回复: CTDB and per record persistency

[Martin Schwenke]

Hi,

On Wed, 3 Jan 2024 07:28:58 +0000, 陈 方进 <sharingfun520 at outlook.com>
wrote:

> The file copying scenario  involves a cluster built using CTDB, with
> two machines sharing a common public IP. When you copy files to the
> shared directory and the node with the public IP experiences a
> failure with subsequent failover, the file copying process continues
> seamlessly without any error displayed on the Windows system. Do you
> have any means to ensure that file copying can proceed normally even
> during transparent failover events?

Now I understand that you were talking about copying data from the
cluster.  I thought you were talking about replicating data between
nodes in the cluster.  Thanks for clarifying!

Currently CTDB uses tickle ACKs and gratuitous ARPs to attempt
failover.  While this works "seamlessly" for some applications, it
doesn't work in some cases.

I wasn't disagreeing with you: with modern SMB, the best way of
achieving transparent failover would be to support persistent file
handles.  Unfortunately, until now, this hasn't reached the top of
anyone's priority queue, so it hasn't been implemented in clustered
Samba.

peace & happiness,
martin