[
https://issues.apache.org/jira/browse/IGNITE-10250?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16695815#comment-16695815
]
Pavel Pereslegin edited comment on IGNITE-10250 at 11/22/18 11:54 AM:
----------------------------------------------------------------------
I managed to reproduce this problem only on atomic partitioned/replicated cache
with a single node.
In case of concurrent notifications from the cache when changing the queue
header, we can observe the following execution order:
(initially queue size = 0, capacity = 1).
{noformat}
T1 (put)
onHeaderChanged("queue full" notification)
T2 (take)
onHeaderChanged("queue empty" notification)
{noformat}
if T2 somehow go ahead T1, then T1 can drain permits that T2 has released.
{noformat}
T2 ("queue empty" notification)
drainPermits()
release(1)
T1 ("queue full" notification)
drainPermits()
{noformat}
As a result, T1 waits infinitely for the semaphore permit to invoke next put.
To resolve this problem we can drain permits only if queue is not full.
was (Author: xtern):
I managed to reproduce this problem only on atomic partitioned/replicated cache
with a single node.
In case of concurrent notifications from the cache when changing the queue
header, we can observe the following execution order:
(initially queue size = 0, capacity = 1).
{noformat}
T3 (put)
onHeaderChanged("queue full" notification)
T4 (take)
onHeaderChanged("queue empty" notification)
{noformat}
if T4 somehow go ahead T3, then T3 can drain permits that T4 has released.
{noformat}
T4 ("queue empty" notification)
drainPermits()
release(1)
T3 ("queue full" notification)
drainPermits()
{noformat}
As a result, T3 waits infinitely for the semaphore permit to invoke next put.
To resolve this problem we can drain permits only if queue is not full.
> Ignite Queue hangs after several read/write operations
> ------------------------------------------------------
>
> Key: IGNITE-10250
> URL: https://issues.apache.org/jira/browse/IGNITE-10250
> Project: Ignite
> Issue Type: Bug
> Components: data structures
> Affects Versions: 2.7
> Reporter: Anton Dmitriev
> Assignee: Pavel Pereslegin
> Priority: Major
> Fix For: 2.8
>
>
> Ignite Queue hangs after several read/write operations. Code to reproduce:
> {code:java}
> try (Ignite ignite = Ignition.start()) {
> IgniteQueue<Integer> queue = ignite.queue("TEST_QUEUE", 1, new
> CollectionConfiguration());
> new Thread(() -> {
> for (int i = 0;; i++) {
> queue.put(i);
> System.out.println("Put: " + i);
> }
> }).start();
> new Thread(() -> {
> for (int i = 0;; i++) {
> queue.take();
> System.out.println("Take: " + i);
> }
> }).start();
> Thread.currentThread().join();
> }
> {code}
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
