Bug 2092950

+++ This bug was initially created as a clone of Bug #1918527 +++

Description of problem:

Suppose that `pcs resource unmanage --monitor` is run on two individual resources in a group, and then `pcs resource manage --monitor` is run on one of those resources.

Both resources' monitor operations get re-enabled in this case. However, only the resource where the `manage --monitor` command was run gets re-managed.

Demo (with dummy1 and dummy2 inside dummygrp):
~~~
# pcs resource unmanage --monitor dummy1 && pcs resource unmanage --monitor dummy2

# pcs resource config dummy1 dummy2 | egrep 'Resource:|Meta|monitor'
 Resource: dummy1 (class=ocf provider=heartbeat type=Dummy)
  Meta Attrs: is-managed=false
              monitor enabled=false interval=10s timeout=20s (dummy1-monitor-interval-10s)
 Resource: dummy2 (class=ocf provider=heartbeat type=Dummy)
  Meta Attrs: is-managed=false
              monitor enabled=false interval=10s timeout=20s (dummy2-monitor-interval-10s)

# pcs resource manage --monitor dummy1

# pcs resource config dummy1 dummy2 | egrep 'Resource:|Meta|monitor'
 Resource: dummy1 (class=ocf provider=heartbeat type=Dummy)
              monitor interval=10s timeout=20s (dummy1-monitor-interval-10s)
 Resource: dummy2 (class=ocf provider=heartbeat type=Dummy)
  Meta Attrs: is-managed=false
              monitor interval=10s timeout=20s (dummy2-monitor-interval-10s)
~~~

IMO, the least surprising behavior would be to both re-manage and re-enable monitor on **only** the resource for which we ran `pcs resource manage --monitor`. Ideally the manage command should perform the inverse of the unmanage command.

However, even if we make the opposite case (to re-manage and re-enable monitor on all resources in the group), then we should be consistent. I don't see a reason to re-enable monitor on all resources when we only re-manage one resource.

-----

The cause:

With some debugging added:
~~~
# pcs resource manage --monitor dummy1
Iterating over _find_resources_expand_tags_or_raise:
resource_el is dummy1
Iterating over to_manage_set:
resource_el is dummy1
resource_el is dummygrp
Iterating over primitives_set:
resource_el is dummy1
resource_el is dummy2
~~~

find_resources_to_manage() grabs both the primitive ID and the group ID in accordance with the comment block below its doc string.
  - (https://github.com/ClusterLabs/pcs/blob/v0.10.7/pcs/lib/cib/resource/common.py#L237-L243)

Then find_primitives() updates primitives_set with all the primitives in dummygrp (dummy1 and dummy2).

We iterate over these primitives and enable all their monitor operations.


Note: I acknowledge that specifying the behavior for pcs resource enable/manage is quite difficult and that decisions often have unintended consequences, as we also discussed in BZ 1875632. In this case, I believe we should enable monitors on a particular primitive only if either (a) the primitive was specified explicitly on the command line or (b) the primitive's group/parent ID was specified explicitly on the command line.

In my example, that would look like: "enable monitor on dummy1, but don't enable monitor for dummy2 because neither dummy2 nor dummygrp was passed as a CLI argument."

(Without looking further, I'm not sure how tags would factor into this.)

-----

Version-Release number of selected component (if applicable):

pcs-0.10.6-4.el8

-----

How reproducible:

Always

-----

Steps to Reproduce:
1. Configure two resources in a group (e.g., dummy1 and dummy2 in dummygrp).
2. Run `pcs resource unmanage --monitor` on both resources.
3. Run `pcs resource manage --monitor` on only one resource (e.g., dummy1).

-----

Actual results:

The monitor operation is re-enabled for both resources. Only the specified resource is re-managed.

-----

Expected results:

The monitor operation is re-enabled only for the specified resource. Only the specified resource is re-managed.

-----

Additional info:

As a workaround, users can omit the `--monitor` keyword and instead manipulate the monitor operation manually.