[sheepdog-users] fix object consistency

刘忠鑫 liuzxzju at gmail.com
Mon Apr 11 12:27:44 CEST 2016


Hi all,

Are there any mechanisms to fix the consistency of the replication except
manual check?

I find following statements from sheepdog wiki (
https://github.com/sheepdog/sheepdog/wiki/Sheepdog-Design):

> The consistency of the replication could be broken if the node crash
during forwarding write I/O requests, so the gateway tries to
> fix the consistency when it reads object for the first time; read a whole
object data from one of the target node and overwrite all
> replicas with it.

But in my simple test, I didn't see the fixing process. I ran a 4-node
3-copies local cluster in my PC, wrote data into a vdi using `dog vdi
write` and killed node 0 (the default gateway) while writing. Then, one
data object became inconsistent.

I restarted node 0,and used `dog vdi read -p $PORT` to read the
inconsistent data object from each node. The data object was still
inconsistent after read. No automatic fixing happened.

Finally I used `dog vdi check`. It fixed the problem successfully.

So, if I do not check manually, will the inconsistency problem stay here
forever?

------------------
Best regards,
Zhongxin Liu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.wpkg.org/pipermail/sheepdog-users/attachments/20160411/6930da14/attachment.html>


More information about the sheepdog-users mailing list