[Drbd-dev] [PATCH] drbd_state: dont clear NEW_CUR_UUID when re-gained quorum

[Philipp Reisner]

Hi Dongsheng,

thanks for that. It looks correct. I have applied it to the drbd-9.0 branch.

best regards,
 Phil

On Fri, Sep 18, 2020 at 7:22 AM Dongsheng Yang
<dongsheng.yang at easystack.cn> wrote:
>
> We cant clear NEW_CUR_UUID when we re-gain quorum, because
> there is a secondary offline.
>
> E.g:
> There is a cluster with 3 nodes, 1 primary (node-1), 2 secondary (node-2, node-3)
>
> (1) all uptodate, primary with quorum=2, quorum-minimum-redundancy=2.
> (2) node-1 network error -> node-1 lost quorum
> (3) node-3 down.
> (4) node-1 network recovery -> node-1 regain quorum, clear NEW_CUR_UUID (node-1 uptodate, node-2 uptodate, node-3 offline)
> (5) write data on primary node. -> as NEW_CUR_UUID cleared, the uuid is old.
> (6) node-3 up. -> as the uuid in primary is old, same with what in node-3. there is no-sync.
>
> Then we will loss the new data in node-3.
>
> To fix it, dont clear NEW_CUR_UUID in (4).
>
> Fixes: aaaa257b837a26ac4a38f2e86632d682fc57a2
> Signed-off-by: Dongsheng Yang <dongsheng.yang at easystack.cn>
> ---
>  drbd/drbd_state.c | 1 -
>  1 file changed, 1 deletion(-)
>
> diff --git a/drbd/drbd_state.c b/drbd/drbd_state.c
> index 5b1744a..358bcc0 100644
> --- a/drbd/drbd_state.c
> +++ b/drbd/drbd_state.c
> @@ -2592,7 +2592,6 @@ static void finish_state_change(struct drbd_resource *resource, struct completio
>
>                 if (!device->have_quorum[OLD] && device->have_quorum[NEW]) {
>                         clear_bit(PRIMARY_LOST_QUORUM, &device->flags);
> -                       clear_bit(NEW_CUR_UUID, &device->flags);
>                 }
>         }
>
> --
> 1.8.3.1
>