Because the ClusterMon and Stateful agents use crm_mon/crm_attribute
respectively.

Mirror upstream commit 57ce5b5e (it no
longer uses libcrmcluster).

Revert 808de7cc, because the upstream
code reorganization ending in ca25b4e9
removed the obstacle.

https://bugs.clusterlabs.org/show_bug.cgi?id=5488

This effectively reverts 7443ebc6,
because upstream commit 6af89fe4 made
it unnecessary.

Update to upstream version '2.1.4'
with Debian dir a77659a34de408641a237187a0037d5749549f57

Low: fencer: assume unknown status from older fencers' history replies

1b9e2896 made stonith_api_history() pull the exit reason from the fencer's
reply by pulling the full result with stonith__xe_get_result() and using only
the exit reason.

However stonith__xe_get_result() will log an error if no result is present,
which will be the case if the client is on a newer Pacemaker Remote node
proxied to an older fencer that doesn't set a result.

That's not a big deal, but avoid it anyway by pulling the exit reason alone.

be0a0b65 added and retrieved fencing results from fencing history entries'
XML. However stonith__xe_get_result() would log an error and set an error
status if no result is set, as is the case with older fencers. So, avoid that
function, and instead parse the results directly. Since a legacy rc was never
added by any version, we don't have to look for one.

Fix OCF agent meta-data

It was undocumented, and the OCF RNG schema requires at least one parameter

The OCF standard uses an RNG now, not a DTD, and validation complains about a
nonexistent DTD.

The idea is this could be run manually at release time. It's not enabled as
part of "make check" because an external RNG file must be supplied.

regression introduced in 2.1.3 by be030855

Release 2.1.4-rc1

In the case of self-fencing with topology, handle_fence_request() will
overwrite F_STONITH_OPERATION in the original request XML, which invalidates
the request.op pointer created by stonith_command(). The fix is to make
request.op a copy.

Regression introduced in 2.1.3 by 067d655e

Regression in 2.1.3 introduced by b09f16eb

This was papered over by the separate regression in cts-lab introduced by
4a6d1879, which caused target-by-attribute to no longer be tested, and which
was recently fixed by caf9a3bc.

This command gets the node ID from a cluster node based on the corosync
config file.  4a6d1879 introduced a bug where the command can fail
because there's no longer a space between the command and the file name
awk is operating on.  So, add that space back.

No idea what has been broken as a result of this, if anything.

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

If a fencing action fails, but a later equivalent action succeeds, crm_mon will
show "(a later attempt succeeded)" when displaying the failed action.

Previously, an action was considered equivalent if it had the same target,
action, and delegate. However the delegate (i.e. the node executing the fence
action) doesn't really matter for this purpose.