• Icon: Bug Bug
    • Resolution: Done-Errata
    • Icon: Major Major
    • rhel-9.4
    • rhel-9.3.0
    • sanlock
    • None
    • sanlock-3.9.1-1.el9
    • None
    • Moderate
    • Rebase
    • sst_logical_storage
    • ssg_platform_storage
    • 17
    • 22
    • 3
    • QE ack, Dev ack
    • False
    • Hide

      None

      Show
      None
    • None
    • None
    • None

      What were you trying to do that didn't work?

       

      We've not updated the sanlock package in a long time (I think since 9.0) since there have not been major issues (also, RHEV is not used in RHEL9, significantly diminishing the sanlock usage.)  But, there are now enough improvements that we should really be including.  The notable changes are:

       

      sanlock: use helper to set max_sectors_kb
      The original addition of setting this sysfs value was testing with the sanlock daemon running as root, but if the daemon is not run as root, then every time it starts a lockspace and tries to set this sysfs value it fails.  These errors appear in the log files and the sysfs values not being set which means the original i/o improvements are not effective.
      
      watchdog timeout configuration
      This was requested from an upstream user.  Like the configurable io timeout, the configurable watchdog timeout will be possible to change, but alternative combinations of either timeout would be up to the user to validate as meeting the requirements of their environment (e.g, the i/o delays that occur on their storage system.)
      
      wdmd: close watchdog when not used
      This is an internal improvement that won't really have any external visibility. If wdmd is running but not being actively used, then the host will not be subject to watchdog resets if the daemon is killed.  It's always good to avoid reseting the host if it's not necessary.
       
      sanlock: skip short delay in delta lease reacquire
      This is a simple optimization to skip the default 20 sec delay in cases where the same host is reacquiring the same host_id.  The original optimization of this delay was too conservative, but since sanlock wasn't typically being configured in way to take advantage of it, we never bothered to improve it.  In future we want to configure constant host names so it will be useful to properly optimize this case.
      
      sanlock: use product_uuid for host name
      Unless the sanlock host name had been configured by the user, sanlock would generate a random UUID as the host name each time the daemon started.  This meant that a host would always have a new name, making it difficult to track, and extending the delta lease delays due to the different name.  Now use the product_uuid (when available) instead of generating random UUIDs for the host name.

       

      
      commit 34af016589c3cb733731ab33b9e249b4943fbcb9
      Author: David Teigland <teigland@redhat.com>
      Date:   Tue Mar 22 17:13:12 2022 -0500
      
      
          sanlock: use helper to set max_sectors_kb
          
          When the sanlock daemon is not run as the root user, it doesn't
          have permission to write to sysfs max_sectors_kb, so use the
          root helper process to do that.
      
      
      commit 748e8325fd0b2e09469c76f584b8e08c1ef03ca6
      Author: David Teigland <teigland@redhat.com>
      Date:   Fri Oct 7 15:51:07 2022 -0500
      
      
          watchdog timeout configuration
          
          Make the watchdog timeout configurable.  The watchdog device must
          support the configured value, and all hosts must use the same value.
          The io_timeout should usually be configured by a similar factor.
          
          Set watchdog_fire_timeout and io_timeout in sanlock.conf, e.g.
          watchdog_fire_timeout=30
          io_timeout=5
          
          The defaults remain watchdog_fire_timeout 60 and io_timeout 10.
          
          wdmd --trytimeout <sec> can be used test if the watchdog device
          supports a certain timeout value.
      
      
      commit 45563a6e0c9d388f58178e03c9dd0b2708b2969c
      Author: David Teigland <teigland@redhat.com>
      Date:   Mon Nov 14 16:04:39 2022 -0600
      
      
          wdmd: close watchdog when not used
          
          Previously, the watchdog would remain open with keepalives
          even after all clients (sanlock lockspaces) were removed.
          This was an unnecessary liability in case it couldn't be
          kept alive while not needed.  It also prevented sanlock
          from being restarted using a new watchdog timeout.
      
      
      commit a239c898c12a92cb5fb74f0eb74bc5919db45b94 (HEAD -> fixed-host-id)
      Author: David Teigland <teigland@redhat.com>
      Date:   Fri Dec 8 17:48:02 2023 -0600
      
      
          sanlock: skip short delay in delta lease reacquire
          
          When a host with the same host name reacquires the
          same host_id, the "fast" delta lease reacquire was
          still applying the short delay (default 20s) which
          should not be necessary.  Reduce this further to
          just 1 second when the lease is also free.
      
      
      commit 5b0eef4aa13e5a28116e815113fe4953d77c2d6a (HEAD -> fixed-host-id)
      Author: David Teigland <teigland@redhat.com>
      Date:   Mon Dec 11 17:16:32 2023 -0600
      
      
          sanlock: use product_uuid for host name
          
          When there is no host name specified on the command line
          or in the config file, use the value from
          /sys/devices/virtual/dmi/id/product_uuid instead of
          generating a random UUID each time the daemon is started.
          If no product_uuid exists, the random UUID will be generated
          for the host name.

       

      Please provide the package NVR for which bug is seen:

      How reproducible:

      Steps to reproduce

      1.  
      2.  
      3.  

      Expected results

      Actual results

        1. testing.txt
          4 kB
          David Teigland

            teigland@redhat.com David Teigland
            teigland@redhat.com David Teigland
            David Teigland David Teigland
            Cluster QE Cluster QE
            Votes:
            0 Vote for this issue
            Watchers:
            5 Start watching this issue

              Created:
              Updated:
              Resolved: