Hazelcast IMap : keySet() versus localKeySet()

Question

I'm using Hazelcast 3.4 in a cluster with 5+ nodes. I have a map where keys are complex objects ( not primitive types). At any point of time, the size of the map may be around 200K.

I understand that the entries/values in a map will be stored in different partitions. However, I would like to know the following.

1. Does each member in the Hazelcast cluster maintain the information about the set of all keys in a given map or only about the subset of those keys in the partition(s) that it owns ?

2 & 3 are follow up questions.

2. Is keySet() a distributed operation ? (i.e. will it surely involve remote machine communication in a cluster environment)

3. Is localKeySet() a distributed operation ?

Some more background on the problem:

I need to inform a third party program about the keys present in the hazelcast map in my program during regular intervals of time by calling a service cachedKeys(K[] keys) exposed by the third party program.

Option 1: call keySet() from any one of the nodes in the cluster to retrieve all the keys in the map and then call cachedKeys(K[] keys) service from that node.

Option 2: call localKeySet() from each of the nodes in the cluster to identify the local keys owned by each nodes, followed by cachedKeys(K[] keys) call from each node.

Above answers will help me to choose the right option.

Answer 1

Members do not maintain a global key set. Each member is aware of local entries only. Hence keySet() is a distributed operation hitting all members while localKeySet() is just a local operation. I hope this info helps you a bit.

Question got answered in google groups: Credits: Jaromir Hamala