[ 
https://issues.apache.org/jira/browse/CASSANDRA-13740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16119799#comment-16119799
 ] 

Aleksey Yeschenko commented on CASSANDRA-13740:
-----------------------------------------------

The patch likely works, but I think we can do better.

Some of the issues I have with it:
1. It introduces a dependency on {{HintsService}} and {{StorageService}} to 
{{HintsWriteExecutor}}
2. It introduces a dependency on {{HintsService}} to {{HintsStore}}

When designing the current iteration of hints I was very careful to design the 
system in a top-down way without any interleaving that’s avoidable. Each class 
is a dumb as possible on its own, and as you go up, you just compose dumb 
classes that by themselves know nothing of layers above them.

As for the problem itself, we do acknowledge that “The worst that can happen if 
we don't get everything right is a hints file (or two) remaining undeleted.” - 
comments to {{excise()}}, it’s more of a known limitation than a bug. But of 
course we can improve on it. What is a problem, however, is the inability to 
programmatically remove those orphan files via JMX. {{nodetool truncatehints}} 
should get results no matter what, and it should be fixed.

If we want to deal with the orphans for sure - and I don’t see why not improve 
this as well - I suggest you do so in a different way. Perhaps as last step of 
{{excise()}} schedule an task - on {{ScheduledExecutors.optionalTasks}} to 
clean up any orphans, if any, after some delay.

> Orphan hint file gets created while node is being removed from cluster
> ----------------------------------------------------------------------
>
>                 Key: CASSANDRA-13740
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-13740
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>            Reporter: Jaydeepkumar Chovatia
>            Assignee: Jaydeepkumar Chovatia
>            Priority: Minor
>             Fix For: 3.0.x
>
>         Attachments: 13740-3.0.15.txt, gossip_hang_test.py
>
>
> I have found this new issue during my test, whenever node is being removed 
> then hint file for that node gets written and stays inside the hint directory 
> forever. I debugged the code and found that it is due to the race condition 
> between [HintsWriteExecutor.java::flush | 
> https://github.com/apache/cassandra/blob/cassandra-3.0/src/java/org/apache/cassandra/hints/HintsWriteExecutor.java#L195]
>  and [HintsWriteExecutor.java::closeWriter | 
> https://github.com/apache/cassandra/blob/cassandra-3.0/src/java/org/apache/cassandra/hints/HintsWriteExecutor.java#L106]
> . 
>  
> *Time t1* Node is down, as a result Hints are being written by 
> [HintsWriteExecutor.java::flush | 
> https://github.com/apache/cassandra/blob/cassandra-3.0/src/java/org/apache/cassandra/hints/HintsWriteExecutor.java#L195]
> *Time t2* Node is removed from cluster as a result it calls 
> [HintsService.java-exciseStore | 
> https://github.com/apache/cassandra/blob/cassandra-3.0/src/java/org/apache/cassandra/hints/HintsService.java#L327]
>  which removes hint files for the node being removed
> *Time t3* Mutation stage keeps pumping Hints through [HintService.java::write 
> | 
> https://github.com/apache/cassandra/blob/cassandra-3.0/src/java/org/apache/cassandra/hints/HintsService.java#L145]
>  which again calls [HintsWriteExecutor.java::flush | 
> https://github.com/apache/cassandra/blob/cassandra-3.0/src/java/org/apache/cassandra/hints/HintsWriteExecutor.java#L215]
>  and new orphan file gets created
> I was writing a new dtest for {CASSANDRA-13562, CASSANDRA-13308} and that 
> helped me reproduce this new bug. I will submit patch for this new dtest 
> later.
> I also tried following to check how this orphan hint file responds:
> 1. I tried {{nodetool truncatehints <node>}} but it fails as node is no 
> longer part of the ring
> 2. I then tried {{nodetool truncatehints}}, that still doesn’t remove hint 
> file because it is not yet included in the [dispatchDequeue | 
> https://github.com/apache/cassandra/blob/cassandra-3.0/src/java/org/apache/cassandra/hints/HintsStore.java#L53]
> Reproducible steps:
> Please find dTest python file {{gossip_hang_test.py}} attached which 
> reproduces this bug.
> Solution:
> This is due to race condition as mentioned above. Since 
> {{HintsWriteExecutor.java}} creates thread pool with only 1 worker, so 
> solution becomes little simple. Whenever we [HintService.java::excise | 
> https://github.com/apache/cassandra/blob/cassandra-3.0/src/java/org/apache/cassandra/hints/HintsService.java#L303]
>  a host, just store it in-memory, and check for already evicted host inside 
> [HintsWriteExecutor.java::flush | 
> https://github.com/apache/cassandra/blob/cassandra-3.0/src/java/org/apache/cassandra/hints/HintsWriteExecutor.java#L215].
>  If already evicted host is found then ignore hints.
> Jaydeep



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

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org

Reply via email to