On 24 Mar 2021, at 7:03, Sven Schwedas wrote:
If that's not acceptable, you need some form of file system/block
layer replication (DRBD, Ceph, Gluster, …) to get the spool data to
a spare.
Have you actually used such a storage system for the Postfix queue?
I've used SAN storage (RAID 10 on the storage presented to hosts as
simple LUNs) for the queue and successfully tested failover to a cold
spare in a pulled-plug scenario, but never had an actual failover occur
with messages in the queue outside of intentional tests. So it *can*
work and *probably* works...
I don't think you can run it as a *hot*spare with both postfix
instances accessing the same spool data, so it'd have to be started on
demand?
That is also my understanding. Given the breakage seen with non-Postfix
software trying to manipulate messages in the queue, I doubt that
Postfix components that expect to be the only one of their kind would
deal well with instances on other machines making changes.
--
Bill Cole
b...@scconsult.com or billc...@apache.org
(AKA @grumpybozo and many *@billmail.scconsult.com addresses)
Not Currently Available For Hire
