Compiling and Building Cubes
This chapter describes how to compile and build cubes. It includes the following topics:
During the build process, users cannot execute queries. (However, if a query is currently running, you can build the cube.)
See Implementing InterSystems Business Intelligence for the following related topics:
Cube synchronization, which updates cubes incrementally (and permits queries to be executed at the same time). See the chapter “Keeping the Cubes Current” in the above listed book.
The InterSystems Business Intelligence Cube Manager utility, which you typically use for production systems. The Cube Manager creates automated tasks that build or synchronize cubes according to your specifications. See the chapter “Keeping the Cubes Current” in the above listed book.
The cube version feature, which enables you to modify a cube definition, build it, and provide it to users, with only a short disruption of running queries. See the appendix “Using Cube Versions” in the above listed book.
Also see “Accessing the Samples Shown in This Book,” in the first chapter.
When to Recompile and Rebuild
If you make any change to a cube class or a subject area class, you must recompile that class before those changes take effect. For many changes to a cube, you must also rebuild the cube before those changes take effect.
The following table lists the required actions after changes:
|Element Type||Type of Change||Required Actions|
|Cube (root element)||Edits to Name or Source class||Recompile and rebuild|
|Filter Value||Other changes that apply to the cube but not to specific elements in the cube||Recompile|
|Measure||Edits to the following options of an existing measure (many other elements have some or all of these common options).
|All other changes, including adding measures||Recompile and perform a selective build for the changed measure(s), or recompile and perform a full cube rebuild|
|Dimension (not a computed dimension)||Edits to the following options of an existing dimension:
|All other changes, including adding dimensions||Recompile and perform a selective build for the levels that make up the changed dimension(s), or recompile and perform a full cube rebuild|
|Computed dimension||All changes||Recompile|
|NLP dimension||All changes||Recompile|
|Hierarchy||Edits to the common options of an existing hierarchy (as listed in “Measure”)||Recompile|
|Hierarchy||All other changes, including adding and deleting hierarchies||Recompile|
|Level||Edits to the following options of an existing level:
|All other changes, including adding levels*||Recompile and perform a selective build for the changed level(s), or recompile and perform a full cube rebuild|
|Property||Edits to the following options of an existing property:
|Property||All other changes, including adding and deleting properties||Recompile and rebuild|
|Calculated member||All changes||Recompile|
|Named set||All changes||Recompile|
|Subject area||All changes||Recompile|
|Compound cube (a kind of subject area)||All changes||Recompile (after recompiling all cubes used in the compound cube)|
|Quality measure||All changes||Recompile the quality measure class|
|KPI or plug-in||All changes||Recompile the KPI or plug-in class|
*The current server locale determines the names of members of a time dimension. (See “Using the Locale to Control the Names of Time Members,” later in this book.) If you change the locale, it is necessary to recompile and rebuild the cube.
Compiling a Cube
To compile a cube class in the Architect:
The system starts to compile the class and displays a dialog box that shows progress.
If you have made changes that you have not yet saved, the system saves them before compiling the cube.
Or open the cube class in Studio and compile it in the same way that you compile other classes.
When you compile a cube class, the system automatically generates the fact table and all related classes if needed. If the fact table already exists, the system regenerates it only if it is necessary to make a structural change.
If there are any cached results for this cube, the system purges them.
Building a Cube
The phrase building a cube refers to two tasks: adding data to the fact table and other tables and building the indices used to access this data.
To perform a full cube build in the Architect:
The system displays a dialog box.
Note that the Build option may be greyed out. In this case, you must compile the cube before performing a build.
Optionally specify a value for Maximum Number of Records to Build.
By default, the system iterates through all records in the source table and builds the same number of records to the fact table. You can override this behavior when you build the cube. If you specify the Maximum Number of Records to Build option, the system iterates through only that number of records. The result is a smaller fact table that the system builds more quickly.
If the Maximum Number of Records to Build field is initialized with a number, that means that the cube class overrides the default behavior. (For details, see the maxFacts attribute for <cube> in the appendix “Reference Information for Cube Classes.”) In this case, you can either use the value provided by the cube class or enter a smaller value.
Select Build Everything in the Build Option section of the dialog box.
The system starts to build the cube and displays progress as it does so. The Compile button is deactivated (greyed out) for the duration of the build.Note:
Clicking the Close button during the build process does not interrupt the build process. If you choose to close the build dialog, you may reopen it at a later time to see the current state of the build process. If the build completes while the dialog is closed, the dialog will reappear to notify the user of build completion.
The cube is then available for use as described in Using the Analyzer.
Using Selective Build
You can use the Selective Build feature to build certain elements in a cube without rebuilding the entire cube and experiencing the attendant downtime. For example, if you recently made changes to a specific dimension or have source data changes that affect only one dimension, you can use Selective Build to build the relevant levels in that dimension. You can also use Selective Build to build a recently added level, measure, or relationship.
You can use Selective Build to build specific levels, measures, or relationships in a cube. More specifically, when you use Selective Build, columns in the fact tables (which each correspond to a level, measure, or relationship) are built. Using Selective Build on a cube does not trigger a need to update that cube’s dependent cubes.
A cube may inherit elements of its definition from another cube. When Selective Build is enabled for a cube inherited by another, the inheriting cube is able to read the designated factNumbers in the supercube definition and assign factNumbers to the subcube definition accordingly. The subcube does not assume that the factNumbers of the supercube remain the same, and therefore regenerates all of its own factNumbers. This protects the current cube from any changes in the supercube that might have been assigned a factNumber that conflicts with a compiled factNumber in the current cube.
Implications of Selective Build
When a selective build is taking place, only the cube elements that are being built are unavailable for queries. Selective Build and cube synchronization cannot happen simultaneously, so while the cube is not entirely inactive as in a full build, cube synchronization is unavailable while a selective build is taking place, and vice versa. Any significant build operation can block a planned synchronize. Furthermore, selective builds take longer than full builds, so budget time accordingly.
It’s good practice to synchronize your cube after a selective build, as Selective Build only performs updates to columns in fact tables and does not insert any new rows into the fact tables.
Multiple selective builds may run at the same time. In this case, each selective build will only build its selected cube elements. You can build multiple columns at once with Selective Build, but you cannot build any column more than once at the same time.
The system handles Selective Build errors the same way it handles errors for full cube builds.
Using Selective Build in the Architect
Selective Build is automatically enabled for all cubes. You must compile your cube before you can use Selective Build.
The following procedure provides an example of using the Selective Build feature:
Navigate to the Analyzer and open the HoleFoods cube.
In the Model Contents pane, expand the Outlet dimension, then expand the Region level. Drag the Region level over to the Rows area. Observe the resulting pivot table.
Next, open the Architect. Click the Region level of the Outlet dimension in the Model Viewer.
In the Details Area to the right, under Source Values, select Expression. Enter the following in the Expression text box:
%source.Outlet.Country.Region.%ID _ "-" _ %source.Outlet.Country.Region.Name
Compile the HoleFoods cube.
Click Build. When the Build Cube dialog appears, note that the system automatically detects that the [Outlet].[H1].[Region] level has changed and preselects a Selective Build for [Outlet].[H1].[Region]. Click Build.
Navigate back to the Analyzer. In the Model Contents pane, expand the Outlet dimension, then expand the Region level. Drag the Region level over to the Rows area. Observe that the resulting pivot table and note the differences for the Region level.
Building the Cube Programmatically
classmethod %BuildCube(pCubeName As %String, pAsync As %Boolean = 1, pVerbose As %Boolean = 1, pIndexOnly As %Boolean = 1, pMaxFacts As %Boolean = 0, pTracking As %Boolean = 0, ByRef pBuildStatistics As %String = "", pFactList As %String) as %Status
pCubeName is the logical name of the cube as given in its XData block; this is not case-sensitive.
pAsync controls whether the system performs the build in multiple background processes. If this argument is true, the system uses multiple processes and does not return until they are all complete. If this argument is false, the system uses a single process and does not return until it is complete.Note:
If you have specified the cube option Initial build order, the system ignores the value of pAsync and uses a single process to build the cube. These options are described in “Specifying Cube Options,” earlier in this book.WARNING:
Custom code for build processes must not invoke HALT. Terminating a DeepSee agent may cause a cascade of lower priority agent terminations and result in the build hanging due to a lack of available agents for new tasks.
pVerbose controls whether the method writes status information. If this argument is 1, the system writes status updates to the command line. (This argument does not affect whether the method writes build errors or other logging information.)
pIndexOnly controls whether the method only updates the indices. If this argument is 1, the system only updates the indices of the fact table.
pMaxFacts specifies the maximum number of rows in the fact table. This determines the number of rows of the base table that the system uses when building the cube.
If pMaxFacts is 0, the default, all rows of the base table are processed.
pTracking is for internal use.
pBuildStatistics returns an array of information about the cube build. This array has the following nodes:
Node Value pBuildStatistics(“elapsedTime”) Elapsed build time, in seconds. pBuildStatistics(“errors”) Number of build errors. pBuildStatistics(“factCount”) Number of facts that were built and indexed. pBuildStatistics(“missingReferences”) Number of missing references. pBuildStatistics(“expressionTime”) Time spent processing sourceExpressions to build the cube elements. pBuildStatistics(“iKnowTime”) Time spent building NLP indices.
pFactList is a list of specific Property names in the cube's fact class. If pFactList is supplied, the build will only update the columns listed in that fact list. This list can have either comma-delimited or $LB format. The specific facts being updated will be individually marked as unavailable for queries and queries referencing dimensions based on those facts will throw an error on prepare.
This method returns a status. If errors occur during the cube build, the status code indicates the number of build errors.
set status = ##class(%DeepSee.Utils).%BuildCube("patients")
This method writes output that indicates the following information:
Number of processors used.
Total elapsed time taken by the build process
Total amount of time spent evaluating source expressions, summed across all processors.
Building cube [patients] Existing cube deleted. Fact table built: 1,000 fact(s) (2 core(s) used) Fact indices built: 1,000 fact(s) (2 core(s) used) Complete Elapsed time: 1.791514s Source expression time: 0.798949s
If Source expression time seems too high, you should re-examine your source expressions to be sure that they are as efficient as possible; in particular, if the expressions use SQL queries, double-check that you have the appropriate indices on the tables that the queries use.
Cube Build Status
If there is a build in progress, you can monitor its progress using the %BuildStatus() method. From the IRIS Terminal call:
If there is no build in progress, %BuildStatus() displays the timestamp of the most recent build: There is no build in progress. Last build was finished on 06/23/2020 11:31:07.
Minimizing Cube Size During Development
While you are developing a cube, you typically recompile and rebuild it frequently. If you are using a large data set, you might want to limit the number of facts in the fact table, in order to force the cube to be rebuilt more quickly. To do this, do one of the following:
If you build the cube in the Architect, specify a value for Maximum Number of Records to Build.
Edit the cube class in Studio and add the maxFacts attribute to the <cube> element. See the appendix “Reference Information for Cube Classes.”
If you do so, be sure to remove this attribute before deployment.
Build the cube in the Terminal and specify the pMaxFacts argument. See “Building the Cube Programmatically” earlier in this chapter.
Note that all these options are ignored during a selective build.
Using Parallel Processing During a Cube Build
If all the following items are true, the system uses multiple cores to perform the build:
You specify pAsync as 1 when you build the cube (see “Building the Cube Programmatically”).
The source for a cube is a persistent class (rather than a data connector). Data connectors are described in Implementing InterSystems Business Intelligence.
The persistent class is bitmap-friendly.
The Initial build order option of the cube has not been set. These options are described in “Specifying Cube Options,” earlier in this book.
When you build a cube asynchronously, the system sets up %SYSTEM.WorkMgr agents to do the work, if it is possible to use parallel processing.
These agents are also used to execute queries.
On rare occasions, you might need to reset these agents. To do so, use the %Reset() method of %DeepSee.Utils. This method also clears any pending tasks and clears the result cache for the current namespace, which would have an immediate impact on any users. This method is intended for use only during development.
When you build a cube, pay attention to any error messages and to the number of facts that it builds and indexes. This section discusses the following topics:
Fact count, which is a useful indicator of build problems in all scenarios
For more information on troubleshooting options, see the InterSystems Developer Community.
Seeing Build Errors
When you build a cube in the Architect or in the Terminal, the system indicates if there are any build errors but does not show all of them. To see all the recorded build errors, do either of the following:
Look for the log file install-dir/mgr/DeepSeeUpdate_cube_name_NAMESPACE.log, where cube_name is the name of the cube, and NAMESPACE is the namespace in which this cube is defined.
The time stamp in this file uses $NOW to write the local date and time, ignoring daylight saving time.
Where cubename is the logical name of the cube, in quotes.
This method displays information about all build errors. For example (with added line breaks):
SAMPLES>do ##class(%DeepSee.Utils).%PrintBuildErrors("holefoods") 1 Source ID: 100000 Time: 05/09/2019 14:12:52 ERROR #5002: ObjectScript error: <DIVIDE>%UpdateFacts+106^HoleFoods.Cube.Fact.1 2 Source ID: 200000 Time: 05/09/2019 14:12:41 ERROR #5002: ObjectScript error: <DIVIDE>%UpdateFacts+106^HoleFoods.Cube.Fact.1 3 Source ID: 300000 Time: 05/09/2019 14:13:13 ERROR #5002: ObjectScript error: <DIVIDE>%UpdateFacts+106^HoleFoods.Cube.Fact.1 ... 10 build error(s) for 'holefoods'
In some cases, the system might not generate an error, so it is important to also check the fact count as discussed in the next section.
Checking the Fact Count
When you build a cube, the system reports the number of facts that it builds and indexes.
Each fact is a record in the fact table. The fact table should have the same as the number of records in the base table, except in the following cases:
You limit the fact count as discussed earlier in this chapter.
The cube class also defines the %OnProcessFact() callback, which you can use to exclude records from the cube. See “Using Advanced Features of Cubes and Subject Areas” in Advanced Modeling for InterSystems Business Intelligence.
Also, when the system builds the indices, the index count should equal the number of records in the fact table. For example, the Architect should show the same number for Building facts and for Building indices. If there is a discrepancy between these numbers, check the log files.
Possible Causes of Build Errors
Examine any levels that use range expressions, and verify that these levels do not drop records. See “Validating Your Levels,” later in this book.
An error of this kind affects the index count but not the fact count.
Try disabling selected dimensions or measures. Then recompile and rebuild to isolate the dimension or measure that is causing the problem.
In some cases, the build log might include errors like the following:
ERROR #5002: ObjectScript error: <STORE>%ConstructIndices+44^Cube.cube_name.Fact.1
This error can occur when a level has a very large number of members. By default, when the system builds the indices, it uses local memory to store the indices in chunks and then write these to disk. If a level has a very large number of members, it is possible to run out of local memory, which causes the <STORE> errors.
To avoid such errors, try either of the following:
Build the cube with a single process. To do so, use %BuildCube() in the Terminal, and use 0 for its second argument.
In the <cube> element, specify bitmapChunkInMemory="false" (this is the default). When this cube is built using background processes, the system will use process-private globals instead of local variables (and will not be limited by local memory).
Missing Reference Errors
If your cubes have relationships to other cubes, the build log might include errors like the following:
ERROR #5001: Missing relationship reference in RelatedCubes/Patients: source ID 1 missing reference to RxHomeCity 4
This error can mean that you have built the cubes in the wrong order. See “Building Cubes That Have Relationships” in Advanced Modeling for InterSystems Business Intelligence. Note that if you use the Cube Manager, the Cube Manager determines an appropriate build order.
The missing relationship reference error can also occur when new source data becomes available during the cube build process — that is, after only some of the cubes have been built. For example, consider the sample cubes RelatedCubes/Cities and RelatedCubes/Patients (which are available in the SAMPLES namespace). Suppose that you build the cube RelatedCubes/Cities, and after that, the source table for RelatedCubes/Patients receives a record that uses a new city. When you build the cube RelatedCubes/Patients, there will be a missing relationship reference error.
If you are certain that you have built the cubes in the correct order, see the next section for information on recovering from the errors.
Recovering from Build Errors
The system provides a way to rebuild only the records that previously generated build errors, rather than rebuilding the entire cube. To do this:
Correct the issues that cause these errors.
Where cubename is the logical name of the cube, in quotes. This method accepts a second argument, which specifies whether to display progress messages; for this argument, the default is true.
Fact '100' corrected Fact '500' corrected Fact '700' corrected 3 fact(s) corrected for 'patients' 0 error(s) remaining for 'patients'
Or rebuild the entire cube.
Business Intelligence Task Log
The system creates an additional log file (apart from the previously described build logs). After it builds the cube or tries to build the cube, the system also writes the DeepSeeTasks_NAMESPACE.log file to the directory install-dir/mgr. You can use the %SetLoggingOptions method of the %DeepSee.WorkMgr class to turn on logging for background agents that the system used during the build process. To do so, make a call like the following:
To see this file from the Management Portal, select Analytics > Admin > Logs.
This file also contains information about runtime errors of various kinds such as listing errors and KPI errors.
The time stamps in this files use the local date and time (taking daylight saving time into account).