On Tue, May 28, 2024 at 4:53 AM Tony Liu <tonyliu0592@hotmail.com> wrote:
Hi,
Say, the source image is being updated and data is mirrored to destination continuously. Suddenly, networking of source is down and destination will be promoted and used to restore the VM. Is that going to cause any FS issue and, for example, fsck needs to be invoked to check and repair FS? Is there any integrity check during journal-based mirror to avoid "partial" update caused by networking issue?
Hi Tony, (Apologies for a late reply, this was marked as spam for some reason.) rbd-mirror daemon on the destination replays the journal in a point-in-time, crash-consistent fashion. If the destination is promoted while the source and/or the network is down, the promotion would have to be forced with "rbd mirror image promote --force" command. As such, fsck may have some work to do because the filesystem is not cleanly unmounted and the source is not cleanly demoted. In general, there is no way to avoid "partial" on-disk state from the filesystem perspective here because crash consistency is weaker than filesystem consistency. Thanks, Ilya