[jira] [Commented] (IGNITE-2145) Async cache operations on primary node might block

["Avihai Berkovitz (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IGNITE-2145?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15055920#comment-15055920 ] 

Avihai Berkovitz commented on IGNITE-2145:
------------------------------------------

I completely agree, but the exact problem is a bit different. EntryProcessors must run atomically, and should be lightweight and not take long. But there might be cases where execution of a single processor is blocked for a couple of seconds (IO, DB access, etc.). In those cases *other* operations that are executed asynchronously should wait for it to finish. But the invocation itself (async.put(), async.get(), etc.) should return without being blocked, because this is the entire point of an async operation.

> Async cache operations on primary node might block
> --------------------------------------------------
>
>                 Key: IGNITE-2145
>                 URL: https://issues.apache.org/jira/browse/IGNITE-2145
>             Project: Ignite
>          Issue Type: Bug
>    Affects Versions: ignite-1.4
>         Environment: Windows 8.1 64 bit
> java version "1.8.0_51"
> Java(TM) SE Runtime Environment (build 1.8.0_51-b16)
> Java HotSpot(TM) 64-Bit Server VM (build 25.51-b03, mixed mode)
> Ignite 1.4.0
>            Reporter: Avihai Berkovitz
>
> When executing async operations on caches the initial function call is supposed to return immediately, without regard to cache state, network traffic, etc. If we reference a cache key on a different node it works. But if the primary node for a cache key is the same node that initiate the operation everything is executed synchronously.
> This is probably an optimization, but can have side effects. For example, if a long running CacheProcessor is invoked on this key, every operation will block.
> Here is some code to illustrate the problem. Notice that there is a 4 seconds delay between "Before async op" and "After async op". If you reverse the isPrimary() condition it won't happen.
> {code}
> IgniteConfiguration igniteConfiguration = new IgniteConfiguration()
> 		.setFailoverSpi(new AlwaysFailoverSpi())
> 		.setGridLogger(new Slf4jLogger())
> 		.setPeerClassLoadingEnabled(false)
> 		.setDeploymentMode(DeploymentMode.CONTINUOUS);
> Ignite ignite1 = Ignition.start(igniteConfiguration);
> Ignite ignite2 = Ignition.start(igniteConfiguration.setGridName("2"));
> assert ignite1.cluster().nodes().size() == 2;
> assert ignite2.cluster().nodes().size() == 2;
> CacheConfiguration<String, Integer> cacheConfiguration = new CacheConfiguration<String, Integer>()
> 		.setName("test")
> 		.setCacheMode(CacheMode.PARTITIONED)
> 		.setAtomicityMode(CacheAtomicityMode.ATOMIC)
> 		.setWriteSynchronizationMode(CacheWriteSynchronizationMode.FULL_SYNC);
> IgniteCache<String, Integer> cache1 = ignite1.getOrCreateCache(cacheConfiguration);
> IgniteCache<String, Integer> cache2 = ignite2.getOrCreateCache(cacheConfiguration);
> // Find key mapped to the first node
> String key = IntStream.range(0, 100)
> 		.mapToObj(value -> "HI" + value)
> 		.filter(s -> ignite1.affinity(cacheConfiguration.getName()).isPrimary(ignite1.cluster().localNode(), s))
> 		.findFirst().get();
> // Run a blocked processor from node 2
> Thread thread = new Thread() {
> 	@Override
> 	public void run() {
> 		System.out.println("Invoking blocked processor");
> 		cache2.invoke(key, (entry, arguments) -> {
> 			System.out.println("IN INVOKE");
> 			try {
> 				Thread.sleep(5 * 1000);
> 			} catch (InterruptedException e) {
> 				e.printStackTrace();
> 			}
> 			return null;
> 		});
> 		System.out.println("Blocked invoke returned");
> 	}
> };
> thread.start();
> Thread.sleep(1000);
> IgniteCache<String, Integer> async1 = cache1.withAsync();
> System.out.println("Before async op");
> //async1.put(key, 1);
> //async1.get(key);
> async1.containsKey(key);
> System.out.println("After async op");
> assert async1.future().isDone();
> thread.join();
> ignite2.close();
> ignite1.close();
> {code}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)