a complete list of supported SQL functions.

Note the following:

  • Pivot SQL supports subset filters, filter tokens, PII masks
  • Subset filters can be expressed as SQL; SQL support for filter tokens will be available in the future
  • Data cube APIs support both SQL and Plywood on SQL data cubes; SQL data cubes can be created using the property: queryMode: “sql”
  • All existing security constraints on user data access applies to SQL queries as well.

Also see the following limitations.

Limitations

The following limitations apply to the use of SQL in Pivot data cubes:

  • altDataSource (also known as native Druid UNION queries) are not currently supported
  • Support for lookups is provided only via the LOOKUP Druid SQL function
  • MilliSecondsInInterval magic constant for rate calculations is not currently supported
  • Resplit measures (e.g., daily active users) must still be expressed using Plywood
  • Druid functions requiring extensions aren't available unless the extension is loaded. For example, using STDDEV requires the stats extension to be loaded.

Edit a data cube

Click the pencil icon in the data cube header to edit a data cube.

Cube edit view

Within the edit view you can change the title, description, and default timezone of the data cube. You can also edit and create dimensions and measures from their respective tabs.

Schema detection

Pivot will automatically detect the columns in your data source and recommend some possible dimensions and measures.

Schema detection can be helpful when you are creating a new data cube in which case it will allow you to initialize the data cube with some dimensions and measures.

The other use for schema detection is when you evolve your schema and add columns over time. If you have just added a column and you want a corresponding dimension to be created for it, you can use schema detection to automatically construct that column for you. This functionality is accessible from the Suggestions tab of the New dimension and New measure modals.

Cube edit new dimension modal

How schema detection works

Pivot looks at the dataset metadata and uses the returned list of columns, their types, and their aggregation (in case of rollup) to determine what dimensions and measures to suggest.

Pivot generates dimensions and measures by applying the following rules to the discovered underlying column types:

  • Time columns get mapped to a dimension with automatic bucketing by default.
  • String columns get mapped to a dimension.
  • Numeric columns get mapped to a SUM measure or an otherwise appropriate measure if the column is marked as being aggregated as part of rollup.
  • ApproximateHistogram columns get mapped as a 98th percentile measure.
  • HyperUnique and ThetaSketch columns get mapped to a countDistinct measure.

Limitations

While schema detection is an invaluable tool for quickly getting you up and running with a new data source, it can never do a perfect job. Once you create a data cube you should play with it and tailor it to be perfectly suited to your needs. Don't hesitate to change and delete the auto-generated dimensions and measures; they will always be there in the Suggestions tab.

In particular, schema detection cannot detect these common scenarios:

  • String columns that you might want to see as countDistinct - number of unique values.
  • The perfect granularities to apply to time and numeric dimensions.
  • Lookups that you might want to apply to certain dimensions.
  • Dimensions that actually correspond to a URL.
  • Measures that are interesting when filtered on something.
  • Measures that should be seen as a ratio, or some other post aggregation.

Defaults

The default settings control what options are selected in the data cube view when a Pivot user first navigates to the data cube. By default, the data cube shows data from the latest day and the first measure in the list of available measures for that data cube.

You can modify these defaults, add filtering conditions, or set a specific dimension to be shown with the default settings.

To create a default view:

  1. Create the view you want to save as your default view.
  2. Click the Options menu in the top right.
  3. Click Update data cube.
  4. Click Update data cube defaults.
  5. Click Save current view as default.
  6. Click Save for all users.

Create a time bucket

The time filtering option includes several default time buckets, such as 1 hour, 6 hours, 1 day, and so on. You can create custom time buckets that appear in the time filtering interface.

To create custom time buckets, open the Defaults tab from the data cube edit page. In the Default shown dimensions field, add an expression that uses the timeBucket function, as follows:

Time buckets

The timeBucket function accepts an ISO 8601 duration expression, and creates the duration that you specify. The ISO 8601 expression for one hour is PT1H. You can use timeBucket in a JSON string like this:

[{"dimension":"__time","bucketAction":{"op":"timeBucket","duration":"PT1H"},
"sortType":"dimension","direction":"ascending"}]

This results in a line chart with markers for the total number of events in every hour of the latest day.

If you need help creating other ISO duration expressions to specify temporal intervals other than one hour, there are online tools to generate these expressions for you.

Advanced options

The following advanced options are available when you expand the More chevron.

Data cube advanced

Subset filter formula: Specify a plywood formula that applies a mandatory, hidden filter to all queries made through this data cube. See row level restriction.

Required filter token: Specify a token that corresponds to a filter token defined in a user role. See token based row level access control.

Primary time dimension: Specify the time dimension that corresponds to when the data point was actually added. This is used for checking the freshness of data and also potentially enforcing a time filter. Defaults to Time.

Enforce time filter: If set to true, ensures that every query is filtered on the primary time dimension. This should be used if time unbounded queries are likely to be slow due to the volume and tearing of data.

Instances: Configure multiple filtered instances within a single data cube via a configuration object. See example configuration.

Refresh time: Configure how the data cube will calculate the data's latest time. Defaults to Query, which means it will query the data to learn about the latest data ingested. This can also be set to predefined in which case latest current time will always be assumed. An alternative 'fixed' time can also be defined as the 'fixedMaxTime' property in the options.

Query caching: Specify query caching behavior. Allowing caching can greatly speed up exploration but can also cause results to be a little out of date especially in realtime rolled up datasets.

Dimension and measure formulae visibility: You can hide formulae for dimensions and measures that appear in the info box of dimensions and measures. Formulae are shown by default. An administrator can select Hide formulae when the formulae might reveal information that is too sensitive for all Pivot users in the organization to view. When the Hide formulae option is selected, an administrator can still view the formula for a dimension or module in its Settings view.

Data cube options

Additional data cube configuration options as JSON with any of the following properties:

  • fixedMaxTime: Specify a fixed time value for data cubes to use as the latest "max time" instead of querying for it or using the current time. This configuration should be used in conjunction with the Refresh time setting set to 'fixed'.
  • timeoutOverride: Override a connection-wide timeout for queries made through this data cube.
  • exactResultsOnly: If true, disallows approximate aggregators and force Pivot to make groupBy over topN queries.
  • customAggregations: Map any Druid aggregation to a Pivot measure. Aggregation definitions should be keyed on measure names.
  • customTransforms: Map any Druid extraction function function to a Pivot dimension. Extraction function definitions should be keyed on dimension names.
  • crossTabBatchSize: Configure the number of cross tab rows loaded at once (defaults to 80).
  • piiMask: Define a threshold below which data is considered Personally identifiable information (PII) and hidden. See the piiMask example.
  • restrictedModeProperties: Define a restricted edit mode for access control which grants users limited edit functionality for a data cube. See the restrictedModeProperties example.
Time filter presets

Each section of presets in the time filter menu can be configured with custom presets. See an example of latest presets.

  • latestPresets: An array of preset objects that will override the LATEST DATA presets in the time filter menu.
  • currentPresets: An array of preset objects that will override the CURRENT presets in the time filter menu.
  • previousPresets: An array of preset objects that will override the PREVIOUS presets in the time filter menu.
Querying options
  • splitLimits: Provide an array of numbers that will override the default options of [5, 10, 25, 50, 100] shown in the limit dropdown of the split menu.
  • ignoreDimensionInExplain: Provide a list of dimensions to ignore when computing an explanation.
  • rankExpression: Configure the measure whose value is used to order dimension values in the dimension filter menu. Defaults to count.
  • druidContext: An object that will be passed in as the Druid query context.
  • priority: A number that will be passed as 'priority' into the Druid query context.
  • maxDownloadLimit: Configure a limit (in number of rows) to downloads.
  • largeDownloadInterval: Configure a limit (as a time interval) to downloads.
Visual options
  • suppressOverallByDefault: If true, hides the "overall" element of a visualization by default.
  • boostPrefixRank: If true, will "boost" dimension values where matches occur at the beginning of the string over values with matches found in other positions in the dimension filter menu.
  • boostSelectedMeasures: If true and the data cube view is in multi-measure selection mode, will "boost" selected measures to the top of the measures list in the measure picker menu.
  • grayOutTrivial: If true, renders "trivial" dimensions in gray in the dimensions panel. A continuous dimension is trivial if it has a max value equal to its min value. A discrete dimension is trivial if it has less than 2 values.
  • alwaysShowCompareControl: If true, will remove the ability to toggle time compare.
  • disablePreview: If true, disables the preview of values in the dimension and measure modals.
  • showExplainBadge: If true, shows a badge with the standard deviation for the sample values used for the explanation.

Examples

Instances
{
  "dimensionFormula": "$language",
  "values": [
    {
      "title": "English",
      "value": "en"
    },
    {
      "title": "Chinese",
      "value": "zh"
    },
    {
      "title": "Spanish",
      "value": "es"
    }
  ]
}

dimensionFormula is a plywood formula that operates on a dimension.

values is a list of objects with keys of title, value where value is the value that will be matched with the dimensionFormula expression and title is the label displayed in the UI.

piiMask
{
  "piiMask": {
    "aggregate": "$main.count()",
    "threshold": 10,
    "maskText": "HIDDEN [%i]",
    "otherText": "Other (%i)",
    "hideAggregates": false
  }
}

Note that %i will be resolved as the row index of the datum

latestPresets

latestPresets appear as time filter values in the data cube view. You can create presets for custom time values, as in the following example:

{
  "latestPresets": [
    { "name": "1M", "pillLabel": "1 minute", "selection": "$m.timeRange('PT1M', -1)" },
    { "name": "5M", "pillLabel": "5 minutes", "selection": "$m.timeRange('PT1M', -5)" },
    { "name": "30M", "pillLabel": "30 minutes", "selection": "$m.timeRange('PT1M', -30)" },
    { "name": "1H", "pillLabel": "1 hour", "selection": "$m.timeRange('PT1H', -1)" },
    { "name": "6H", "pillLabel": "6 hours", "selection": "$m.timeRange('PT1H', -6)" }
  ]
}

In the UI, the value of name appears as the UI label in the presets dropdown, while the optional pillLabel appears as the label when the value is selected. If pillLabel is not provided, name is used in both places. The selection value is the interval of time selected. The sample presets above would appear as follows:

Data cube latest time presets

In the selection value, $m is resolved as the data's maximum time and $n as the current time.

Since $m is a time value, if you would like to add a preset to view all data for the last seven days, with full data for each day, you would need to adjust $m as shown in the following example:

{
  "latestPresets": [
    { "name": "7D", "pillLabel": "7 days", "selection": "$m.timeFloor('P1D').timeRange('P1D', -7)" }
  ]
}
restrictedModeProperties

When set, this property will expose an additional access control list configuration in the data cube's "Access" section, enabling the ability to grant users and/or roles a subset of full edit functionality.

{
  "restrictedModeProperties": [
    "dimensions",
    "measures"
  ]
}

The configuration above would allow users and roles granted "Restricted edit access" on the "Access" screen the ability to edit the data cube's dimensions and measures, but nothing else.

The values that can be set for this array are:

  • "title": The data cube name
  • "description": The data cube description
  • "theme": The data cube theme
  • "dimensions": Dimensions
  • "measures": Measures
  • "group": Dimension groups
  • "defaultDuration": Default time filter duration
  • "defaultSelectedMeasures": Default selected measures
  • "defaultPinnedDimensions": Default pinned dimensions
  • "defaultRefreshRate": Default refresh rate
  • "defaultTimezone": Default time zone
  • "defaultFilter": Default filter
  • "defaultSplits": Default shown dimensions
  • "subsetFormula": Subset filter formula
  • "filterTokens": Required filter token
  • "specialTimeDimension": Primary time dimension
  • "enforceTimeFilter": Enforce time filter
  • "instances": The possible instances for the data cube
  • "refreshTimeWith": Refresh time
  • "queryCaching": Query caching
Last updated on 1/24/2022
← About data cubesVisualizations →