MCOL-5806: added ability to start node in read-only mode

Author: mariadb-AlexanderPresniakov

cmapi/cmapi_server/constants.py

@@ -4,6 +4,7 @@
 """
 import os
 from typing import NamedTuple
+from enum import Enum
 
 
 # default MARIADB ColumnStore config path
@@ -51,6 +52,16 @@
     CMAPI_INSTALL_PATH, 'cmapi_server/SingleNode.xml'
 )
 
+class MCSProgs(Enum):
+    STORAGE_MANAGER = 'StorageManager'
+    WORKER_NODE = 'workernode'
+    CONTROLLER_NODE = 'controllernode'
+    PRIM_PROC = 'PrimProc'
+    EXE_MGR = 'ExeMgr'
+    WRITE_ENGINE_SERVER = 'WriteEngineServer'
+    DML_PROC = 'DMLProc'
+    DDL_PROC = 'DDLProc'
+
 # constants for dispatchers
 class ProgInfo(NamedTuple):
     """NamedTuple for some additional info about handling mcs processes."""
@@ -64,17 +75,17 @@ class ProgInfo(NamedTuple):
 # on top level of process handling
 # mcs-storagemanager starts conditionally inside mcs-loadbrm, but should be
 # stopped using cmapi
-ALL_MCS_PROGS = {
+ALL_MCS_PROGS: dict[str, ProgInfo] = {
     # workernode starts on primary and non primary node with 1 or 2 added
     # to subcommand (DBRM_Worker1 - on primary, DBRM_Worker2 - non primary)
-    'StorageManager': ProgInfo(15, 'mcs-storagemanager', '', False, 1),
-    'workernode': ProgInfo(13, 'mcs-workernode', 'DBRM_Worker{}', False, 1),
-    'controllernode': ProgInfo(11, 'mcs-controllernode', 'fg', True),
-    'PrimProc': ProgInfo(5, 'mcs-primproc', '', False, 1),
-    'ExeMgr': ProgInfo(9, 'mcs-exemgr', '', False, 1),
-    'WriteEngineServer': ProgInfo(7, 'mcs-writeengineserver', '', False, 3),
-    'DMLProc': ProgInfo(3, 'mcs-dmlproc', '', False),
-    'DDLProc': ProgInfo(1, 'mcs-ddlproc', '', False),
+    MCSProgs.STORAGE_MANAGER.value: ProgInfo(15, 'mcs-storagemanager', '', False, 1),
+    MCSProgs.WORKER_NODE.value: ProgInfo(13, 'mcs-workernode', 'DBRM_Worker{}', False, 1),
+    MCSProgs.CONTROLLER_NODE.value: ProgInfo(11, 'mcs-controllernode', 'fg', True),
+    MCSProgs.PRIM_PROC.value: ProgInfo(5, 'mcs-primproc', '', False, 1),
+    MCSProgs.EXE_MGR.value: ProgInfo(9, 'mcs-exemgr', '', False, 1),
+    MCSProgs.WRITE_ENGINE_SERVER.value: ProgInfo(7, 'mcs-writeengineserver', '', False, 3),
+    MCSProgs.DML_PROC.value: ProgInfo(3, 'mcs-dmlproc', '', False),
+    MCSProgs.DDL_PROC.value: ProgInfo(1, 'mcs-ddlproc', '', False),
 }
 
 # constants for docker container dispatcher

cmapi/cmapi_server/controllers/endpoints.py

@@ -401,7 +401,8 @@ def put_config(self):
                 MCSProcessManager.stop_node(
                     is_primary=node_config.is_primary_node(),
                     use_sudo=use_sudo,
-                    timeout=request_timeout
+                    timeout=request_timeout,
+                    is_read_only=node_config.is_read_only(),
                 )
             except CMAPIBasicError as err:
                 raise_422_error(
@@ -417,7 +418,7 @@ def put_config(self):
             )
             if in_maintenance_state():
                 module_logger.info(
-                    'Maintaninance state is active in new config. '
+                    'Maintenance state is active in new config. '
                     'MCS processes should not be started.'
                 )
                 cherrypy.engine.publish('failover', False)
@@ -430,6 +431,7 @@ def put_config(self):
                     MCSProcessManager.start_node(
                         is_primary=node_config.is_primary_node(),
                         use_sudo=use_sudo,
+                        is_read_only=node_config.is_read_only(),
                     )
                 except CMAPIBasicError as err:
                     raise_422_error(
@@ -633,7 +635,8 @@ def put_start(self):
         try:
             MCSProcessManager.start_node(
                 is_primary=node_config.is_primary_node(),
-                use_sudo=use_sudo
+                use_sudo=use_sudo,
+                is_read_only=node_config.is_read_only(),
             )
         except CMAPIBasicError as err:
             raise_422_error(
@@ -668,7 +671,8 @@ def put_shutdown(self):
             MCSProcessManager.stop_node(
                 is_primary=node_config.is_primary_node(),
                 use_sudo=use_sudo,
-                timeout=timeout
+                timeout=timeout,
+                is_read_only=node_config.is_read_only(),
             )
         except CMAPIBasicError as err:
             raise_422_error(

cmapi/cmapi_server/failover_agent.py

@@ -95,7 +95,7 @@ def enterStandbyMode(self, test_mode = False):
         try:
             # TODO: remove test_mode condition and add mock for testing
             if not test_mode:
-                MCSProcessManager.stop_node(is_primary=nc.is_primary_node())
+                MCSProcessManager.stop_node(is_primary=nc.is_primary_node(), is_read_only=nc.is_read_only())
             logger.info(
                 'FA.enterStandbyMode(): successfully stopped node.'
             )

cmapi/cmapi_server/handlers/cluster.py

@@ -196,7 +196,8 @@ def process_shutdown():
     @staticmethod
     def add_node(
         node: str, config: str = DEFAULT_MCS_CONF_PATH,
-        logger: logging.Logger = logging.getLogger('cmapi_server')
+        logger: logging.Logger = logging.getLogger('cmapi_server'),
+        read_only: bool = False,
     ) -> dict:
         """Method to add node to MCS CLuster.
 
@@ -207,6 +208,8 @@ def add_node(
         :type config: str, optional
         :param logger: logger, defaults to logging.getLogger('cmapi_server')
         :type logger: logging.Logger, optional
+        :param read_only: add node in read-only mode, defaults to False
+        :type read_only: bool, optional
         :raises CMAPIBasicError: on exception while starting transaction
         :raises CMAPIBasicError: if transaction start isn't successful
         :raises CMAPIBasicError: on exception while adding node
@@ -216,7 +219,7 @@ def add_node(
         :return: result of adding node
         :rtype: dict
         """
-        logger.debug(f'Cluster add node command called. Adding node {node}.')
+        logger.debug('Cluster add node command called. Adding node %s in %s mode.', node, 'read-only' if read_only else 'read-write')
 
         response = {'timestamp': str(datetime.now())}
         transaction_id = get_id()
@@ -238,13 +241,18 @@ def add_node(
         try:
             add_node(
                 node, input_config_filename=config,
-                output_config_filename=config
+                output_config_filename=config,
+                read_only=read_only,
             )
             if not get_dbroots(node, config):
-                add_dbroot(
-                    host=node, input_config_filename=config,
-                    output_config_filename=config
-                )
+                if not read_only:  # Read-only nodes don't own dbroots
+                    add_dbroot(
+                        host=node, input_config_filename=config,
+                        output_config_filename=config
+                    )
+                else:
+                    logger.debug("Node %s is read-only, skipping dbroot addition", node)
+
         except Exception as err:
             rollback_transaction(transaction_id, cs_config_filename=config)
             raise CMAPIBasicError('Error while adding node.') from err

cmapi/cmapi_server/helpers.py

@@ -513,6 +513,10 @@ def get_desired_nodes(config=DEFAULT_MCS_CONF_PATH):
     return [ node.text for node in nodes ]
 
 
+def get_read_only_nodes(root) -> list[str]:
+    return [node.text for node in root.findall("./ReadOnlyNodes/Node")]
+
+
 def in_maintenance_state(config=DEFAULT_MCS_CONF_PATH):
     nc = NodeConfig()
     root = nc.get_current_config_root(config, upgrade=False)
@@ -549,6 +553,7 @@ def get_dbroots(node, config=DEFAULT_MCS_CONF_PATH):
     dbroots = []
     smc_node = root.find('./SystemModuleConfig')
     mod_count = int(smc_node.find('./ModuleCount3').text)
+
     for i in range(1, mod_count+1):
         ip_addr = smc_node.find(f'./ModuleIPAddr{i}-1-3').text
         hostname = smc_node.find(f'./ModuleHostName{i}-1-3').text
@@ -568,6 +573,12 @@ def get_dbroots(node, config=DEFAULT_MCS_CONF_PATH):
                 dbroots.append(
                     smc_node.find(f"./ModuleDBRootID{i}-{j}-3").text
                 )
+
+    if dbroots and nc.is_read_only():
+        logger = logging.getLogger("dbroots")
+        logger.warning("Config contains dbroots %s for this read-only node, ignoring", dbroots)
+        return []
+
     return dbroots
 
 

cmapi/cmapi_server/managers/process.py

@@ -7,7 +7,8 @@
 import psutil
 
 from cmapi_server.exceptions import CMAPIBasicError
-from cmapi_server.constants import MCS_INSTALL_BIN, ALL_MCS_PROGS
+from cmapi_server.constants import MCS_INSTALL_BIN, ALL_MCS_PROGS, MCSProgs
+from cmapi_server.process_dispatchers.base import BaseDispatcher
 from cmapi_server.process_dispatchers.systemd import SystemdDispatcher
 from cmapi_server.process_dispatchers.container import (
     ContainerDispatcher
@@ -18,7 +19,7 @@
 from mcs_node_control.models.process import Process
 
 
-PROCESS_DISPATCHERS = {
+PROCESS_DISPATCHERS: dict[str, type[BaseDispatcher]] = {
     'systemd': SystemdDispatcher,
     # could be used in docker containers and OSes w/o systemd
     'container': ContainerDispatcher,
@@ -404,37 +405,46 @@ def is_node_processes_ok(
         return set(node_progs) == set(p['name'] for p in running_procs)
 
     @classmethod
-    def start_node(cls, is_primary: bool, use_sudo: bool = True):
+    def start_node(cls, is_primary: bool, use_sudo: bool = True, is_read_only: bool = False) -> None:
         """Start mcs node processes.
 
         :param is_primary: is node primary or not, defaults to True
         :type is_primary: bool
         :param use_sudo: use sudo or not, defaults to True
         :type use_sudo: bool, optional
+        :param is_read_only: if true, doesn't start WriteEngine
+        :type is_read_only: bool, optional
         :raises CMAPIBasicError: immediately if one mcs process not started
         """
         for prog_name in cls._get_sorted_progs(is_primary):
             if (
                     cls.dispatcher_name == 'systemd'
-                    and prog_name == 'StorageManager'
+                    and prog_name == MCSProgs.STORAGE_MANAGER.value
             ):
                 # TODO: MCOL-5458
                 logging.info(
                     f'Skip starting {prog_name} with systemd dispatcher.'
                 )
                 continue
             # TODO: additional error handling
-            if prog_name == 'controllernode':
+            if prog_name == MCSProgs.CONTROLLER_NODE.value:
                 cls._wait_for_workernodes()
-            if prog_name in ('DMLProc', 'DDLProc'):
+            if prog_name in (MCSProgs.DML_PROC.value, MCSProgs.DDL_PROC.value):
                 cls._wait_for_controllernode()
+            if is_read_only and prog_name == MCSProgs.WRITE_ENGINE_SERVER.value:
+                logging.debug('Node is in read-only mode, not starting WriteEngine')
+                continue
             if not cls.start(prog_name, is_primary, use_sudo):
                 logging.error(f'Process "{prog_name}" not started properly.')
                 raise CMAPIBasicError(f'Error while starting "{prog_name}".')
 
     @classmethod
     def stop_node(
-        cls, is_primary: bool, use_sudo: bool = True, timeout: int = 10
+        cls,
+        is_primary: bool,
+        use_sudo: bool = True,
+        timeout: int = 10,
+        is_read_only: bool = False,
     ):
         """Stop mcs node processes.
 
@@ -444,20 +454,25 @@ def stop_node(
         :type use_sudo: bool, optional
         :param timeout: timeout for DMLProc gracefully stop using DBRM, seconds
         :type timeout: int
+        :param is_read_only: if true, doesn't stop WriteEngine
+        :type is_read_only: bool, optional
         :raises CMAPIBasicError: immediately if one mcs process not stopped
         """
         # Every time try to stop all processes no matter primary it or slave,
         # so use full available list of processes. Otherwise, it could cause
         # undefined behaviour when primary gone and then recovers (failover
         # triggered 2 times).
         for prog_name in cls._get_sorted_progs(True, reverse=True):
+            if is_read_only and prog_name == MCSProgs.WRITE_ENGINE_SERVER.value:
+                logging.debug('Node is in read-only mode, not stopping WriteEngine')
+                continue
             if not cls.stop(prog_name, is_primary, use_sudo):
                 logging.error(f'Process "{prog_name}" not stopped properly.')
                 raise CMAPIBasicError(f'Error while stopping "{prog_name}"')
 
     @classmethod
-    def restart_node(cls, is_primary: bool, use_sudo: bool):
+    def restart_node(cls, is_primary: bool, use_sudo: bool, is_read_only: bool = False):
         """TODO: For next releases."""
         if cls.get_running_mcs_procs():
-            cls.stop_node(is_primary, use_sudo)
-        cls.start_node(is_primary, use_sudo)
+            cls.stop_node(is_primary, use_sudo, is_read_only)
+        cls.start_node(is_primary, use_sudo, is_read_only)

cmapi/cmapi_server/node_manipulation.py

@@ -61,7 +61,8 @@ def switch_node_maintenance(
 def add_node(
     node: str, input_config_filename: str = DEFAULT_MCS_CONF_PATH,
     output_config_filename: Optional[str] = None,
-    rebalance_dbroots: bool = True
+    rebalance_dbroots: bool = True,
+    read_only: bool = False,
 ):
     """Add node to a cluster.
 
@@ -95,14 +96,23 @@ def add_node(
     try:
         if not _replace_localhost(c_root, node):
             pm_num = _add_node_to_PMS(c_root, node)
-            _add_WES(c_root, pm_num, node)
+
+            if not read_only:
+                _add_WES(c_root, pm_num, node)
+            else:
+                logging.info("Node is read-only, skipping WES addition")
+                _add_read_only_node(c_root, node)
+
             _add_DBRM_Worker(c_root, node)
             _add_Module_entries(c_root, node)
             _add_active_node(c_root, node)
             _add_node_to_ExeMgrs(c_root, node)
             if rebalance_dbroots:
-                _rebalance_dbroots(c_root)
-                _move_primary_node(c_root)
+                if not read_only:
+                    _rebalance_dbroots(c_root)
+                    _move_primary_node(c_root)
+                else:
+                    logging.debug("Node is read-only, skipping dbroots rebalancing")
     except Exception:
         logging.error(
             'Caught exception while adding node, config file is unchanged',
@@ -140,7 +150,11 @@ def remove_node(
 
         if len(active_nodes) > 1:
             pm_num = _remove_node_from_PMS(c_root, node)
-            _remove_WES(c_root, pm_num)
+
+            is_read_only = node in helpers.get_read_only_nodes(c_root)
+            if not is_read_only:
+                _remove_WES(c_root, pm_num)
+
             _remove_DBRM_Worker(c_root, node)
             _remove_Module_entries(c_root, node)
             _remove_from_ExeMgrs(c_root, node)
@@ -151,7 +165,7 @@ def remove_node(
                 # TODO: unspecific name, need to think of a better one
                 _remove_node(c_root, node)
 
-            if rebalance_dbroots:
+            if rebalance_dbroots and not is_read_only:
                 _rebalance_dbroots(c_root)
                 _move_primary_node(c_root)
         else:
@@ -359,12 +373,16 @@ def __remove_helper(parent_node, node):
 
 def _remove_node(root, node):
     '''
-    remove node from DesiredNodes, InactiveNodes, and ActiveNodes
+    remove node from DesiredNodes, InactiveNodes, ActiveNodes and (if present) ReadOnlyNodes
     '''
 
     for n in (root.find("./DesiredNodes"), root.find("./InactiveNodes"), root.find("./ActiveNodes")):
         __remove_helper(n, node)
 
+    read_only_nodes = root.find("./ReadOnlyNodes")
+    if read_only_nodes is not None:
+        __remove_helper(read_only_nodes, node)
+
 
 # This moves a node from ActiveNodes to InactiveNodes
 def _deactivate_node(root, node):
@@ -972,6 +990,19 @@ def _add_WES(root, pm_num, node):
     etree.SubElement(wes_node, "Port").text = "8630"
 
 
+def _add_read_only_node(root, node) -> None:
+    """Add node name to ReadOnlyNodes if it's not already there"""
+    read_only_nodes = root.find("./ReadOnlyNodes")
+    if read_only_nodes is None:
+        read_only_nodes = etree.SubElement(root, "ReadOnlyNodes")
+    else:
+        for n in read_only_nodes.findall("./Node"):
+            if n.text == node:
+                return
+
+    etree.SubElement(read_only_nodes, "Node").text = node
+
+
 def _add_DBRM_Worker(root, node):
     '''
     find the highest numbered DBRM_Worker entry, or one that isn't used atm
@@ -1074,7 +1105,7 @@ def _add_node_to_PMS(root, node):
 
     return new_pm_num
 
-def _replace_localhost(root, node):
+def _replace_localhost(root: etree.Element, node: str) -> bool:
     # if DBRM_Controller/IPAddr is 127.0.0.1 or localhost,
     # then replace all instances, else do nothing.
     controller_host = root.find('./DBRM_Controller/IPAddr')