starlingx/update
Code Issues Proposed changes

Merge "Restrict ceph mon hook execution"

Browse Source
This commit is contained in:
Zuul
2024-11-13 19:23:04 +00:00
committed by Gerrit Code Review
parent e1b18b9235 ddd9410d2a
commit e3e454c208
1 changed files with 31 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
software/software/agent_hooks.py
View File

@@ -304,34 +304,44 @@ class RemoveKubernetesConfigSymlinkHook(BaseHook):
raise raise
# TODO(heitormatsui): delete in the future, not needed for stx-10 -> <future-releases>
class RemoveCephMonHook(BaseHook): class RemoveCephMonHook(BaseHook):
""" """
Remove additional ceph-mon added for each controller Remove additional ceph-mon added for each controller
""" """
PMON_FILE = "/etc/pmon.d/ceph-fixed-mon.conf" PMON_FILE = "/etc/pmon.d/ceph-fixed-mon.conf"
def __init__(self, attrs):
super().__init__(attrs)
self._major_release = None
if "major_release" in attrs:
self._major_release = attrs.get("major_release")
def run(self): def run(self):
system_type = utils.get_platform_conf("system_type") # (DX only) on 22.12 there is 1 mon, on 24.09 there are 3
system_mode = utils.get_platform_conf("system_mode") # so only in 24.09 -> 22.12 rollback this hook is needed
nodetype = utils.get_platform_conf("nodetype") if self._major_release == "22.12":
# additional monitors were added only for AIO-DX system_type = utils.get_platform_conf("system_type")
if (system_type == constants.SYSTEM_TYPE_ALL_IN_ONE and system_mode = utils.get_platform_conf("system_mode")
system_mode != constants.SYSTEM_MODE_SIMPLEX and nodetype = utils.get_platform_conf("nodetype")
nodetype == constants.CONTROLLER): # additional monitors were added only for AIO-DX
cmd_remove_mon_controller_0 = ["timeout", "30", "ceph", "mon", "rm", "controller-0"] if (system_type == constants.SYSTEM_TYPE_ALL_IN_ONE and
cmd_remove_mon_controller_1 = ["timeout", "30", "ceph", "mon", "rm", "controller-1"] system_mode != constants.SYSTEM_MODE_SIMPLEX and
try: nodetype == constants.CONTROLLER):
subprocess.check_call(cmd_remove_mon_controller_0) cmd_remove_mon_controller_0 = ["timeout", "30", "ceph", "mon", "rm", "controller-0"]
subprocess.check_call(cmd_remove_mon_controller_1) cmd_remove_mon_controller_1 = ["timeout", "30", "ceph", "mon", "rm", "controller-1"]
LOG.info("Removed mon.controller-0 and mon.controller-1 from ceph cluster.") try:
except subprocess.CalledProcessError as e: subprocess.check_call(cmd_remove_mon_controller_0)
LOG.exception("Failure removing mon.controller-0 and mon.controller-1 from ceph cluster: %s" % str(e)) subprocess.check_call(cmd_remove_mon_controller_1)
raise LOG.info("Removed mon.controller-0 and mon.controller-1 from ceph cluster.")
try: except subprocess.CalledProcessError as e:
os.unlink(self.PMON_FILE) LOG.exception("Failure removing mon.controller-0 and mon.controller-1 from ceph cluster: %s" % str(e))
LOG.info("Removed %s from pmon." % self.PMON_FILE) raise
except FileNotFoundError: try:
pass # ignore if link doesn't exist os.unlink(self.PMON_FILE)
LOG.info("Removed %s from pmon." % self.PMON_FILE)
except FileNotFoundError:
pass # ignore if link doesn't exist
class RestartKubeApiServer(BaseHook): class RestartKubeApiServer(BaseHook):