[
https://issues.apache.org/jira/browse/CASSANDRA-8446?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14239994#comment-14239994
]
Sylvain Lebresne commented on CASSANDRA-8446:
---------------------------------------------
I'm afraid nothing in this description indicates a bug. Ligthweight
transactions can throw write timeouts and they will under heavy contention. And
a greater replica factor does make it somewhat more likely. It's something to
be aware of and that your application should defend against by handling those
timeouts (which is not really specific to lightweigth transactions). This is
why lightweigth transactions are not recommended if you have lots of
contention: they just don't perform very well in this case as of the current
implementation.
> Lost writes when using lightweight transactions
> -----------------------------------------------
>
> Key: CASSANDRA-8446
> URL: https://issues.apache.org/jira/browse/CASSANDRA-8446
> Project: Cassandra
> Issue Type: Bug
> Components: API, Core
> Environment: Cassandra 2.1.2 on Java(TM) SE Runtime Environment
> (build 1.7.0_65-b17)
> Reporter: Jochen Zeischka
> Fix For: 2.1.3
>
>
> There's a serious problem when using lightweight transactions. Whenever th
> cluster gets any load, write timeout exceptions start to occur and the client
> has no way to know whether the write actually succeeded or not.
> I a simple test (https://gist.github.com/anonymous/4c83f2962b57fce4c3df) this
> results in large percentages of lost writes (0 - 50%, depending on the load).
> The problem was described in
> https://stackoverflow.com/questions/27313360/how-can-we-avoid-losing-writes-when-using-lightweight-transactions-cas-in-cass
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
