rolling_update: disable FSMap sanity checks

Disable FSMap sanity checks before upgrade to prevent
mon crash(1)

(1) https://access.redhat.com/solutions/7020523

fixes: https://bugzilla.redhat.com/show_bug.cgi?id=2277756

Signed-off-by: Teoman ONAY <tonay@ibm.com>

diff --git a/infrastructure-playbooks/rolling_update.yml b/infrastructure-playbooks/rolling_update.yml
index 5914880b477872c053b0caf7b7c76fff733997e1..e90d048cd0d421041613aa64e8dfc2250749b345 100644 (file)
--- a/infrastructure-playbooks/rolling_update.yml
+++ b/infrastructure-playbooks/rolling_update.yml
@@ -148,6 +148,15 @@
   become: True
   gather_facts: false
   tasks:
+    # workaround to prevent mon crash https://access.redhat.com/solutions/7020523
+    - name: Disable FSMap sanity checks before starting the upgrade
+      run_once: true
+      ceph_config:
+        action: set
+        who: mon
+        option: mon_mds_skip_sanity
+        value: true
+
     - name: upgrade ceph mon cluster
       block:
         - name: remove ceph aliases
@@ -1194,6 +1203,19 @@
         path: /etc/ceph/{{ cluster }}-crushmap
         state: absent
 
+- name: Remove FSMap sanity checks after the upgrade
+  hosts: "{{ mon_group_name|default('mons') }}"
+  become: True
+  gather_facts: false
+  tasks:
+    # workaround to prevent mon crash https://access.redhat.com/solutions/7020523
+    - name: Remove FSMap sanity checks
+      run_once: true
+      ceph_config:
+        action: rm
+        who: mon
+        option: mon_mds_skip_sanity
+
 - name: show ceph status
   hosts: "{{ mon_group_name|default('mons') }}"
   tags: always

diff --git a/library/ceph_config.py b/library/ceph_config.py
new file mode 100644 (file)
index 0000000..c82c8cf
--- /dev/null
+++ b/library/ceph_config.py
@@ -0,0 +1,208 @@
+# Copyright Red Hat
+# SPDX-License-Identifier: Apache-2.0
+# Author: Guillaume Abrioux <gabrioux@redhat.com>
+
+from __future__ import absolute_import, division, print_function
+__metaclass__ = type
+
+from ansible.module_utils.basic import AnsibleModule  # type: ignore
+try:
+    from ansible.module_utils.ca_common import exit_module, generate_cmd, fatal, is_containerized  # type: ignore
+except ImportError:
+    from module_utils.ca_common import exit_module, generate_cmd, fatal, is_containerized  # type: ignore
+
+import datetime
+import json
+
+ANSIBLE_METADATA = {
+    'metadata_version': '1.1',
+    'status': ['preview'],
+    'supported_by': 'community'
+}
+
+DOCUMENTATION = '''
+---
+module: ceph_config
+short_description: set ceph config
+version_added: "2.10"
+description:
+    - Set Ceph config options.
+options:
+    fsid:
+        description:
+            - the fsid of the Ceph cluster to interact with.
+        required: false
+    image:
+        description:
+            - The Ceph container image to use.
+        required: false
+    action:
+        description:
+            - whether to get or set the parameter specified in 'option'
+        required: false
+        default: 'set'
+    who:
+        description:
+            - which daemon the configuration should be set to
+        required: true
+    option:
+        description:
+            - name of the parameter to be set
+        required: true
+    value:
+        description:
+            - value of the parameter
+        required: true if action is 'set'
+
+author:
+    - Guillaume Abrioux <gabrioux@redhat.com>
+'''
+
+EXAMPLES = '''
+- name: set osd_memory_target for osd.0
+  ceph_config:
+    action: set
+    who: osd.0
+    option: osd_memory_target
+    value: 5368709120
+
+- name: set osd_memory_target for host ceph-osd-02
+  ceph_config:
+    action: set
+    who: osd/host:ceph-osd-02
+    option: osd_memory_target
+    value: 5368709120
+
+- name: get osd_pool_default_size value
+  ceph_config:
+    action: get
+    who: global
+    option: osd_pool_default_size
+    value: 1
+'''
+
+RETURN = '''#  '''
+
+
+def set_option(module,
+               who,
+               option,
+               value,
+               container_image=None):
+
+    args = []
+    args.extend([who, option, value])
+
+    cmd = generate_cmd(sub_cmd=['config', 'set'],
+                       args=args,
+                       cluster=module.params.get('cluster'),
+                       container_image=container_image)
+
+    rc, out, err = module.run_command(cmd)
+
+    return rc, cmd, out.strip(), err
+
+
+def rm_option(module,
+              who,
+              option,
+              container_image=None):
+
+    args = []
+    args.extend([who, option])
+
+    cmd = generate_cmd(sub_cmd=['config', 'rm'],
+                       args=args,
+                       cluster=module.params.get('cluster'),
+                       container_image=container_image)
+
+    rc, out, err = module.run_command(cmd)
+
+    return rc, cmd, out.strip(), err
+
+
+def get_config_dump(module, container_image=None):
+    cmd = generate_cmd(sub_cmd=['config', 'dump', '--format', 'json'],
+                       args=[],
+                       cluster=module.params.get('cluster'),
+                       container_image=container_image)
+    rc, out, err = module.run_command(cmd)
+    if rc:
+        fatal(message=f"Can't get current configuration via `ceph config dump`.Error:\n{err}", module=module)
+    out = out.strip()
+    return rc, cmd, out, err
+
+
+def get_current_value(who, option, config_dump):
+    for config in config_dump:
+        if config['section'] == who and config['name'] == option:
+            return config['value']
+    return None
+
+
+def main() -> None:
+    module = AnsibleModule(
+        argument_spec=dict(
+            who=dict(type='str', required=True),
+            action=dict(type='str', required=False, choices=['get', 'set', 'rm'], default='set'),
+            option=dict(type='str', required=True),
+            value=dict(type='str', required=False),
+            fsid=dict(type='str', required=False),
+            image=dict(type='str', required=False),
+            cluster=dict(type='str', required=False, default='ceph')
+        ),
+        supports_check_mode=True,
+        required_if=[['action', 'set', ['value']]]
+    )
+
+    # Gather module parameters in variables
+    who = module.params.get('who')
+    option = module.params.get('option')
+    value = module.params.get('value')
+    action = module.params.get('action')
+
+    container_image = is_containerized()
+
+    if module.check_mode:
+        module.exit_json(
+            changed=False,
+            stdout='',
+            cmd=[],
+            stderr='',
+            rc=0,
+            start='',
+            end='',
+            delta='',
+        )
+
+    startd = datetime.datetime.now()
+    changed = False
+
+    rc, cmd, out, err = get_config_dump(module, container_image=container_image)
+    config_dump = json.loads(out)
+    current_value = get_current_value(who, option, config_dump)
+
+    if action == 'set':
+        if current_value and value.lower() == current_value.lower():
+            out = 'who={} option={} value={} already set. Skipping.'.format(who, option, value)
+        else:
+            rc, cmd, out, err = set_option(module, who, option, value, container_image=container_image)
+            changed = True
+    elif action == 'get':
+        if current_value is None:
+            out = ''
+            err = 'No value found for who={} option={}'.format(who, option)
+        else:
+            out = current_value
+    elif action == 'rm':
+        if current_value:
+            rc, cmd, out, err = rm_option(module, who, option, container_image=container_image)
+            changed = True
+
+    exit_module(module=module, out=out, rc=rc,
+                cmd=cmd, err=err, startd=startd,
+                changed=changed)
+
+
+if __name__ == '__main__':
+    main()