<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <p>Thanks a lot for your suggestions (Robert and Lars), it took a
      while before I was able to try them on virtual machines. I hope
      you don't mind that I reply to both of you in one mail -- I messed
      up my mail delivery options (now corrected).<br>
    </p>
    <p>I've added my latest drbd config below, for reference.<br>
    </p>
    <blockquote type="cite"
      cite="mid:6A2938E1-5AAD-45BF-BCB5-B7C6520CFB0C@linbit.com">
      <blockquote type="cite">
        <pre class="moz-quote-pre" wrap="">I can't find any sequence of commands that can convince drbd (or pacemaker) that I *want* to use outdated data.
</pre>
      </blockquote>
      <pre class="moz-quote-pre" wrap="">This should work:
drbdadm del-peer tapas:fims1
drbdadm primary —force tapas
</pre>
    </blockquote>
    <p>This seems to work (get to UpToDate state) briefly, until the
      next time the pacemaker drbd monitor runs, which 'demotes' the
      resource again to it's original state.<br>
    </p>
    <p><tt>Failed Resource Actions:</tt><tt><br>
      </tt><tt>* drbd_monitor_20000 on vmnbiaas2 'master' (8): call=84,
        status=complete, exitreason='',</tt><tt><br>
      </tt><tt>    last-rc-change='Wed Oct  9 14:49:06 2019',
        queued=0ms, exec=0ms</tt><br>
    </p>
    <p>The corosync logs are difficult to follow, so I'm not sure how I
      can get pacemaker to accept the trickery done behind its back..<br>
    </p>
    <p>Lars wrote:<br>
    </p>
    <pre>Alternatively, you could *add* a suitable fencing constraint to your sole survivor node, which should make the fencing succeed.

You could tell the crm-fence-peer.9.sh fencing handler that an --unreachable-peer-is-outdated.
(Manually. From a root shell. That switch is not effective from within the drbd configuration; for reasons).
</pre>
    <p>I tried this, after finding what the command should look like in
      /var/log/messages:</p>
    <p><tt>DRBD_BACKING_DEV_0=/dev/mapper/centos-drbd
        DRBD_CONF=/etc/drbd.conf DRBD_LL_DISK=/dev/mapper/centos-drbd
        DRBD_MINOR=0 DRBD_MINOR_0=0 DRBD_MY_ADDRESS=172.17.5.62
        DRBD_MY_AF=ipv4 DRBD_MY_NODE_ID=1 DRBD_NODE_ID_0=vmnbiaas1
        DRBD_NODE_ID_1=vmnbiaas2 DRBD_PEER_ADDRESS=172.17.5.61
        DRBD_PEER_AF=ipv4 DRBD_PEER_NODE_ID=0 DRBD_RESOURCE=tapas
        DRBD_VOLUME=0 UP_TO_DATE_NODES=0x00000002
        /usr/lib/drbd/crm-fence-peer.9.sh --unreachable-peer-is-outdated</tt></p>
    <p>This failed as follows:<br>
    </p>
    <p><tt>Oct  9 14:29:42 vmnbiaas2 crm-fence-peer.9.sh[6153]: WARNING
        Found &lt;cib crm_feature_set="3.0.14"
        validate-with="pacemaker-2.10" epoch="48" num_updates="23"
        admin_epoch="0" cib-last-written="Wed Oct  9 14:07:22 2019"
        update-origin="vmnbiaas1" update-client="cibadmin"
        update-user="root" have-quorum="0" dc-uuid="1"</tt></p>
    <p><tt>Oct  9 14:29:42 vmnbiaas2 crm-fence-peer.9.sh[6153]: WARNING
        I don't have quorum; did not place the constraint!</tt></p>
    <p>OK, while I'm experimenting, I quick-hacked the script to use<br>
    </p>
    <p><tt>fail_if_no_quorum=false</tt></p>
    <p>After which the error changes to<br>
    </p>
    <p><tt>Oct  9 14:38:13 vmnbiaas2 crm-fence-peer.9.sh[7579]: WARNING
        some peer is UNCLEAN, my disk is not UpToDate, did not place the
        constraint!</tt><br>
      <br>
    </p>
    <p>Cheers!</p>
    <p>     Rob<br>
    </p>
    <p><br>
    </p>
    <p>resource tapas {<br>
        protocol C;<br>
      <br>
        startup {<br>
          wfc-timeout            0;    ## Infinite!<br>
          outdated-wfc-timeout    120;<br>
          degr-wfc-timeout        120;  ## 2 minutes.<br>
        }<br>
      <br>
        disk {<br>
          on-io-error detach;<br>
        }<br>
      <br>
        handlers {<br>
          split-brain "/opt/sol/tapas/bin/split-brain-helper.sh";<br>
      <br>
          fence-peer "/usr/lib/drbd/crm-fence-peer.9.sh";<br>
          unfence-peer "/usr/lib/drbd/crm-unfence-peer.9.sh";<br>
        }<br>
      <br>
        net {<br>
          fencing resource-only;<br>
      <br>
      #    after-sb-0pri       discard-least-changes;<br>
        }<br>
      <br>
        device        /dev/drbd0;<br>
        disk            /dev/mapper/centos-drbd;<br>
        meta-disk        internal;<br>
      <br>
        on vmnbiaas1 {<br>
          address    172.17.5.61:7789;<br>
        }<br>
      <br>
        on vmnbiaas2 {<br>
          address    172.17.5.62:7789;<br>
        }<br>
      }<br>
      <br>
    </p>
  </body>
</html>