On Mon, Sep 24, 2018 at 9:24 AM Germano Veit Michel <germano(a)redhat.com> wrote:
Hi,
I'm working an a tool (vdsm-tool update-volume) to make modifying SD metadata easier
and more importantly, safer. This is very useful to recover from failed LSMs or snapshot
issues.
The plan is to use the VDSM API (modified by some of these patches) and add a tool
(vdsm-tool) that talks to the API and modifies the volumes metadata as required by the
user. Currently this is done manually, i.e.: looking at MD_XXX tags, doing dd, sed and
then dd back to the storage. Any wrong argument (like a skip in place of a seek) can ruin
the entire metadata, so this tool can be quite handy.
The code is not necessarily 100% finished yet, but I've been testing this for some
time and it seems ok from a functional point of view. I'm just not sure everything I
did (especially inside VDSM, example 94366) is correct. Your comments on what can/should
be improved are very welcome at this point. Please see this series and help reviewing it.
https://gerrit.ovirt.org/#/q/topic:update-volume+(status:open+OR+status:m...
https://gerrit.ovirt.org/#/c/93258/
I am not a maintainer of vdsm.storage, but I can say that I'm missing
a high-level description of the change that you are suggesting, and
its motivation. When do you see a need to manually change the metadata
of volumes? Shouldn't we fix the bug that causes this need?
I personally have a deep resentment to a "force" flags - not just
here, but everywhere. It is never clear what is being forced. Some
things cannot or should not be forced.
One last note: a CI+1 gives a positive psychological vibe to your reviewer.