Avoid incorrectly marking zfs tasks as changed (#2454)

* Avoid incorrectly marking zfs tasks as changed

The zfs module will incorrectly mark certain tasks as having been
changed. For example, if a dataset has a quota of "1G" and the user
changes it to "1024M", the actual quota vale has not changed, but since
the module is doing a simple string comparison between "1G" and "1024M",
it marks the step as "changed".

Instead of trying to handle all the corner cases of zfs (another example
is when the zpool "altroot" property has been set), this change simply
compares the output of "zfs-get" from before and after "zfs-set" is
called

* update changelog format

* Update changelogs/fragments/2454-detect_zfs_changed.yml

Co-authored-by: Felix Fontein <felix@fontein.de>

* add note about check_mode

* Update plugins/modules/storage/zfs/zfs.py

Co-authored-by: Felix Fontein <felix@fontein.de>

* Update plugins/modules/storage/zfs/zfs.py

Co-authored-by: Felix Fontein <felix@fontein.de>

* clarify check mode qualifications

* rephrase to avoid hypothetical

Co-authored-by: Felix Fontein <felix@fontein.de>

changelogs/fragments/2454-detect_zfs_changed.yml

@@ -0,0 +1,2 @@
+bugfixes:
+  - zfs - certain ZFS properties, especially sizes, would lead to a task being falsely marked as "changed" even when no actual change was made (https://github.com/ansible-collections/community.general/issues/975, https://github.com/ansible-collections/community.general/pull/2454).

plugins/modules/storage/zfs/zfs.py

@@ -37,6 +37,12 @@ options:
       - A dictionary of zfs properties to be set.
       - See the zfs(8) man page for more information.
     type: dict
+notes:
+  - C(check_mode) is supported, but in certain situations it may report a task
+    as changed that will not be reported as changed when C(check_mode) is disabled.
+    For example, this might occur when the zpool C(altroot) option is set or when
+    a size is written using human-readable notation, such as C(1M) or C(1024K),
+    instead of as an unqualified byte count, such as C(1048576).
 author:
 - Johan Wiren (@johanwiren)
 '''
@@ -184,9 +190,7 @@ class Zfs(object):
             return
         cmd = [self.zfs_cmd, 'set', prop + '=' + str(value), self.name]
         (rc, out, err) = self.module.run_command(cmd)
-        if rc == 0:
-            self.changed = True
-        else:
+        if rc != 0:
             self.module.fail_json(msg=err)
 
     def set_properties_if_changed(self):
@@ -194,15 +198,25 @@ class Zfs(object):
         for prop, value in self.properties.items():
             if current_properties.get(prop, None) != value:
                 self.set_property(prop, value)
+        if self.module.check_mode:
+            return
+        updated_properties = self.get_current_properties()
+        for prop in self.properties:
+            value = updated_properties.get(prop, None)
+            if value is None:
+                self.module.fail_json(msg="zfsprop was not present after being successfully set: %s" % prop)
+            if current_properties.get(prop, None) != value:
+                self.changed = True
 
     def get_current_properties(self):
-        cmd = [self.zfs_cmd, 'get', '-H']
+        cmd = [self.zfs_cmd, 'get', '-H', '-p', '-o', "property,value,source"]
         if self.enhanced_sharing:
             cmd += ['-e']
         cmd += ['all', self.name]
         rc, out, err = self.module.run_command(" ".join(cmd))
         properties = dict()
-        for prop, value, source in [l.split('\t')[1:4] for l in out.splitlines()]:
+        for line in out.splitlines():
+            prop, value, source = line.split('\t')
             # include source '-' so that creation-only properties are not removed
             # to avoids errors when the dataset already exists and the property is not changed
             # this scenario is most likely when the same playbook is run more than once