cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ariel Weisberg (JIRA)" <>
Subject [jira] [Commented] (CASSANDRA-13265) Expiration in OutboundTcpConnection can block the reader Thread
Date Mon, 13 Mar 2017 17:44:41 GMT


Ariel Weisberg commented on CASSANDRA-13265:

The right way to do it is create a branch for all the versions where this is going to be fixed.
Start at 2.2, merge to 3.0, merge to 3.11, then merge to trunk. 

You can get away with one field. Check the next expiration time, CAS it to {{Long.MAX_VALUE}},
then when you are done store the next expiration time in it. Doing it with two fields also
works. I wouldn't bother changing it.

If you set the default value via a property it will work fine. It will set it once when it
loads the class at startup and then overwrite it with YAML contents or JMX invocations. Generally
we do set the default value in config via assignment. Doing it via property gives yet another
way to set the value, but it's the least important. It's more useful for things which aren't
in the YAML. Adding YAML properties adds a bit of boiler plate.

* [A smaller value could potentially expire messages slightly sooner at the expense of more
CPU time and queue contention while iterating the backlog of messages.
* [You shouldn't need the check for null? Usually we "just" make sure its not null and skip
the boilerplate.|]
* [Avoid unrelated whitespace changes.|]
*  [I still think it's a good idea to avoid hard coding this kind of value so operators have
options without recompiling.|]
* Fun fact. You don't need {{backlogNextExpirationTime}} to be volatile. You can piggyback
on {{backlogExpirationActive}} to get the desired effects from the Java memory model. A store
to {{backlogExpirationActive}} makes a prior stores (by the current thread) to {{backlogNextExpirationTime}}
visible. A read of {{backlogExpirationActive}} would make prior stores to {{backlogNextExpirationTimeVisible}}
by the last writer to {{backlogExpirationActive}} visible. The volatile read is close to free
so I wouldn't change it just so it's not sensitive to the order the fields are accessed in.
* [Breaking out the uber bike shedding this could be maybeExpireMessages.|]
* [Swap the order of these two stores so it doesn't do extra expirations.|]
* [Using a boxed integer makes it a bit confusing because now everyone has to know how null
is handled. What's the diff between null and 0? Better to let 0 be disabled and not have null.|]
* [This is not quite correct you can't count drainCount as dropped because some of the drained
messages may have been sent during iteration.|]

> Expiration in OutboundTcpConnection can block the reader Thread
> ---------------------------------------------------------------
>                 Key: CASSANDRA-13265
>                 URL:
>             Project: Cassandra
>          Issue Type: Bug
>         Environment: Cassandra 3.0.9
> Java HotSpot(TM) 64-Bit Server VM version 25.112-b15 (Java version 1.8.0_112-b15)
> Linux 3.16
>            Reporter: Christian Esken
>            Assignee: Christian Esken
>             Fix For: 2.2.x, 3.0.x, 3.11.x, 4.x
>         Attachments: cassandra.pb-cache4-dus.2017-02-17-19-36-26.chist.xz,
> I observed that sometimes a single node in a Cassandra cluster fails to communicate to
the other nodes. This can happen at any time, during peak load or low load. Restarting that
single node from the cluster fixes the issue.
> Before going in to details, I want to state that I have analyzed the situation and am
already developing a possible fix. Here is the analysis so far:
> - A Threaddump in this situation showed  324 Threads in the OutboundTcpConnection class
that want to lock the backlog queue for doing expiration.
> - A class histogram shows 262508 instances of OutboundTcpConnection$QueuedMessage.
> What is the effect of it? As soon as the Cassandra node has reached a certain amount
of queued messages, it starts thrashing itself to death. Each of the Thread fully locks the
Queue for reading and writing by calling, making the situation worse and worse.
> - Writing: Only after 262508 locking operation it can progress with actually writing
to the Queue.
> - Reading: Is also blocked, as 324 Threads try to do, and fully lock
the Queue
> This means: Writing blocks the Queue for reading, and readers might even be starved which
makes the situation even worse.
> -----
> The setup is:
>  - 3-node cluster
>  - replication factor 2
>  - Consistency LOCAL_ONE
>  - No remote DC's
>  - high write throughput (100000 INSERT statements per second and more during peak times).

This message was sent by Atlassian JIRA

View raw message