Coherence can perform queries and indexes against currently cached data that meets a given set of criteria. Queries and indexes can be simple, employing filters packaged with Coherence, or they can be run against multi-value attributes such as collections and arrays.

This section contains the following information:

Query Functionality

Coherence provides the ability to search for cache entries that meet a given set of criteria. The result set may be sorted if desired. Queries are evaluated with Read Committed isolation.

It should be noted that queries apply only to currently cached data (and will not use the CacheLoader interface to retrieve additional data that may satisfy the query). Thus, the dataset should be loaded entirely into cache before queries are performed. In cases where the dataset is too large to fit into available memory, it may be possible to restrict the cache contents along a specific dimension (e.g. date) and manually switch between cache queries and database queries based on the structure of the query. For maintainability, this is usually best implemented inside a cache-aware data access object (DAO).

Indexing requires the ability to extract attributes on each Partitioned cache node; in the case of dedicated CacheServer instances, this implies (usually) that application classes must be installed in the CacheServer classpath.

For Local and Replicated caches, queries are evaluated locally against unindexed data. For Partitioned caches, queries are performed in parallel across the cluster, using indexes if available. Coherence includes a Cost-Based Optimizer (CBO). Access to unindexed attributes requires object deserialization (though indexing on other attributes can reduce the number of objects that must be evaluated).

Simple Queries

Querying cache content is very simple:

Filter filter = new GreaterEqualsFilter("getAge", 18);

for (Iterator iter = cache.entrySet(filter).iterator(); iter.hasNext(); )
    {
    Map.Entry entry = (Map.Entry) iter.next();
    Integer key = (Integer) entry.getKey();
    Person person = (Person) entry.getValue();
    System.out.println("key=" + key + " person=" + person);
    }

Coherence provides a wide range of filters in the com.tangosol.util.filter package.

Querying Partitioned Caches The Partitioned Cache implements this method using the Parallel Query feature, which is only available in Coherence Enterprise Edition or higher. When working with a Partitioned Cache in Coherence Standard Edition, this method will retrieve the data set to the client for processing.

Querying Near Caches Although queries can be executed through a near cache, the query will not use the front portion of a near cache. If using a near cache with queries, the best approach is to use the following sequence: Set setKeys = cache.keySet(filter); Map mapResult = cache.getAll(setKeys);

Any queryable attribute may be indexed with the addIndex method of the QueryMap class:

// addIndex(ValueExtractor extractor, boolean fOrdered, Comparator comparator)
cache.addIndex(extractor, true, null);

The fOrdered argument specifies whether the index structure is sorted. Sorted indexes are useful for range queries, including "select all entries that fall between two dates" and "select all employees whose family name begins with 'S'". For "equality" queries, an unordered index may be used, which may have better efficiency in terms of space and time.

The comparator argument can be used to provide a custom java.util.Comparator for ordering the index.

This method is only intended as a hint to the cache implementation, and as such it may be ignored by the cache if indexes are not supported or if the desired index (or a similar index) already exists. It is expected that an application will call this method to suggest an index even if the index may already exist, just so that the application is certain that index has been suggested. For example in a distributed environment, each server will likely suggest the same set of indexes when it starts, and there is no downside to the application blindly requesting those indexes regardless of whether another server has already requested the same indexes. Indexes are a feature of Coherence Enterprise Edition or higher. This method will have no effect when using Coherence Standard Edition.

Note that queries can be combined by Coherence if necessary, and also that Coherence includes a cost-based optimizer (CBO) to prioritize the usage of indexes. To take advantage of an index, queries must use extractors that are equal ((Object.equals()) to the one used in the query.

A list of applied indexes can be retrieved from the StorageManagerMBean via JMX.

Query Concepts

This section goes into more detail on the design of the query interface, building up from the core components.

The concept of querying is based on the ValueExtractor interface. A value extractor is used to extract an attribute from a given object for the purpose of querying (and similarly, indexing). Most developers will need only the ReflectionExtractor implementation of this interface. The ReflectionExtractor uses reflection to extract an attribute from a value object by referring to a method name, typically a "getter" method like getName().

ValueExtractor extractor = new ReflectionExtractor("getName");

Any "void argument" method can be used, including Object methods like toString() (useful for prototyping/debugging). Indexes may be either traditional "field indexes" (indexing fields of objects) or "functional indexes" (indexing "virtual" object attributes). For example, if a class has field accessors "getFirstName" and "getLastName", the class may define a function "getFullName" which concatenates those names, and this function may be indexed.

To query a cache that contains objects with "getName" attributes, a Filter must be used. A filter has a single method which determines whether a given object meets a criterion.

Filter filter = new EqualsFilter(extractor, "Bob Smith");

Note that the filters also have convenience constructors that accept a method name and internally construct a ReflectionExtractor:

Filter filter = new EqualsFilter("getName", "Bob Smith");

To select the entries of a cache that satisfy a particular filter:

for (Iterator iter = cache.entrySet(filter).iterator(); iter.hasNext(); )
    {
    Map.Entry entry = (Map.Entry)iter.next();
    Integer key = (Integer)entry.getKey();
    Person person = (Person)entry.getValue();
    System.out.println("key=" + key + " person=" + person);
    }

To select and also sort the entries:

// entrySet(Filter filter, Comparator comparator) 
Iterator iter = cache.entrySet(filter, null).iterator();

The additional null argument specifies that the result set should be sorted using the "natural ordering" of Comparable objects within the cache. The client may explicitly specify the ordering of the result set by providing an implementation of Comparator. Note that sorting places significant restrictions on the optimizations that Coherence can apply, as sorting requires that the entire result set be available prior to sorting.

Batching Queries and Memory Usage

In order to preserve memory on the client issuing a query, there are various techniques that can be used to retrieve query results in batches.

Using the keySet form of the queries – combined with getAll() – will reduce memory consumption since the entire entry set is not deserialized on the client at once. It will also take advantage of near caching:

// keySet(Filter filter)
Set setKeys = cache.keySet(filter);
Set setPageKeys = new HashSet();
int PAGE_SIZE = 100;
for (Iterator iter = setKeys.iterator(); iter.hasNext();)
    {
    setPageKeys.add(iter.next());
    if (setKeyPage.size() == PAGE_SIZE || !iter.hasNext())
        {
        // get a block of values
        Map mapResult = cache.getAll(setPageKeys);

        // process the block
        // ...

        setPageKeys.clear();
        }
    }

A LimitFilter may be used to limit the amount of data sent to the client, and also to provide "paging" for users:

int pageSize = 25;
Filter filter = new GreaterEqualsFilter("getAge", 18);

// get entries 1-25
Filter limitFilter = new LimitFilter(filter, pageSize);
Set entries = cache.entrySet(limitFilter);

// get entries 26-50
limitFilter.nextPage();
entries = cache.entrySet(limitFilter);

When using a distributed/partitioned cache, queries can be targeted to partitions and cache servers using a PartitionedFilter. This is the most efficient way of batching query results as each query request will be targeted to a single cache server, thus reducing the number of servers that must respond to a request and making the most efficient use of the network.

PartitionedFilter Restrictions Use of PartitionedFilter is limited to cluster members; it cannot be used by Coherence*Extend clients. Coherence*Extend clients may use one of the two techniques described above, or these queries can be implemented as an Invocable and executed remotely by a Coherence*Extend client.

To execute a query partition by partition:

DistributedCacheService service =
     (DistributedCacheService) cache.getCacheService();
int cPartitions = service.getPartitionCount();

PartitionSet parts = new PartitionSet(cPartitions);
for (int iPartition = 0; iPartition < cPartitions; iPartition++)
    {
    parts.add(iPartition);

    Filter filterPart = new PartitionedFilter(filter, parts);

    Set setEntriesPart = cache.entrySet(filterPart);

    // process the entries ...

    parts.remove(iPartition);
    }

Queries can also be executed on a server by server basis:

DistributedCacheService service =
    (DistributedCacheService) cache.getCacheService();
int cPartitions = service.getPartitionCount();

PartitionSet partsProcessed = new PartitionSet(cPartitions);
for (Iterator iter = service.getStorageEnabledMembers().iterator();
        iter.hasNext();)
    {
    Member member = (Member) iter.next();

    PartitionSet partsMember = service.getOwnedPartitions(member);

    // due to a redistribution some partitions may have already been processed
    partsMember.remove(partsProcessed);

    Filter filterPart = new PartitionedFilter(filter, partsMember);
    Set setEntriesPart = cache.entrySet(filterPart);

    // process the entries ...

    partsProcessed.add(partsMember);
    }

// due to a possible redistribution, some partitions may have been skipped
if (!partsProcessed.isFull())
    {
    partsProcessed.invert();
    Filter filter = new PartitionedFilter(filter, partsProcessed);

    // process the remaining entries ...
    }

Queries Involving Multi-Value Attributes

Coherence supports indexing and querying of multi-value attributes including collections and arrays. When an object is indexed, Coherence will check to see if it is a multi-value type, and will then index it as a collection rather than a singleton. The ContainsAllFilter, ContainsAnyFilter and ContainsFilter are used to query against these collections.

Set searchTerms = new HashSet();
searchTerms.add("java");
searchTerms.add("clustering");
searchTerms.add("books");

// The cache contains instances of a class "Document" which has a method
// "getWords" which returns a Collection<String> containing the set of
// words that appear in the document.
Filter filter = new ContainsAllFilter("getWords", searchTerms);

Set entrySet = cache.entrySet(filter);

// iterate through the search results
// ...

ChainedExtractor

The ChainedExtractor implementation allows chained invocation of zero-argument (accessor) methods.

ValueExtractor extractor = new ChainedExtractor("getName.length");

In the example above, the extractor will first use reflection to call getName() on each cached Person object, and then use reflection to call length() on the returned String. This extractor could be passed into a query, allowing queries (for example) to select all people with names not exceeding 10 letters. Method invocations may be chained indefinitely, for example "getName.trim.length".