[ 
https://issues.apache.org/jira/browse/TEPHRA-297?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Philippe Laflamme updated TEPHRA-297:
-------------------------------------
    Affects Version/s:     (was: 0.14.0-incubating)
                       0.15.0-incubating

> Inconsistent reads
> ------------------
>
>                 Key: TEPHRA-297
>                 URL: https://issues.apache.org/jira/browse/TEPHRA-297
>             Project: Tephra
>          Issue Type: Bug
>    Affects Versions: 0.15.0-incubating
>            Reporter: Philippe Laflamme
>            Assignee: Poorna Chandra
>            Priority: Major
>
> It is possible to get inconsistent reads in the following test setup:
>  * modify the BalanceBooks example to verify the balance periodically (say 
> every second) as opposed to only at the end.
>  * start multiple transaction servers
>  * start the modified BalanceBooks example
>  * randomly restart the transaction servers
> Running this for long enough will eventually lead to the balance verification 
> to fail, reading uncommitted or partially committed data.
> The modified BalanceBooks example can be found here:
> https://github.com/plaflamme/incubator-tephra/commit/6caeff5d76f193cdf2e0ad861059bc19cf894705
> The logs of the various services can be found here: 
> [https://gist.github.com/plaflamme/238a6539da9da1ac3a2e313e05ee82eb]
> In my experimentation, I've found that this problem usually happens after the 
> transaction server fails to write its snapshot during shutdown. It may be 
> unrelated, but this exception usually shows up in the logs around the time 
> when the issue is seen on the client.
> This is the relevant section in the logs: 
> [https://gist.github.com/plaflamme/238a6539da9da1ac3a2e313e05ee82eb#file-tephra-service-hbase-master-log-L1360-L1878]
> I've sent a message to the mailing list about this problem: 
> [https://mail-archives.apache.org/mod_mbox/tephra-dev/201812.mbox/%3CCAFYs7S9BgPs_A-1E-K3Q06BfiO7W%3Dh3mZD6-vkjb14R29C-bKw%40mail.gmail.com%3E]
>  Sorry for the noise, but perhaps a JIRA issue is more appropriate for 
> discussing this.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to