lp:drbdmanage
- Get this branch:
- bzr branch lp:drbdmanage
Branch information
Import details
This branch is an import of the HEAD branch of the Git repository at git://git.linbit.com/drbdmanage.git.
Last successful import was .
Recent revisions
- 878. By Roland Kammerer
-
add more function to wrappers
These wrappers are used to ping/pong between server and cli, especially
to tell the client that the server is currently not ready.When drbdmanaged was shut down, and it was started with "edit-config",
it ran into a dbus error because the signature was wrong because the
entry in the wrappers was missing.Fix up the ones used in "edit-config".
- 877. By Roland Kammerer
-
leader: implement different error handling strategies
So far actions are retried until a (configurable) maximum error count is
reached. Then the resource/assignment is ignored until a "resume-all"
gets triggered manually.This commit introduces a more fine-grained contol over the
error-handling strategy. The strategy is set via "edit-config", where
the following settings are possible:"error-
handling- strategy" is one of:
"resume-no": the current behavior
"resume-max-backoff" : exponential backoff until a maximum is reached
"resume-interval" : resume with the given interval The given strategy is also the key to set a value for
"resume-max-backoff" and "resume-interval". Times are given in minutes. For example:
error-handling-strategy = resume-interval
resume-interval = 30Issues the equivalent of "resume-all" every 30 minutes if there are
failures.error-handling-
strategy = resume-max-backoff
resume-max-backoff = 65The first resume is triggered after 2 minutes, the next after 4
additional minutes, the next after 8 minutes and so on. If the next
target time is larger than "resume-max-backoff" (in the given example
when it is set to 128), it is reset to 2.The strategy is only executed if there are failures drbdmanage detected,
otherwise all counters are reset to 0. - 876. By Roland Kammerer
-
allow to keep ctrlvol on shutdown
This adds a new option to "drbdadm shutdown", namely "-c" to not touch
the control volume. This might be handy if the current leader is already
selected (== .drbdctrl is Primary) and the leader should be restarted
without triggering a leader election. - 873. By Roland Kammerer
-
allow longer snapshot names
In the wild (OpenNebula in this case), we saw that these tools like to
generate very long names (UUIDs over UUIDs over...). Sometimes these are
too long for our rather arbitrary limit. Increase it. - 872. By Roland Kammerer
-
provide zvol2/zvol_thinlv2 storage plugins
The existing zvol plugin mimics the LVM world. There a snapshot also
creates a block device. To have the same behavior in the zfs world, a
snapshot also creates a zfs clone (i.e, the block device). On restore,
we created a second snapshot from that block device and then used that
one as the new backing device for the new resource.This mimicking of LVM is not necessary. Create one snapshot on
"create-snapshot". On "restore-snapshot" use that one to create the new
backing device.The old version works, but generates unnecessary snapshots/clones. New
users should use this version, for example for the thinly-provisioned
version:
storage-plugin = drbdmanage.storage. zvol2_thinlv. ZvolThinLv2 These two versions of the zfs plugin are incompatible. If you only used
zfs (without snapshots), you can switch to the new one. - 871. By Roland Kammerer
-
storage backends add notes about remove_snapshot()
The idea was that every plugin can provide/overwrite a method hat is
called to remove a snapshot. In the current design all of them called
remove_blockdevice( ). The storage core actually already called
remove_blockdevice( ). Strictly speaking this is a bug, because it
bypasses the plugins remove_snapshot(). As all plugins only provide a
wrapper, just add a note and don't touch all the existing plugins. This
can be fixed when we need it, but document it for now.
Branch metadata
- Branch format:
- Branch format 7
- Repository format:
- Bazaar repository format 2a (needs bzr 1.16 or later)