Why shouldn't you prefer the default way of doing things in the software you use? If you don't trust in their choice, why should you run the software at all?
ZFS / cache=writeback / safe ?
- Thread starter trendco
- Start date
In my understanding that's expected behavior.
The "write-back"-mechanism just tells your application that data has been written - while in fact it is not!
Default and recommended is "none" as caching mode for your VMs when using ZFS. ZFS will use its ARC as read cache anyways, so no need to cache the same data multiple times in RAM.
And you can always lose data on an power outage, when not forcing every async write to be handled as a sync write, as your hosts RAM and the RAM cache of all HDDs and consumer SSDs will loose the cached async writes. Thats why you should always run your server with an UPS. They don't cost that much, starting at 50€ for something usable. By far the cheapest option to minimize data loss if you compare that to buying alot of enterprise grade hardware like ECC RAM, enterprise SSDs, redundant PSUs, HW raid with cache and BBU, ... . And don't forget to set up something like NUT, so the server will do a automatic graceful shutdown in case of an power outage.
And you can always lose data on an power outage, when not forcing every async write to be handled as a sync write, as your hosts RAM and the RAM cache of all HDDs and consumer SSDs will loose the cached async writes. Thats why you should always run your server with an UPS. They don't cost that much, starting at 50€ for something usable. By far the cheapest option to minimize data loss if you compare that to buying alot of enterprise grade hardware like ECC RAM, enterprise SSDs, redundant PSUs, HW raid with cache and BBU, ... . And don't forget to set up something like NUT, so the server will do a automatic graceful shutdown in case of an power outage.
Last edited: