HomeClusterLabs Projects

Doc: Clusters from Scratch: Improve DRBD walkthrough

Description

Doc: Clusters from Scratch: Improve DRBD walkthrough

  • Remove syncer section, which was made obsolete in DRBD 8.4. Move verify-alg to net section.
  • Use new-style configuration instead of legacy 8.4-style. These changes aren't necessary but seem like best practices (although the user guide itself becomes inconsistent later, reverting to what it calls the old style at times).
    • Use device minor 1 instead of device /dev/drbd1.
    • Move protocol C to the net section where it belongs.
    • Assign explicit an node-id to each node.
    • Use a connection section instead of specifying addresses in on <host> sections.
    • Quote strings.
    • Ref: https://linbit.com/drbd-user-guide/drbd-guide-9_0-en/#s-drbdconf-example, especially Listings 2 and 4 and the bullet points beneath.
  • Integrate with Pacemaker fencing.
    • Use fencing resource-and-stonith.
    • Use handlers for fence-peer and unfence-peer.
      • Note: The DRBD user guide instructs to use the handler outdate-peer /sbin/kill-other-node.sh. But that script does not exist, and the DRBD source code comments say outdate-peer was replaced by fence-peer years ago. I'm using the settings from the fencing resource-only description. The crm-fence-peer.9.sh script seems to have at least been designed to work with fencing resource-and-stonith.

Signed-off-by: Reid Wahl <nrwahl@protonmail.com>

Details

Provenance
nrwahl2Authored on Aug 3 2022, 3:45 AM
Parents
rPdde6e4278d0e: Doc: Clusters from Scratch: Correct node for mkfs.gfs2
Branches
Unknown
Tags
Unknown