[Discuss] systemd reboot

[smallm at sdf.org (Mike Small)]

Richard Pieri <richard.pieri at gmail.com> writes:

> The old "sync;sync;sync;halt" mantra is folklore from the days before we
> had a shutdown/reboot command which does this for us. The first sync
> flushes any dirty buffers, the second blocks waiting for the first to
> complete ensuring that there are no dirty buffers when the system goes
> down, and the third... makes us feel good (it has no technical benefit).
>
> This doesn't work as expected today because most drives lie about
> committing writes to permanent storage. The second sync won't block
> unless the size of data in dirty kernel buffers exceeds the drives'
> write cache capacity and then it will block only long enough for that
> ratio to flip. If the system restarts, loses power, whatever, when the
> drives' on-board caches have not been committed then there will be data
> loss. The Linux kernel code which guarantees that writes are committed
> doesn't actually work because it relies on drives not lying about their
> cache commits.

Thanks, this is great info. Curiously, in my local reproduction of the
issue the lying hardware involved is qemu's virtio simulated disk. So
maybe their simulation is super realistic, eh?

-- 
Mike Small
smallm at sdf.org