[
https://issues.apache.org/jira/browse/FLINK-12852?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16878461#comment-16878461
]
Piotr Nowojski commented on FLINK-12852:
----------------------------------------
Just for the record & FYI, I'm currently working in parallel on changing when
{{requestPartitions}} call is happening (I'm trying to move it from
\{{SingleInputGate#get/poll}} to {{SingleInputGate#setup()}} )
https://issues.apache.org/jira/browse/FLINK-13013 .
> is only allowed to grab extra resources once it knows that it can make a
>progress
When writing this I forgot that it indeed requires not only the immediate
successor operator to be ready, but for all of them as [~gaoyunhaii] wrote, so
we would have to indeed initialize/grab resources in the order from the sinks
up to the sources.
Do I understand it correctly, that we can have three solutions:
# Grab only required resources from sinks up to the sources and once that
completes grab additional resources.
# Make buffers completely not shared between tasks - for example disable slot
sharing and make buffer pools per slot
# Make optional buffers "revocable", for example by spilling.
[~zjwang] which one of those three were you referring to when writing "Network
resource matching in slot" and "Lazy allocation buffers on producer side"?
> Deadlock occurs when requiring exclusive buffer for RemoteInputChannel
> ----------------------------------------------------------------------
>
> Key: FLINK-12852
> URL: https://issues.apache.org/jira/browse/FLINK-12852
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Network
> Affects Versions: 1.7.2, 1.8.1, 1.9.0
> Reporter: Yun Gao
> Assignee: Yun Gao
> Priority: Blocker
> Labels: pull-request-available
> Fix For: 1.9.0
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> When running tests with an upstream vertex and downstream vertex, deadlock
> occurs when submitting the job:
> {code:java}
> "Sink: Unnamed (3/500)" #136 prio=5 os_prio=0 tid=0x00007f2cca81b000
> nid=0x38845 waiting on condition [0x00007f2cbe9fe000]
> java.lang.Thread.State: TIMED_WAITING (parking)
> at sun.misc.Unsafe.park(Native Method)
> - parking to wait for <0x000000073ed6b6f0> (a
> java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject)
> at java.util.concurrent.locks.LockSupport.parkNanos(LockSupport.java:233)
> at
> java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject.awaitNanos(AbstractQueuedSynchronizer.java:2078)
> at java.util.concurrent.ArrayBlockingQueue.poll(ArrayBlockingQueue.java:418)
> at
> org.apache.flink.runtime.io.network.buffer.NetworkBufferPool.requestMemorySegments(NetworkBufferPool.java:180)
> at
> org.apache.flink.runtime.io.network.buffer.NetworkBufferPool.requestMemorySegments(NetworkBufferPool.java:54)
> at
> org.apache.flink.runtime.io.network.partition.consumer.RemoteInputChannel.assignExclusiveSegments(RemoteInputChannel.java:139)
> at
> org.apache.flink.runtime.io.network.partition.consumer.SingleInputGate.assignExclusiveSegments(SingleInputGate.java:312)
> - locked <0x000000073fbc81f0> (a java.lang.Object)
> at
> org.apache.flink.runtime.io.network.partition.consumer.SingleInputGate.setup(SingleInputGate.java:220)
> at
> org.apache.flink.runtime.taskmanager.Task.setupPartionsAndGates(Task.java:836)
> at org.apache.flink.runtime.taskmanager.Task.run(Task.java:598)
> at java.lang.Thread.run(Thread.java:834)
> {code}
> This is due to the required and max of local buffer pool is not the same and
> there may be over-allocation, when assignExclusiveSegments there are no
> available memory.
>
> The detail of the scenarios is as follows: The parallelism of both upstream
> vertex and downstream vertex are 1500 and 500 respectively. There are 200 TM
> and each TM has 10696 buffers( in total and has 10 slots. For a TM that runs
> 9 upstream tasks and 1 downstream task, the 9 upstream tasks start first with
> local buffer pool \{required = 500, max = 2 * 500 + 8 = 1008}, it produces
> data quickly and each occupy about 990 buffers. Then the DownStream task
> starts and try to assigning exclusive buffers for 1500 -9 = 1491
> InputChannels. It requires 2981 buffers but only 1786 left. Since not all
> downstream tasks can start, the job will be blocked finally and no buffer can
> be released, and the deadlock finally occurred.
>
> I think although increasing the network memory solves the problem, the
> deadlock may not be acceptable. Fined grained resource management
> Flink-12761 can solve this problem, but AFAIK in 1.9 it will not include the
> network memory into the ResourceProfile.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
