part2_data-grid.md

Data Grid

In this part we will learn how to create and use different distributed caches in our cluster.

Ignite ensures cache data is serialized and partitioned across server nodes and available to any node in cluster.

Example

Creating a distributed cache in our cluster is as easy as:

import org.apache.ignite.*;

public class App {

    public static void main(String[] args) {

        Ignite ignite = Ignition.start();

        CacheConfiguration<Key, Value> configuration = new CacheConfiguration<>("my-cache");

        IgniteCache<Key, Value> cache = ignite.getOrCreateCache(configuration);
    }
}

Cache queries

Cache can be queried by key or using SQL, Text or Scan queries.

Complete TODOs in Step1_CacheQuery to fix all tests in Step1_CacheQueryTest.

Cache Modes

Cache entry distribution across server nodes can be adjusted by 3 means:

• Partitioned Cache: each entry is in one primary node node only
• Replicated Cache: each entry is in one primary node and n-1 backup nodes (n is cluster size).
• Cache Backups: each entry is in one primary node and x backup nodes (x is backup size).

Complete TODOs in Step2_CacheMode to fix all tests in Step2_CacheModeTest.

Compute affinity

Compute Affinity allows sending computations according to a cache entry location.

Affinity avoids cross-node cache queries by sending computations only to relevant nodes. For example:

import org.apache.ignite.*;

public class App {

    public static void main(String[] args) {

        Ignite ignite = Ignition.start();

        IgniteCache<Key, Value> cache = ignite.getOrCreateCache("my-cache");
        cache.put("key1", "value1");
        cache.put("key2", "value2");

        String value = ignite.compute()
            .affinityCall("my-cache", "key2", () -> cache.get("key2"));
    }
}

Here Ignite does no network call on cache.get("key2") call because affinityCall() is executed on node where "key2" is stored.

Complete TODOs in Step3_ComputeAffinity to fix all tests in Step3_ComputeAffinityTest.

Data affinity

AffinityKey allows grouping multiple cache entries in same node.

Data is distributed across nodes with a specific key: affinity key, which differs from the key used for cache. For example:

import org.apache.ignite.*;
import org.apache.ignite.cache.affinity.*;

public class App {

    public static void main(String[] args) {

        Ignite ignite = Ignition.start();

        IgniteCache<Object, Value> cache = ignite.getOrCreateCache("my-cache");
        cache.put(new AffinityKey<>("key1", "group1"), "value1");
        cache.put(new AffinityKey<>("key2", "group1"), "value2");
        cache.put(new AffinityKey<>("key3", "group2"), "value3");
        cache.put(new AffinityKey<>("key4", "group2"), "value4");

        String value = ignite.compute()
            .affinityCall("my-cache", "group1", () -> cache.get("key1") + cache.get("key2"));
    }
}

Ignite ensures key1 and key2 are hosted on the same node since they are stored with same affinity key - "group1".

Here Ignite does no network call on cache.get("key1") and cache.get("key2") calls because affinityCall() is executed on node where keys associated to "group1" are stored.

Complete TODOs in Step4_DataAffinity to fix all tests in Step4_DataAffinityTest.

Spot the difference

Can you spot the difference between case 1 and 2 ?

import org.apache.ignite.*;
import org.apache.ignite.cache.*;

public class App {

    public static void main(String[] args) {

        Ignite ignite = Ignition.start();
        IgniteCache<String, String> cache = ignite.getOrCreateCache("test");

        // Case 1
        Collection<Integer> sizes1 = ignite.compute().broadcast(() -> cache.localSize());

        // Case 2
        Collection<Integer> sizes2 = new ArrayList<>();
        ignite.compute().broadcast(() -> sizes2.add(cache.localSize()));
    }
}

Solution

• Case 1: each server node computes cache.localSize() and results are reduced by Ignite to a collection in client node.
• Case 2: each server node computes cache.localSize() but results are added to a copy of the sizes2 collection defined in client node. Original collection is never updated.

Home | Back | Next