A newer version of Hazelcast Platform is available.

View latest

CPMap

Hazelcast CPMap is a distributed implementation of a minimal key-value interface. It is a CP data structure, so unlike the traditional maps built for availability (AP), CPMap guarantees data consistency even during network partitions, ensuring that every cluster member always sees the same data. This reliability makes it ideal for storing sensitive information such as user bank balances and system configurations, or managing mission-critical workflows.

CPMap data structure supports the following atomic operations:

  • compareAndSet

  • delete

  • get

  • put

  • putIfAbsent

  • remove

  • set

Due to CPMaps distributed nature, above operations involve remote calls. This means that the calls might take longer to complete than non-distributed counterparts, such as HashMaps.

CPMap is only available in the Enterprise Edition. Your license must include ADVANCED_CP to activate this feature and you must use the Enterprise Edition client.

There is no unsafe variant of CPMap, unlike other CP data structures. Therefore, CP must be enabled before using CPMap.

The following example creates an instance of a CPMap in the default CP group, then updates its contents using the operations listed previously.

        HazelcastInstance hazelcastInstance = Hazelcast.newHazelcastInstance();
        // creates 'capitalCities' CPMap in the 'default' CP Group
        CPMap<String, String> capitalCities = hazelcastInstance.getCPSubsystem().getMap("capitalCities");
        // prefer 'set' over 'put' when the previous value associated with a key is not required
        capitalCities.set("England", "London");
        assert "London".equals(capitalCities.get("England"));
        assert null == capitalCities.put("France", "Paris");
        assert "London".equals(capitalCities.remove("England"));
        // prefer 'delete' over 'remove' when the value associated with the key is not required
        capitalCities.delete("France");
        capitalCities.set("Germany", "Munich");
        assert capitalCities.compareAndSet("Germany", "Munich", "Berlin");
        assert "Berlin".equals(capitalCities.get("Germany"));
        assert null == capitalCities.putIfAbsent("Ireland", "Dublin");
        assert "Dublin".equals(capitalCities.putIfAbsent("Ireland", "Cork"));
        assert "Dublin".equals(capitalCities.get("Ireland"));
        capitalCities.destroy();

The mutations over the CPMap capitalCities in the above example are replicated across each of the members of the default CP Group.

Use set to associate a key with a value instead of put when the previously-associated value with the respective key is not required. This is because the network cost of set is less than put. Use delete instead of remove for the same reason.

CPMaps are not automatically removed. If an instance is not used anymore, Hazelcast does not automatically perform garbage collection on it. This can lead to an OutOfMemoryError. If you create CPMaps on the fly, ensure that they are destroyed. See Destroying Objects and CP Data Structures.

A CPMap has a default capacity of 100MB. This capacity can be adjusted for a CPMap through CPMap Options. The backing data structure of a CPMap is stored on the JVM heap.

Since CPMap is a CP data structure, it takes part in a process called snapshotting. When defining the capacity of a CPMap you must consider the following memory requirements:

  • On-heap capacity for the CPMap

  • On-heap capacity for a CPMap snapshot for each follower within the respective CP Group

Therefore, at a minimum, for a CPMap with a capacity of 5MB within a 3-member CP Group, we must conservatively ensure that we have on-heap headroom for 15MB. This capacity planning must be considered for all CPMap instances you intend to create.

A CPMap, like other CP data structures, resides within a user specified CP Group. Each CP Group is an instance of Raft and consequently has its own Raft Log. Co-locating CP data structures within the same CP Group must be undertaken with care as the operations of the co-located CP data structures are totally ordered with respect to one another via Raft. Snapshotting also takes place at the granularity of a CP Group. Therefore, it is highly recommended that you carefully consider the envisaged access frequency and patterns of a CPMap when deciding the CP Group a CPMap will reside.