[jira] [Comment Edited] (IGNITE-6380) Exception should be thrown on cache creation attempt inside transaction

[Anton Vinogradov (JIRA)]

    [ 
https://issues.apache.org/jira/browse/IGNITE-6380?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16173316#comment-16173316
 ] 

Anton Vinogradov edited comment on IGNITE-6380 at 9/20/17 3:14 PM:
-------------------------------------------------------------------

[~xtern],

Simplified reproducer:
{noformat}
    public void test() throws Exception {
        Ignite ignite = startGrid();

        final IgniteCache<Object, Object> cache1 = ignite.getOrCreateCache(new 
CacheConfiguration<>()
            .setAtomicityMode(CacheAtomicityMode.TRANSACTIONAL)
            .setName("cache1"));

        final IgniteCache<Object, Object> cache2 = ignite.getOrCreateCache(new 
CacheConfiguration<>()
            .setAtomicityMode(CacheAtomicityMode.TRANSACTIONAL)
            .setName("cache2"));

        final CountDownLatch latch1 = new CountDownLatch(1);

        Thread t = new Thread(new Runnable() {
            @Override public void run() {
                Lock lock = cache2.lock("fake");

                try {
                    lock.lock();

                    System.out.println("Locked");

                    latch1.countDown();

                    Thread.sleep(1000);

                    cache1.clear();
                }
                catch (Exception e) {
                    e.printStackTrace();

                    throw new RuntimeException(e);
                }
                finally {
                    lock.unlock();
                }
            }
        });

        t.start();

        try {
            latch1.await();
        }
        catch (InterruptedException e) {
            e.printStackTrace();
        }

        ignite.getOrCreateCache(new CacheConfiguration<>()
            .setAtomicityMode(CacheAtomicityMode.TRANSACTIONAL)
            .setName("cache3"));

        System.out.println("No deadlock");
    }
{noformat}

{{cache.clear()}} cause {{ctx.kernalContext().task().execute(...)}} which will 
be mapped to pending topology.

We should restrict such call when lock is locked and there is pending topology.
Please make sure we have no such issues under started optimistic transaction.

Simple addition of 
{{throw new IgniteException("")}}
imitating found locked lock
to  
{{org.apache.ignite.internal.processors.task.GridTaskProcessor#execute(org.apache.ignite.compute.ComputeTask<T,R>,
 T, boolean, java.lang.String)}}
right after {{lock.readLock();}}
fixing reproducer.


was (Author: avinogradov):
[~xtern],

Simplified reproducer:
{noformat}
    public void test() throws Exception {
        Ignite ignite = startGrid();

        final IgniteCache<Object, Object> cache1 = ignite.getOrCreateCache(new 
CacheConfiguration<>()
            .setAtomicityMode(CacheAtomicityMode.TRANSACTIONAL)
            .setName("cache1"));

        final IgniteCache<Object, Object> cache2 = ignite.getOrCreateCache(new 
CacheConfiguration<>()
            .setAtomicityMode(CacheAtomicityMode.TRANSACTIONAL)
            .setName("cache2"));

        final CountDownLatch latch1 = new CountDownLatch(1);

        Thread t = new Thread(new Runnable() {
            @Override public void run() {
                Lock lock = cache2.lock("fake");

                try {
                    lock.lock();

                    System.out.println("Locked");

                    latch1.countDown();

                    Thread.sleep(1000);

                    cache1.clear();
                }
                catch (Exception e) {
                    e.printStackTrace();

                    throw new RuntimeException(e);
                }
                finally {
                    lock.unlock();
                }
            }
        });

        t.start();

        try {
            latch1.await();
        }
        catch (InterruptedException e) {
            e.printStackTrace();
        }

        ignite.getOrCreateCache(new CacheConfiguration<>()
            .setAtomicityMode(CacheAtomicityMode.TRANSACTIONAL)
            .setName("cache3"));

        System.out.println("No deadlock");
    }
{noformat}

{{cache.clear()}} cause {{ctx.kernalContext().task().execute(...)}} which will 
be mapped to pending topology.

We should restrict such call when lock is locked and there is pending topology.
Please make sure we have no such issues under started optimistic transaction.

Simple addition of 
{{throw new IgniteException("")}}
imitating found locked lock
to  
{{org.apache.ignite.internal.processors.task.GridTaskProcessor#execute(org.apache.ignite.compute.ComputeTask<T,R>,
 T, boolean, java.lang.String)}}
fixing reproducer.

> Exception should be thrown on cache creation attempt inside transaction
> -----------------------------------------------------------------------
>
>                 Key: IGNITE-6380
>                 URL: https://issues.apache.org/jira/browse/IGNITE-6380
>             Project: Ignite
>          Issue Type: Improvement
>            Reporter: Yakov Zhdanov
>            Assignee: Pavel Pereslegin
>              Labels: newbie, usability
>
> Exception should be thrown on cache creation attempt inside transaction to 
> prevent deadlocks since cache start triggers exchange and exchange cannot 
> finish until all txs are finished.
> We need to check if thread owns a tx before starting cache and if it does 
> then IllegalStateException should be thrown.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)