Configuration Guide


  1. Properties
    1. Property list
  2. Connect strings
    1. Syntax
    2. Connect string properties
  3. Cache management
    1. Schema cache
  4. Memory management
    1. Out of memory
  5. Logging
    1. Configuring log4j within Mondrian's test environment
    2. MDX and SQL Statement Logging

1. Properties 

Mondrian has a properties file to allow you to configure how it executes. The file is loaded when the executing Mondrian JAR detects it needs properties, but can also be done explicitly in your code. It looks in several places, in the following order:

  1. In the directory where you started your JVM (Current working directory for JVM process, java.exe on Win32, java on UNIX/Linux).
  2. If there isn't under current working directory of JVM process, Class MondrianProperties's classloader will try to locate in all of its classpaths. So you may put under /WEB-INF/classes when you pack Mondrian into a Java web application. The demonstration web applications have this configuration.

These properties are stored as system properties, so they can be set during JVM startup via -D<property>=<value>.

1.1 Property list 

The following properties in effect the operations of Mondrian.

Not all of the properties in this table are of interest to the end-user. For example, those in the 'Testing' are only applicable if are running Mondrian's suite of regression tests.

Property Type Default value Description

mondrian.foodmart.jdbcURL string "jdbc:odbc:Mondrian FoodMart" Property containing the JDBC URL of the FoodMart database. The default value is to connect to an ODBC data source called "MondrianFoodMart".
mondrian.query.limit int 40 Maximum number of simultaneous queries the system will allow.

Oracle fails if you try to run more than the 'processes' parameter in init.ora, typically 150. The throughput of Oracle and other databases will probably reduce long before you get to their limit.

mondrian.jdbcDrivers string mondrian.jdbcDrivers,
oracle.jdbc.Oracle Driver,
A list of JDBC drivers to load automatically. Must be a comma-separated list of class names, and the classes must be on the class path.
mondrian.result.highCardChunkSize int 1 When reading high cardinality dimension elements, number of elements read from database in each step.
When dealing with high cardinality dimensions, elements are retrieved in blocks: first N elements are retrieved, when these elements are read, next ones (from N+1 to 2N) are got. N number is this property.
Setting great values for this property increases performance but may overload memory. Values should be prime with mondrian.result.limit
mondrian.result.limit int 0 If set to a value greater than zero, limits the maximum size of a result set.

If a query exceeds the limit, you will get an error such as:

Mondrian result limit exceeded: Mondrian Error: Size of CrossJoin result (53,463) exceeded limit (50,000)


Number of members to be read exceeded limit 50,000

and Mondrian throws a mondrian.olap.ResourceLimitExceededException. See also limit properties.

mondrian.rolap.evaluate.MaxEvalDepth int 10 Maximum number of passes allowable while evaluating an MDX expression. If evaluation exceeds this depth (for example, while evaluating a very complex calculated member), Mondrian will throw an error.
mondrian.rolap.SparseSegmentValueThreshold int 1,000 The values of the mondrian.rolap. SparseSegment ValueThreshold (countThreshold) and mondrian.rolap. SparseSegment DensityThreshold (densityThreshold) properties determine whether to choose a sparse or dense representation when storing collections of cell values in memory.

When storing collections of cell values in memory, Mondrian has to choose between a sparse and a dense representation, based upon the possible and actual number of values. The density is defined by the formula

density = actual / possible

Mondrian uses a sparse representation if

(possible - countThreshold) * densityThreshold > actual

For example, at the default values (countThreshold = 1000 and densityThreshold = 0.5), Mondrian use a dense representation for

  • (1000 possible, 0 actual), or
  • (2000 possible, 500 actual), or
  • (3000 possible, 1000 actual).

Any fewer actual values, or any more possible values, and Mondrian will use a sparse representation.

mondrian.rolap.SparseSegmentDensityThreshold double 0.5 See mondrian.rolap.SparseSegmentValueThreshold.
mondrian.olap.triggers.enable boolean true Whether to notify the Mondrian system when a property value changes.

This allows objects dependent on Mondrian properties to react (that is, reload), when a given property changes via, say,



MondrianProperties.instance().QueryLimit.set(50); boolean false Controls whether the MDX parser resolves uses case-sensitive matching when looking up identifiers.
mondrian.rolap.compareSiblingsByOrderKey boolean false Controls whether sibling members are compared according to the order key value fetched from their ordinal expression. The default value of false means ordinal expressions are only used for ORDER BY, and the actual values produced are ignored inside of Mondrian. Changing this to true allows Mondrian to correctly order members when native filtering is used, but requires that the values produced by the DBMS are non-null instances of java.lang.Comparable which yield the desired ordering when their Comparable.compareTo method is called.
mondrian.rolap.localePropFile string null
Name of the class whose resource bundle is to be used to for this schema. For example, if the class is com.acme.MyResource, mondrian will look for a resource bundle called com/acme/MyResource_<locale>.properties on the class path. (This property has a confusing name because in a previous release it actually held a file name.)

Used for the LocalizingDynamicSchemaProcessor; see Internationalization for more details.

mondrian.rolap.queryTimeout int 0 If set to a value greater than zero, limits the number of seconds a query executes before it is aborted.

If a query exceeds the limit, mondrian throws a mondrian.olap.QueryTimeoutExceptiony, resulting in an error such as:

Query timeout of 10 seconds reached

See also limit properties.

mondrian.rolap.nonempty boolean false If true, each query axis implicit has the NON EMPTY option set (and in fact there is no way to display empty cells).
mondrian.rolap.ignoreInvalidMembers boolean false If set to true, during schema load, invalid members are ignored and will be treated as a null member if they are later referenced in a query.
mondrian.rolap.ignoreInvalidMembersDuringQuery boolean false If set to true, during query validation, invalid members are ignored and will be treated as a null member.
mondrian.olap.NullMemberRepresentation string #null Defines how a null Member is represented in the result output. MSAS 2000 shows empty value while MSAS 2005 shows "(null)"
mondrian.rolap.IterationLimit int 0 If > 0, the maximum number of iterations allowed when evaluating an aggregate. If 0, there is no limit.

If a query exceeds the limit, mondrian throws a mondrian.olap.ResourceLimitExceededException, resulting in an error such as:

Number of iterations exceeded limit of 100

See also limit properties. int 0 If a crossjoin input list's size is larger than this property's value and the axis has the "NON EMPTY" qualifier, then the crossjoin non-empty optimizer is applied. Setting this value to '0' means that for all crossjoin input lists in non-empty axes will have the optimizer applied. On the other hand, if the value is set larger than any possible list, say Integer.MAX_VALUE, then the optimizer will never be applied.
mondrian.olap.NullOrZeroDenominatorProducesNull boolean false If a division has a non-null numerator and a null denominator, it evaluates to "Infinity", which conforms to MSAS behavior. However, the old semantics of evaluating this to NULL(non MSAS conforming), is useful in some applications. This properties controls whether the result should be NULL when denominator is NULL or zero.
mondrian.olap.elements.NeedDimensionPrefix boolean false

Property determines if elements of dimension (levels, hierarchies, members) need to be prefixed with dimension name in MDX query. For example when the property is true, the following queries will error out. The same queries will work when this property is set to false.

  • select {[M]} on 0 from sales
  • select {[USA]} on 0 from sales
  • select {[USA].[CA].[Santa Monica]} on 0 from sales

When the property is set to true, any query where elements are prefixed with dimension name as below will work

  • select {[Gender].[F]} on 0 from sales
  • select {[Customers].[Santa Monica]} on 0 from sales

Please note that this property does not govern the behaviour where in

  • [Gender].[M]
  • is resolved into a fully qualified

  • [Gender].[All Gender].[M]
  • In a scenario where the schema is very large and dimensions have large number of members a MDX query that has a invalid member in it will cause mondrian to to go through all the dimensions, levels, hierarchies, members and properties trying to resolve the element name. This behaviour consumes considerable time and resources on the server. Setting this property to true will make it fail fast in a scenario where it is desirable

    mondrian.olap.agg.IgnoreMeasureForNonJoiningDimension boolean false If there are unrelated dimensions to a measure in context during aggregation, the measure is ignored in the evaluation context. This behaviour kicks in only if the CubeUsage for this measure has IgnoreUnrelatedDimensions attribute set to false

    For example, Gender doesn't join with [Warehouse Sales] measure.

    With mondrian.olap.agg.IgnoreMeasureForNonJoiningDimension=true

    Warehouse Sales gets eliminated and is ignored in the aggregate value

    SUM({Product.members * Gender.members})
    [Store Sales] + [Warehouse Sales] = 7,913,333.82

    With mondrian.olap.agg.IgnoreMeasureForNonJoiningDimension=false

    Warehouse Sales is part of the aggregate value.

    SUM({Product.members * Gender.members})
    [Store Sales] + [Warehouse Sales] = 9,290,730.03

    On a report where Gender M, F and All members exist a user will see a large aggregated value compared to the aggregated value that can be arrived at by suming up values against Gender M and F. This can be confusing to the user. This feature can be used to eliminate such a situation.

    mondrian.rolap.SolveOrderMode string absolute

    Controls solve_order evaluation behavior.

    Valid values are:

    • absolute - The SOLVE_ORDER value is absolute regardless of where it is defined; e.g. a query defined calculated member with a SOLVE_ORDER of 1 always takes precedence over a cube defined value of 2.
    • scoped - Cube calculated members are resolved before any session scope calculated members, and session scope members are resolved before any query defined calculation. The SOLVE_ORDER value only applies within the scope in which it was defined.

    The default behavior (scoped) is compatible with Microsoft SQL Server Analysis Services 2005 and later. Use {@code absolute} to be compatible with Analysis Services 2000 and Mondrian 3.0 and earlier.

    mondrian.native.NativizeMaxResults int 150000

    Limits the maximum size of a NativizeSet result set.

    If a query exceeds this limit you will get a LimitExceededDuringCrossjoin error.

    mondrian.native.NativizeMinThreshold int 100000

    Disables native evaluation of a NativizeSet espression when the expression's expected cardinality falls below this value.

    mondrian.test.Name string null Property which determines which tests are run. This is a Java regular expression. If this property is specified, only tests whose names match the pattern in its entirety will be run.
    mondrian.test.Class string - Property which determines which test class to run. This is the name of the class which either implements interface  junit.framework.Test or has a method public static junit.framework.Test suite().
    mondrian.test.connectString string - Property containing the connect string which regresssion tests should use to connect to the database.

    See the connect string specification for more details.

    mondrian.test.QueryFilePattern string - (not documented)
    mondrian.test.QueryFileDirectory string - (not documented)
    mondrian.test.Iterations int 1 (not documented)
    mondrian.test.VUsers int 1 (not documented)
    mondrian.test.TimeLimit int 0 The time limit for the test run in seconds. If the test is running after that time, it is terminated.
    mondrian.test.Warmup boolean false Whether this is a "warmup test".
    mondrian.catalogURL string - The URL of the catalog to be used by CmdRunner and XML/A Test.
    mondrian.test.ExpDependencies int 0 Whether to test operators' dependencies, and how much time to spend doing it.

    If this property is positive, Mondrian's test framework allocates an expression evaluator which evaluates each expression several times, and makes sure that the results of the expression are independent of dimensions which the expression claims to be independent of.
    mondrian.test.random.seed int 1234 Seed for random number generator used by some of the tests.

    Any value besides 0 or -1 gives deterministic behavior. The default value is 1234: most users should use this. Setting the seed to a different value can increase coverage, and therefore may uncover new bugs.

    If you set the value to 0, the system will generate its own pseudo-random seed.

    If you set the value to -1, Mondrian uses the next seed from an internal random-number generator. This is a little more deterministic than setting the value to 0.

    mondrian.test.jdbcURL string - Property containing the JDBC URL of a test database. There is no default value.
    mondrian.test.jdbcUser string - Property containing the JDBC user of a test database. The default value is null, to cope with DBMSs that don't need this.
    mondrian.test.jdbcPassword string - Property containing the JDBC password of a test database. The default value is null, to cope with DBMSs that don't need this.
    mondrian.test.WarnIfNoPatternForDialect string NONE Property which controls if warning messages should be printed if a sql comparison tests do not contain expected sqls for the specified dialect. The tests are skipped if no expected sqls are found for the current dialect.

    Possible values are the following:

    • "NONE": no warning (default)
    • "ANY": any dialect
    • "ACCESS"
    • "DERBY"
    • "LUCIDDB"
    • "MYSQL"
    • ... and any Dialect enum in SqlPattern.Dialect

    Specific tests can overwrite the default setting. The priority between test setting and property file setting is:

    Settings besides "NONE" in file < Any setting in the test < "ANY"

    Aggregate tables
    mondrian.rolap.aggregates.Use boolean false Whether to use aggregate tables.

    If true, then Mondrian uses aggregate tables. This property is queried prior to each aggregate query so that changing the value of this property dynamically (not just at startup) is meaningful.

    Aggregates can be read from the database using the mondrian.rolap. aggregates.Read property but will not be used unless this property is set to true.

    mondrian.rolap.aggregates.optimizePredicates boolean true Determines whether Mondrian optimizes predicates or not

    If set to true, Mondrian intelligently optimizes certain predicates.

    If set to false, then predicates are optimized only when all the members of a dimension are included.


    For the in MDX query which selects all but one member of 1997 quarter:

    select {[Time].[1997].[Q1],[Time].[1997].[Q2],[Time].[1997].[Q3]} on 0 from sales predicate optimization would result in sql that retrives all 4 quarters, instead of just 3. The fourth quarter is cached along with others for future use:

    ...where`time_by_day`.`the_year` = 1997 group by `time_by_day`.`the_year`, `time_by_day`.`quarter`...

    (note the where in clause involving quarters is missing)

    when predicate optimization is disabled, Mondrian would retrive no more data than specified by the query, In the above case it would retrive only data for first 3 quarters:

    ...`time_by_day`.`the_year` = 1997 and `time_by_day`.`quarter` in ('Q1', 'Q2', 'Q3') group by `time_by_day`.`the_year`, `time_by_day`.`quarter`...

    mondrian.rolap.aggregates.Read boolean false Whether to read aggregate tables.

    If set to true, then Mondrian scans the database for aggregate tables. Unless mondrian.rolap. aggregates.Use is set to true, the aggregates found will not be used.

    mondrian.rolap.aggregates.ChooseByVolume boolean false Whether to choose an aggregate tables based volume or row count.

    If true, Mondrian uses the aggregate table with the smallest volume (number of rows multiplied by number of columns); if false, Mondrian uses the aggregate table with the fewest rows.

    mondrian.rolap.aggregates.rules string "/DefaultRules.xml"
    (which is in the mondrian.rolap.aggmatcher package in mondrian.jar)
    Name of the file which defines the rules for recognizing an aggregate table.

    Can be either a resource in the Mondrian jar or a URL. See aggregate table rules for details.

    Normally, this property is not set by a user.

    mondrian.rolap.aggregates.rule.tag string default The AggRule element's tag value.

    Normally, this property is not set by a user.
    mondrian.rolap.aggregates.generateSql boolean false Whether to print the SQL code generated for aggregate tables.

    If set, then as each aggregate request is processed, both the lost and collapsed dimension create and insert sql code is printed. This is for use in the CmdRunner allowing one to create aggregate table generation sql.
    mondrian.rolap.aggregates.jdbcFactoryClass string null If specified, overrides the default StdFactory factory class which specifies the JdbcSchema class to use for determining the tables and columns of a data source. The JdbcSchema class is used as part aggregate table matcher algorithm.

    Caching boolean false Whether to clear a RolapStar's data cache after each query.

    If true, RolapStar does not cache aggregate data from one query to the next: the cache is cleared after each query.

    mondrian.expCache.enable boolean true Controls whether to use a cache for the results of frequently evaluated expressions.

    With the cache disabled, an expression like:

    Rank([Product]. CurrentMember,
         Order([Product].MEMBERS, [Measures].[Unit Sales]))

    would perform many redundant sorts.

    mondrian.rolap.RolapResult.flushAfterEachQuery boolean false Obsolete.
    mondrian.rolap.EnableRolapCubeMemberCache boolean true Determines whether to cache RolapCubeMember objects, each of which associates a member of a shared hierarchy with a particular cube in which it is being used.

    The default is true, that is, use a cache. If you wish to use the member cache control aspects of mondrian.olap.CacheControl you must set this property to false.

    SQL generation
    mondrian.native.crossjoin.enable boolean true
    If enabled, some NON EMPTY CrossJoin MDX statements will be computed in the database and not within Mondrian/Java
    mondrian.native.topcount.enable boolean true
    If enabled, some TopCount MDX statements will be computed in the database and not within Mondrian/Java
    mondrian.native.filter.enable boolean true
    If enabled, some Filter() MDX statements will be computed in the database and not within Mondrian/Java
    mondrian.native.nonempty.enable boolean true
    If enabled, some NON EMPTY MDX set operations like member.children, level.members and member.descendants will be computed in the database and not within Mondrian/Java
    mondrian.native.ExpandNonNative boolean false
    If set to true, expand non native sub-expressions of a native expression into MemberLists
    mondrian.native.unsupported.alert string OFF Alerting action {OFF, WARN, ERROR} to take in case native evaluation of a function is enabled but not supported for that function's usage in a particular query. (No alert is ever raised in cases where native evaluation would definitely have been wasted effort.) Currently, alerts are only raised for the NonEmptyCrossJoin function.
    mondrian.rolap.generate.formatted.sql boolean false Whether to pretty-print SQL generated statements.

    If true, Mondrian generates SQL strings are generated in the log or output in pretty-print mode, formatted for ease of reading.

    mondrian.rolap.maxConstraints int 500 Max number of constraints in a single `IN' SQL clause.

    This value may be variant among database products and their runtime settings. Oracle, for example, gives the error "ORA-01795: maximum number of expressions in a list is 1000".

    Recommended values:

    • Oracle: 1,000
    • DB2: 2,500
    • Other: 10,000
    mondrian.rolap.groupingsets.enable boolean false If enabled, some rollup queries will be combined using group by grouping sets. Even if this property is set to true, grouping sets will be used only on supported Databases.

    Currently grouping sets will be used only against Oracle, DB2 and Teradata.

    mondrian.xmla.drillthroughTotalCount.enable boolean true If enabled, first row in the result of an XML/A drill-through request will be filled with the total count of rows in underlying database.
    mondrian.xmla.drillthroughMaxRows int 1,000 Limit on the number of rows returned by XML/A drill through request.

    Memory monitoring
    mondrian.util.memoryMonitor.enable boolean true If enabled and one is using Java5 or above, then Mondrian will use the Java5 memory monitoring capability - a MemoryLimitExceededException exception ought be throw rather than a OutOfMemoryError when memory get low and the JVM will not crash.
    mondrian.util.memoryMonitor.percentage.threshold int 90 At what level of memory usage should Mondrian be notified that memory is running low.

    mondrian.util.MemoryMonitor.class string - By default this property does not have a value. If set, then it is the class name of an implementation of the MemoryMonitor class and is used by the MemoryMonitorFactory to create the single instance.
    mondrian.calc.ExpCompiler.class string - By default this property does not have a value. If set, then it is the class name of an implementation of the ExpCompiler class and is used by the ExpCompiler.Factory to create each instance.
    mondrian.rolap.RolapMember.PropertyValueMapFactory.class string mondrian.rolap.RolapMember.DefaultPropertyValueMapFactory If set, the specified factory class is used to create the Map that stores member property values in mondrian.rolap.RolapMember. The factory class must implement mondrian.rolap.RolapMember.PropertyValueMapFactory.
    mondrian.rolap.SqlMemberSource.ValuePoolFactory.class string - If set, the specified factory class is used to create a Map that is used to pool member property values in mondrian.rolap.SqlMemberSource. If unset, no value pooling is done. The factory class must implement mondrian.rolap.SqlMemberSource.ValuePoolFactory.
    mondrian.spi.dataSourceResolverClass string mondrian.spi.impl.JndiDataSourceResolver If set, the specified plugin is used to resolve data source names in the connect string into JDBC data source (javax.sql.DataSource) objects. If not set, the default implementation uses JDNI to resolve.

    Limit properties 

    Properties mondrian.result.limit, mondrian.rolap.iterationLimit and mondrian.rolap.queryTimeout enforce runtime limits on the time or space required to execute a query. If any of these limits are exceeded, mondrian throws an exception which extends mondrian.olap.ResultLimitExceededException.

    Connect strings 

    Connect string syntax 

    Mondrian connect strings are a connection of property/value pairs, of the form 'property=value;property=value;...'.

    Values can be enclosed in single-quotes, which allows them to contain spaces and punctuation. See the the OLE DB connect string syntax specification.

    The supported properties are described below.

    Connect string properties 

    Name Required? Description
    Provider Yes

    Must have the value "Mondrian".

    Jdbc Exactly one

    The URL of the JDBC database where the data is stored. You must specify either DataSource or Jdbc.


    The name of a data source loaded via JNDI. The name must be a valid JNDI name, and the object referenced must implement the javax.sql.DataSource interface. You must specify either DataSource or Jdbc.

    JdbcDrivers Yes

    Comma-separated list of JDBC driver classes, for example,

    JdbcUser No

    The name of the user to log on to the JDBC database. (If your JDBC driver allows you to specify the user name in the JDBC URL, you don't need to set this property.)

    JdbcPassword No

    The name of the password to log on to the JDBC database. (If your JDBC driver allows you to specify the password in the JDBC URL, you don't need to set this property.)

    Catalog Exactly one

    The URL of the catalog, an XML file which describes the schema: cubes, hierarchies, and so forth. For example,


    Catalogs are described in the Schema Guide. See also CatalogContent.


    An XML string representing the schema: cubes, hierarchies, and so forth. For example,

    CatalogContent=<Schema name="MySchema"><Cube name="Cube1"> ... </Schema>

    Catalogs are described in the Schema Guide. See also Catalog.

    CatalogName No

    Not used. If, in future, Mondrian supports multiple catalogs, this property will specify which catalog to use. See also Catalog.

    PoolNeeded No

    Tells Mondrian whether to add a layer of connection pooling.

    If the value "true" is specified, or no value is specified, Mondrian assumes that:

    • connections created via the Jdbc property are not pooled, and therefore need to be pooled;
    • connections created via the DataSource are already pooled.

    If the value "false" is specified, Mondrian does not apply connection-pooling to any connection.

    Role No

    The name of the role to adopt for access-control purposes. If not specified, the connection uses a role which has access to every object in the schema.

    This property can contain multiple role names separated by commas. If so, queries in the connection execute with the sum of the privileges of all of the rules; the effect is the same as running under a union role, defined using the <Union> element in the schema file.

    If a role name contains a comma, escape the comma using an extra comma. For example, a connection created with

    Role='Pacific region manager,Europe,, Middle East and Africa manager'

    will execute with the combined privileges of the roles "Pacific region manager", and "Europe, Middle East and Africa manager".

    jdbc.* No

    Any property whose name begins with "jdbc." will be added to the JDBC connection properties, after removing this prefix. This allows you to specify connection properties without a URL.

    For example, given the properties

    jdbc.Timeout=50; jdbc.CacheSize=1m

    Mondrian will create a JDBC connection using the properties {Timeout="50", CacheSize="1m"}.

    UseContentChecksum No

    Allows mondrian to work with dynamically changing schema. If this property is set to true and schema content has changed (previous checksum doesn't equal with current), schema would be reloaded. The default is false.

    Could be used in combination with DynamicSchemaProcessor property.

    UseSchemaPool No

    Controls whether a new connection use a schema from the schema cache. If true, the default, a connection shares a schema definition (and hence also a cache of aggregate data retrieved by previous queries) with other connections which have a textually identical schema definition.

    If false, the connection has a private schema definition and cache.

    DynamicSchemaProcessor No

    The name of a class which is called at runtime in order to modify the schema content. The class must implement the mondrian.spi.DynamicSchemaProcessor interface. For example,

    DynamicSchemaProcessor = mondrian.i18n.LocalizingDynamicSchemaProcessor

    uses the builtin schema processor class mondrian.i18n.LocalizingDynamicSchemaProcessor to replace variables in the schema file, according to resource files and the current locale (see the Locale property).

    Locale No

    The requested Locale for the current session. The locale determines the formatting of numbers and date/time values, and Mondrian's error messages.

    Example values are "en" (English), "en_US" (United States English), "hu" (Hungarian). If Locale is not specified, then the name of system's default will be used, as per java.util.Locale#getDefault().

    Connect string properties are also documented in the RolapConnectionProperties class.

    Cache management

    Schema cache

    To flush all schema definitions, use the mondrian.olap.CacheControl.flushSchemaCache() method:

    import mondrian.olap.*;

    Connection connection;
    CacheControl cacheControl = connection.getCacheControl(null);

    The cache is only used when creating new connections; existing connections retain their schemas.

    There are four connect string properties that control the use of the Schema cache: UseSchemaPool, UseContentChecksum, CatalogContent and DynamicSchemaProcessor.

    The UseSchemaPool property controls whether or not the cache is used regardless of the values of any of the other properties. If UseSchemaPool is "false", then the cache is not used; each request for a new schema object creates a new one (entailing the re-parsing of the schema definition and re-scanning of the database for meta data and aggregate tables - very slow, and, in addition, there is no reuse of the in-memory aggregate cache).

    Next, if UseContentChecksum is "true", then a check sum (MD5) is created from the schema definition content (not URL) and it is this check sum that is used as the key to lookup previously cached versions of the schema definition. If two schema definitions produce different check sums, then one can safely assume that they are different schemas (of course, it is possible that only a comment or some whitespace in the schema definition changed in which case the two schemas would actually be the same, but because their check sums are different, different schema objects are used). If UseContentChecksum is "false", then no check sum is created and used as the lookup key, rather, a combination of the connection attributes "catalogUrl", "connectionKey", "jdbcUser", "dataSourceStr" or "catalogUrl", "dataSource" are used to create the key.

    If the CatalogContent is specified, then it is used as the schema definition content. If, in fact, it is specified, then the value of DynamicSchemaProcessor, if any, is ignored.

    Finally, the DynamicSchemaProcessor connection string property is the class name of a class that implements the DynamicSchemaProcessor interface. If set, an instance of the class is created for each schema request and its "processSchema" method is called which returns the schema definition content.

    Memory management

    Out Of Memory

    Java OutOfMemoryError errors have always been an issue with applications. When the JVM throws an Error as opposed to an Exception it is telling the application that its world has ended and it has no recourse but to die. Prior to Java5 there was not much one could do other than buy 64-bit machines with lots of RAM and hope for the best. For a multi-user, Mondrian environment with potentially very large data-sets and clients that can generate queries requesting arbitrarily large amounts of that data, this can be an issue. This is especially the case when Mondrian is being hosted on some corporate web-server; applications that kill web-servers are not looked upon favorably by IT.

    With Java5 (and Java6, etc.) there is alternative. An application cay take advantage of a new feature in Java5 allowing the application to be notified when memory starts running low. This allows the application to take preemptive action prior to an OutOfMemoryError being generated by the Java runtime.

    Mondrian takes advantage of this new feature. Rather than passing an OutOfMemoryError to its client, it will now stop processing the present query, free up data structures associated with the present query and return a MemoryLimitExceededException to the client. The MemoryLimitExceededException is one of Mondrian's ResultLimitExceededException which are used to communicate with clients that a limit has been exceeded, in this case, memory usage.

    By default, for Mondrian running under Java5, this feature is enabled and the "safety limit" is set at 90 percent, when memory usage gets to with 90 percent of the maximum possible, the the processing of the current query is stopped and a MemoryLimitExceededException is return to the client. See the Memory monitoring properties above on this page for additional information.

    Lastly, the gorilla in the closet. Java5 in its wisdom only allows for one memory threshold notification level to be registered with the JVM. What this means is if within the same JVM, some code registers one level, say, at 80% (here I use percentages for ease of presentation rather than number of bytes which is what the Java5 API actually supports) and some other code later on registers a level of 90%, then it is the 90% that the JVM knows about - it knows nothing of the previously registered 80%. What this means is that the code expecting to be notified when the memory level crosses 80%, won't be notified!

    For many applications that don't share their JVM with other applications, this is not a problem, but for Mondrian is it potentially an issue. Mondrian can be running in a Webserver and Webservers can have more than one independent applications. Each such application can register a different memory threshold notification level. In general, application-containing applications such as web-servers or application-servers are a problem with the current Java5 memory threshold notification approach. At the current time, I do not know a way around this problem.


    Mondrian uses log4j for all information and debug logging. When running within an application server, Mondrian's log4j configuration is determined by the server's or web application's log4j configuration. Please see log4j's documentation for a additional details.

    Configuring log4j within Mondrian's test environment

    When running outside an application server, log4j determines the location of the log4j.xml file via the log4j.configuration java system property. log4j treats this string as a URL, so to have it detect the log4j file on the file system, you must use the syntax "file:DIR/log4j.xml". Relative paths are acceptible, so if you have your log4j.xml file in the root directory of mondrian, "file:log4j.xml" will load the correct file. You may specify the log4j.configuration property in, because Mondrian's ant build file explicitly sets the property as a JVM system property when running JUnit tests.

    MDX and SQL Statement Logging

    The default log4j.xml file is configured so that a separate log file is created for both MDX and SQL statement logging. In the code, the MDX and SQL strings are logged at the debug level, so to disable them you can set the log level to INFO or any other level above debug. Statement logging occurs within the log4j categories "mondrian.mdx" and "mondrian.sql". These categories log the statements and how long they took to execute. The SQL log also records the number of results returned in the result set.

    For example, to trace both MDX and SQL statements, create a file in the directory where you started mondrian with the following contents:

    # Set root logger level to DEBUG and its only appender to MONDRIAN.
    log4j.rootLogger=WARN, MONDRIAN
    # MONDRIAN is set to be a ConsoleAppender.
    # MONDRIAN uses PatternLayout.
    log4j.appender.MONDRIAN.layout.ConversionPattern=%-4r [%t] %-5p %c %x - %m%n
    # Trace MDX and SQL statements
    log4j.category.mondrian.mdx=DEBUG, MONDRIAN
    log4j.category.mondrian.sql=DEBUG, MONDRIAN

    Then mondrian with the argument on the Java command line.

    Consider setting the property mondrian.rolap.generate.formatted.sql=true in to make the format more readable.

    Author: Julian Hyde; last modified December, 2009.
    Version: $Id: //open/mondrian-release/3.2/doc/configuration.html#1 $ (log)
    Copyright (C) 2006-2009 Julian Hyde