]> git-server-git.apps.pok.os.sepia.ceph.com Git - ceph-ci.git/commitdiff
mgr: isolated CherryPy to prevent global state sharing wip-nia-isolate-cherrypy
authorNizamudeen A <nia@redhat.com>
Thu, 5 Feb 2026 10:42:47 +0000 (16:12 +0530)
committerNizamudeen A <nia@redhat.com>
Fri, 6 Feb 2026 07:42:23 +0000 (13:12 +0530)
as the modules are now being loaded onto the main interpreter (see
https://github.com/ceph/ceph/pull/66244), the
cherrypy is getting hit with an issue where its global state is being
affecting all the modules updating the cherrypy config simultaneously in
the same tree.

So i am adding a CherryPyMgr which manages all the independent servers
that will be created across all modules. This CherryPyMgr will create
its own server instances by utilizing cherrypy's WSGI Server and
eliminates the global state sharing. Each module or app can create their
own tree and start an adapter which will open an independent server for
that app.

Fixes: https://tracker.ceph.com/issues/74643, https://tracker.ceph.com/issues/74543
Signed-off-by: Nizamudeen A <nia@redhat.com>
13 files changed:
.github/labeler.yml
ceph.spec.in
debian/ceph-mgr.install
src/pybind/mgr/CMakeLists.txt
src/pybind/mgr/cephadm/agent.py
src/pybind/mgr/cephadm/http_server.py
src/pybind/mgr/cephadm/services/service_discovery.py
src/pybind/mgr/cherrypy_mgr.py [new file with mode: 0644]
src/pybind/mgr/dashboard/module.py
src/pybind/mgr/dashboard/services/auth/auth.py
src/pybind/mgr/dashboard/tools.py
src/pybind/mgr/prometheus/module.py
src/pybind/mgr/tox.ini

index 932b2a23018c7a3a35e75021607712ae96f6036a..1dcdf585963a0eb55a738cf82e9a2f9282e5dba8 100644 (file)
@@ -55,6 +55,7 @@ mgr:
   - src/pybind/mgr/ceph_module.pyi
   - src/pybind/mgr/mgr_module.py
   - src/pybind/mgr/mgr_util.py
+  - src/pybind/mgr/cherrypy_mgr.py
   - src/pybind/mgr/object_format.py
   - src/pybind/mgr/requirements.txt
   - src/pybind/mgr/tox.ini
index 18cd57b26833e0181627f59ff6f3c901c0b6d298..e9a3bac48dc5f16a4437e6402c044e693877f46c 100644 (file)
@@ -1920,6 +1920,7 @@ fi
 %{_datadir}/ceph/mgr/mgr_module.*
 %{_datadir}/ceph/mgr/mgr_util.*
 %{_datadir}/ceph/mgr/object_format.*
+%{_datadir}/ceph/mgr/cherrypy_mgr.*
 %{_unitdir}/ceph-mgr@.service
 %{_unitdir}/ceph-mgr.target
 %attr(750,ceph,ceph) %dir %{_localstatedir}/lib/ceph/mgr
index 11a4a9ce4e2adf6aaf69ba8dcf1191736f2791bc..803dcf6bcd8f7673619a0d12a200e2cec3c59427 100644 (file)
@@ -3,3 +3,4 @@ usr/bin/ceph-mgr
 usr/share/ceph/mgr/mgr_module.*
 usr/share/ceph/mgr/mgr_util.*
 usr/share/ceph/mgr/object_format.*
+usr/share/ceph/mgr/cherrypy_mgr.*
index 9e900f859d701c1536b98266772eaff008ee3327..b83ba1b387ea237c4c860f6544253421fbb2250d 100644 (file)
@@ -58,5 +58,5 @@ set(mgr_modules
 install(DIRECTORY ${mgr_modules}
   DESTINATION ${CEPH_INSTALL_DATADIR}/mgr
   ${mgr_module_install_excludes})
-install(FILES mgr_module.py mgr_util.py object_format.py
+install(FILES mgr_module.py mgr_util.py object_format.py cherrypy_mgr.py
   DESTINATION ${CEPH_INSTALL_DATADIR}/mgr)
index cec4ab0ab81a50cad7d404d9cb77eeba97d85265..8c7111e5f497d46fb89dd569126bbf2bfbcea8af 100644 (file)
@@ -1,11 +1,4 @@
-try:
-    import cherrypy
-    from cherrypy._cpserver import Server
-except ImportError:
-    # to avoid sphinx build crash
-    class Server:  # type: ignore
-        pass
-
+import cherrypy
 import json
 import logging
 import socket
@@ -33,18 +26,6 @@ if TYPE_CHECKING:
     from cephadm.module import CephadmOrchestrator
 
 
-def cherrypy_filter(record: logging.LogRecord) -> bool:
-    blocked = [
-        'TLSV1_ALERT_DECRYPT_ERROR'
-    ]
-    msg = record.getMessage()
-    return not any([m for m in blocked if m in msg])
-
-
-logging.getLogger('cherrypy.error').addFilter(cherrypy_filter)
-cherrypy.log.access_log.propagate = False
-
-
 CEPHADM_AGENT_CERT_DURATION = (365 * 5)
 
 
@@ -57,13 +38,21 @@ class AgentEndpoint:
         self.key_file: IO[bytes]
         self.cert_file: IO[bytes]
 
-    def configure_routes(self) -> None:
-        conf = {'/': {'tools.trailing_slash.on': False}}
-
-        cherrypy.tree.mount(self.host_data, '/data', config=conf)
-        cherrypy.tree.mount(self.node_proxy_endpoint, '/node-proxy', config=conf)
-
-    def configure_tls(self, server: Server) -> None:
+    def get_cherrypy_config(self) -> Dict:
+        config = {
+            '/': {
+                'tools.trailing_slash.on': False
+            }
+        }
+        return config
+
+    def configure_routes(self, config) -> List[tuple]:
+        return [
+            (self.host_data, '/data', config),
+            (self.node_proxy_endpoint, '/node-proxy', config),
+        ]
+
+    def configure_tls(self) -> Dict[str, str]:
         self.mgr.cert_mgr.register_self_signed_cert_key_pair(CephadmAgent.TYPE)
         tls_pair = self._get_agent_certificates()
         self.cert_file = tempfile.NamedTemporaryFile()
@@ -75,7 +64,10 @@ class AgentEndpoint:
         self.key_file.flush()  # pkey_tmp must not be gc'ed
 
         verify_tls_files(self.cert_file.name, self.key_file.name)
-        server.ssl_certificate, server.ssl_private_key = self.cert_file.name, self.key_file.name
+        return {
+            'cert': self.cert_file.name,
+            'key': self.key_file.name,
+        }
 
     def _get_agent_certificates(self) -> TLSCredentials:
         host = self.mgr.get_hostname()
@@ -97,12 +89,14 @@ class AgentEndpoint:
                 self.server_port += 1
         self.mgr.log.error(f'Cephadm agent could not find free port in range {max_port - 150}-{max_port} and failed to start')
 
-    def configure(self) -> None:
-        self.host_data = HostData(self.mgr, self.server_port, self.server_addr)
-        self.configure_tls(self.host_data)
+    def configure(self, tree) -> None:
+        self.host_data = HostData(self.mgr)
+        ssl_info = self.configure_tls()
         self.node_proxy_endpoint = NodeProxyEndpoint(self.mgr)
-        self.configure_routes()
+        config = self.get_cherrypy_config()
+        mount_specs = self.configure_routes(config)
         self.find_free_port()
+        return config, ssl_info, mount_specs, (self.server_addr, self.server_port)
 
 
 class NodeProxyEndpoint:
@@ -636,22 +630,11 @@ class NodeProxyEndpoint:
         return results
 
 
-class HostData(Server):
+class HostData:
     exposed = True
 
-    def __init__(self, mgr: "CephadmOrchestrator", port: int, host: str):
+    def __init__(self, mgr: "CephadmOrchestrator"):
         self.mgr = mgr
-        super().__init__()
-        self.socket_port = port
-        self.socket_host = host
-        self.subscribe()
-
-    def stop(self) -> None:
-        # we must call unsubscribe before stopping the server,
-        # otherwise the port is not released and we will get
-        # an exception when trying to restart it
-        self.unsubscribe()
-        super().stop()
 
     @cherrypy.tools.allow(methods=['POST'])
     @cherrypy.tools.json_in()
index baa00a3eb5ac833a6af788989e12c60a653ebb0a..68198e2a2645fd64d579a5ca31d6be73a7a856e5 100644 (file)
@@ -2,6 +2,8 @@ import cherrypy
 import threading
 import logging
 from typing import TYPE_CHECKING
+from cherrypy import _cptree
+from cherrypy_mgr import CherryPyMgr
 
 from cephadm.agent import AgentEndpoint
 from cephadm.services.service_discovery import ServiceDiscovery
@@ -12,42 +14,20 @@ if TYPE_CHECKING:
     from cephadm.module import CephadmOrchestrator
 
 
-def cherrypy_filter(record: logging.LogRecord) -> bool:
-    blocked = [
-        'TLSV1_ALERT_DECRYPT_ERROR'
-    ]
-    msg = record.getMessage()
-    return not any([m for m in blocked if m in msg])
-
-
-logging.getLogger('cherrypy.error').addFilter(cherrypy_filter)
-cherrypy.log.access_log.propagate = False
-
-
 class CephadmHttpServer(threading.Thread):
     def __init__(self, mgr: "CephadmOrchestrator") -> None:
         self.mgr = mgr
         self.agent = AgentEndpoint(mgr)
         self.service_discovery = ServiceDiscovery(mgr)
         self.cherrypy_shutdown_event = threading.Event()
+        self.cherrypy_restart_event = threading.Event()
         self._service_discovery_port = self.mgr.service_discovery_port
         security_enabled, _, _ = self.mgr._get_security_config()
         self.security_enabled = security_enabled
+        self.agent_adapter = None
+        self.server_adapter = None
         super().__init__(target=self.run)
 
-    def configure_cherrypy(self) -> None:
-        cherrypy.config.update({
-            'environment': 'production',
-            'engine.autoreload.on': False,
-        })
-
-    def configure(self) -> None:
-        self.configure_cherrypy()
-        self.agent.configure()
-        self.service_discovery.configure(self.mgr.service_discovery_port,
-                                         self.mgr.get_mgr_ip(),
-                                         self.security_enabled)
-
     def config_update(self) -> None:
         self.service_discovery_port = self.mgr.service_discovery_port
         security_enabled, _, _ = self.mgr._get_security_config()
@@ -76,27 +56,73 @@ class CephadmHttpServer(threading.Thread):
         self.restart()
 
     def restart(self) -> None:
-        cherrypy.engine.stop()
-        cherrypy.server.httpserver = None
-        self.configure()
-        cherrypy.engine.start()
+        self.cherrypy_restart_event.set()
 
     def run(self) -> None:
+        def start_servers():
+            # start service discovery server
+            sd_config, sd_ssl_info = self.service_discovery.configure(
+                self.mgr.service_discovery_port,
+                self.mgr.get_mgr_ip(),
+                self.security_enabled
+            )
+            sd_port = self._service_discovery_port
+            sd_ip = self.mgr.get_mgr_ip()
+            self.mgr.log.info(f'Starting service discovery server on {sd_ip}:{sd_port}...')
+
+            sd_tree = _cptree.Tree()
+            sd_tree.mount(self.service_discovery, "/sd", config=sd_config)
+            adapter_sd, _ = CherryPyMgr.mount(
+                sd_tree,
+                (sd_ip, int(sd_port)),
+                ssl_info=sd_ssl_info
+            )
+
+            # start agent server
+            agent_config, agent_ssl_info, agent_mounts, bind_addr = self.agent.configure()
+            self.mgr.log.info(f'Starting agent server on {bind_addr[0]}:{bind_addr[1]}...')
+
+            agent_tree = _cptree.Tree()
+            agent_tree.mount(self.agent, "/", config=agent_config)
+
+            for app, path, conf in agent_mounts:
+                agent_tree.mount(app, path, config=conf)
+
+            adapter_agent, _ = CherryPyMgr.mount(
+                agent_tree,
+                bind_addr,
+                ssl_info=agent_ssl_info
+            )
+            return adapter_sd, adapter_agent
+
         try:
-            self.mgr.log.debug('Starting cherrypy engine...')
-            self.configure()
-            cherrypy.server.unsubscribe()  # disable default server
-            cherrypy.engine.start()
-            self.mgr.log.debug('Cherrypy engine started.')
-            self.mgr._kick_serve_loop()
-            # wait for the shutdown event
-            self.cherrypy_shutdown_event.wait()
-            self.cherrypy_shutdown_event.clear()
-            cherrypy.engine.stop()
-            cherrypy.server.httpserver = None
-            self.mgr.log.debug('Cherrypy engine stopped.')
+            self.server_adapter, self.agent_adapter = start_servers()
+            self.mgr.log.info('Cherrypy server started successfully.')
         except Exception as e:
-            self.mgr.log.error(f'Failed to run cephadm http server: {e}')
+            self.mgr.log.error(f'Failed to start cherrypy server: {e}')
+            if self.server_adapter: self.server_adapter.stop()
+            if self.agent_adapter: self.agent_adapter.stop()
+            return
+
+        while not self.cherrypy_shutdown_event.is_set():
+            if self.cherrypy_restart_event.wait(timeout=0.5):
+                self.cherrypy_restart_event.clear()
+                self.mgr.log.debug('Restarting cherrypy server...')
+                if self.server_adapter:
+                    self.server_adapter.stop()
+                if self.agent_adapter:
+                    self.agent_adapter.stop()
+                try:
+                    self.server_adapter, self.agent_adapter = start_servers()
+                    self.mgr.log.debug('Cherrypy server restarted successfully.')
+                except Exception as e:
+                    self.mgr.log.error(f'Failed to restart cherrypy server: {e}')
+                    continue
+
+        if self.server_adapter:
+            self.server_adapter.stop()
+        if self.agent_adapter:
+            self.agent_adapter.stop()
 
     def shutdown(self) -> None:
         self.mgr.log.debug('Stopping cherrypy engine...')
index 2c0478cb6684e69d0d60d1c44fd9949829acfa5f..6d1d86837e7991dd9c7056ff57c7a2a90a88dabe 100644 (file)
@@ -1,11 +1,4 @@
-try:
-    import cherrypy
-    from cherrypy._cpserver import Server
-except ImportError:
-    # to avoid sphinx build crash
-    class Server:  # type: ignore
-        pass
-
+import cherrypy
 import logging
 
 import orchestrator  # noqa
@@ -28,16 +21,6 @@ if TYPE_CHECKING:
     from cephadm.module import CephadmOrchestrator
 
 
-def cherrypy_filter(record: logging.LogRecord) -> bool:
-    blocked = [
-        'TLSV1_ALERT_DECRYPT_ERROR'
-    ]
-    msg = record.getMessage()
-    return not any([m for m in blocked if m in msg])
-
-
-logging.getLogger('cherrypy.error').addFilter(cherrypy_filter)
-cherrypy.log.access_log.propagate = False
 logger = logging.getLogger(__name__)
 
 
@@ -61,28 +44,34 @@ class ServiceDiscovery:
 
     def validate_password(self, realm: str, username: str, password: str) -> bool:
         return (password == self.password and username == self.username)
+    
+    def get_cherrypy_config(self, enable_auth: bool) -> Dict:
+        config = {
+            '/': {
+                'request.dispatch': 4,
+                'environment': 'production',
+                'tools.gzip.on': True,
+                'engine.autoreload.on': False,
+            }
+        }
+        if enable_auth:
+            config['/'].update({
+                'tools.auth_basic.on': True,
+                'tools.auth_basic.realm': 'localhost',
+                'tools.auth_basic.checkpassword': self.validate_password,
+            })
+        return config
 
-    def configure_routes(self, server: Server, enable_auth: bool) -> None:
+    def configure_routes(self, root: 'Root') -> cherrypy.dispatch.RoutesDispatcher:
         ROUTES = [
-            Route('index', '/', server.index),
-            Route('sd-config', '/prometheus/sd-config', server.get_sd_config),
-            Route('rules', '/prometheus/rules', server.get_prometheus_rules),
+            Route('index', '/', root.index),
+            Route('sd-config', '/prometheus/sd-config', root.get_sd_config),
+            Route('rules', '/prometheus/rules', root.get_prometheus_rules),
         ]
         d = cherrypy.dispatch.RoutesDispatcher()
         for route in ROUTES:
             d.connect(**route._asdict())
-        if enable_auth:
-            conf = {
-                '/': {
-                    'request.dispatch': d,
-                    'tools.auth_basic.on': True,
-                    'tools.auth_basic.realm': 'localhost',
-                    'tools.auth_basic.checkpassword': self.validate_password
-                }
-            }
-        else:
-            conf = {'/': {'request.dispatch': d}}
-        cherrypy.tree.mount(None, '/sd', config=conf)
+        return d
 
     def enable_auth(self) -> None:
         self.username = self.mgr.get_store('service_discovery/root/username')
@@ -93,7 +82,7 @@ class ServiceDiscovery:
             self.mgr.set_store('service_discovery/root/password', self.password)
             self.mgr.set_store('service_discovery/root/username', self.username)
 
-    def configure_tls(self, server: Server) -> None:
+    def configure_tls(self) -> Dict[str, str]:
         addr = self.mgr.get_mgr_ip()
         host = self.mgr.get_hostname()
         tls_pair = self.mgr.cert_mgr.generate_cert(host, addr, duration_in_days=CEPHADM_SVC_DISCOVERY_CERT_DURATION)
@@ -106,40 +95,35 @@ class ServiceDiscovery:
         self.key_file.flush()  # pkey_tmp must not be gc'ed
 
         verify_tls_files(self.cert_file.name, self.key_file.name)
-
-        server.ssl_certificate, server.ssl_private_key = self.cert_file.name, self.key_file.name
+        return {
+            'cert': self.cert_file.name,
+            'key': self.key_file.name,
+        }
 
     def configure(self, port: int, addr: str, enable_security: bool) -> None:
         # we create a new server to enforce TLS/SSL config refresh
-        self.root_server = Root(self.mgr, port, addr)
+        self.root_server = Root(self.mgr)
         self.root_server.ssl_certificate = None
         self.root_server.ssl_private_key = None
+        ssl_info = None
         if enable_security:
             self.enable_auth()
-            self.configure_tls(self.root_server)
-        self.configure_routes(self.root_server, enable_security)
+            ssl_info = self.configure_tls() 
+        config = self.get_cherrypy_config(enable_security)
+        dispatcher = self.configure_routes(self.root_server)
+        config['/'].update({'request.dispatch': dispatcher})
+        return config, ssl_info
 
 
-class Root(Server):
+class Root:
 
     # collapse everything to '/'
     def _cp_dispatch(self, vpath: str) -> 'Root':
         cherrypy.request.path = ''
         return self
 
-    def stop(self) -> None:
-        # we must call unsubscribe before stopping the server,
-        # otherwise the port is not released and we will get
-        # an exception when trying to restart it
-        self.unsubscribe()
-        super().stop()
-
-    def __init__(self, mgr: "CephadmOrchestrator", port: int = 0, host: str = ''):
+    def __init__(self, mgr: "CephadmOrchestrator"):
         self.mgr = mgr
-        super().__init__()
-        self.socket_port = port
-        self.socket_host = host
-        self.subscribe()
 
     @cherrypy.expose
     def index(self) -> str:
diff --git a/src/pybind/mgr/cherrypy_mgr.py b/src/pybind/mgr/cherrypy_mgr.py
new file mode 100644 (file)
index 0000000..b4a83f6
--- /dev/null
@@ -0,0 +1,122 @@
+"""
+CherryPyMgr is a utility class to encapsulate the CherryPy server instance
+into a standalone component. Unlike standard cherrypy which relies on global state
+and a single engine, CherryPyMgr allows for multiple independent server instances
+to be created and managed within the same process. So we can run multiple servers
+in each modules without worrying about their global state interfering with each other.
+
+Usage:
+    # Create a tree and mount your WSGI app on it
+    from cherrypy import _cptree
+    tree = _cptree.Tree()
+    tree.mount(my_wsgi_app, config=config)
+
+    # Mount your WSGI app on the manager
+    adapter, app = CherryPyMgr.mount(
+        tree,
+        addr,
+        ssl_info={'cert': 'path/to/cert.pem', 'key': 'path/to/key.pem', 'context': ssl_context}
+        )
+
+    # The adapter can be used to stop the server when needed
+    adapter.stop()
+"""
+import logging
+import cherrypy
+from cherrypy.process.servers import ServerAdapter
+from cheroot.wsgi import Server as WSGIServer
+from cheroot.ssl.builtin import BuiltinSSLAdapter
+from cherrypy._cptree import Tree
+from typing import Any, Tuple, Optional, Dict
+
+logger = logging.getLogger(__name__)
+
+
+def cherrypy_filter(record: logging.LogRecord) -> bool:
+    blocked = [
+        'TLSV1_ALERT_DECRYPT_ERROR'
+    ]
+    msg = record.getMessage()
+    return not any([m for m in blocked if m in msg])
+
+
+class CherryPyMgr:
+    @classmethod
+    def mount(
+        cls,
+        tree: Tree,
+        bind_addr: Tuple[str, int],
+        ssl_info: Optional[Dict[str, Any]] = None,
+        conf: Optional[Dict[str, Any]] = None
+    ) -> Tuple[ServerAdapter, Any]:
+        """
+        :param bind_addr: Tuple (host, port)
+        :param ssl_info: Dict containing {'cert': path, 'key': path, 'context': ssl_context}
+        :param conf: Optional CherryPy config dict for the mounted app
+        """
+        if not hasattr(cherrypy, '_mgr_engine_started'):
+            if hasattr(cherrypy, 'server'):
+                cherrypy.server.unsubscribe()
+            if hasattr(cherrypy.engine, 'autoreload'):
+                cherrypy.engine.autoreload.unsubscribe()
+            if hasattr(cherrypy.engine, 'signal_handler'):
+                cherrypy.engine.signal_handler.unsubscribe()
+
+            cherrypy.config.update({
+                'engine.autoreload.on': False,
+                'checker.on': False,
+                'tools.log_headers.on': False,
+                'log.screen': False
+            })
+            try:
+                cherrypy.engine.start()
+                cherrypy._mgr_engine_started = True
+                logger.info('Cherrypy engine started successfully.')
+            except Exception as e:
+                logger.error(f'Failed to start cherrypy engine: {e}')
+                raise e
+
+        cls.configure_logging()
+        adapter = cls.create_adapter(tree, bind_addr, ssl_info)
+        cls.subscribe_adapter(adapter)
+        adapter.start()
+
+        return adapter, tree
+
+    @staticmethod
+    def configure_logging() -> None:
+        cherrypy.log.access_log.propagate = False
+        cherrypy.log.error_log.propagate = False
+
+        error_log = logging.getLogger('cherrypy.error')
+
+        # make sure we only add the filter once
+        has_filter = any(f.__name__ == 'cherrypy_filter' for f in error_log.filters if hasattr(f, '__name__'))
+        if not has_filter:
+            error_log.addFilter(cherrypy_filter)
+
+    @staticmethod
+    def create_adapter(
+        app: Any,
+        bind_addr: Tuple[str, int],
+        ssl_info: Optional[Dict[str, Any]] = None,
+    ) -> ServerAdapter:
+        server = WSGIServer(
+            bind_addr=bind_addr,
+            wsgi_app=app,
+            numthreads=30,
+            server_name='Ceph-Mgr'
+        )
+
+        if ssl_info:
+            adapter = BuiltinSSLAdapter(ssl_info['cert'], ssl_info['key'])
+            if ssl_info.get('context'):
+                adapter.context = ssl_info['context']
+            server.ssl_adapter = adapter
+
+        adapter = ServerAdapter(cherrypy.engine, server, bind_addr)
+        return adapter
+
+    @staticmethod
+    def subscribe_adapter(adapter: ServerAdapter) -> None:
+        adapter.subscribe()
index 9ce6d12ea503d6c41e004dbef577ad61b516200d..a1eeb7e88103bf0a28248ec8df431d96a9344a29 100644 (file)
@@ -14,6 +14,7 @@ import threading
 import time
 from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple
 from urllib.parse import urlparse
+from cherrypy import _cptree
 
 from .controllers.multi_cluster import MultiCluster
 
@@ -28,6 +29,7 @@ from mgr_module import CLIReadCommand, CLIWriteCommand, HandleCommandResult, \
     MgrModule, MgrStandbyModule, NotifyType, Option, _get_localized_key
 from mgr_util import ServerConfigException, build_url, \
     create_self_signed_cert, get_default_addr, verify_tls_files
+from cherrypy_mgr import CherryPyMgr
 
 from . import mgr
 from .controllers import nvmeof  # noqa # pylint: disable=unused-import
@@ -90,10 +92,32 @@ class CherryPyConfig(object):
     def url_prefix(self):
         return self._url_prefix
 
-    @staticmethod
-    def update_cherrypy_config(config):
-        PLUGIN_MANAGER.hook.configure_cherrypy(config=config)
-        cherrypy.config.update(config)
+    def update_cherrypy_config(self, config):
+        if '/' not in config:
+            config['/'] = {}
+
+        defaults = { 
+            'response.headers.server': 'Ceph-Dashboard',
+            'response.headers.content-security-policy': "frame-ancestors 'self';",
+            'response.headers.x-content-type-options': 'nosniff',
+            'response.headers.strict-transport-security': 'max-age=63072000; includeSubDomains; preload',
+            'engine.autoreload.on': False,
+            'tools.request_logging.on': True,
+            'tools.gzip.on': True,
+            'tools.gzip.mime_types': [
+                'text/html', 'text/plain', 'application/json',
+                'application/*+json', 'application/javascript', 'text/css'
+            ],
+            'tools.json_in.on': True,
+            'tools.json_in.force': True,
+            'tools.plugin_hooks_filter_request.on': True,
+            'error_page.default': json_error_page,
+            'tools.sessions.on': True
+        }
+
+        config['/'].update(defaults)
+        PLUGIN_MANAGER.hook.configure_cherrypy(config=config['/'])
+        config['/']['request.show_tracebacks'] = True
 
     # pylint: disable=too-many-branches
     def _configure(self):
@@ -120,7 +144,8 @@ class CherryPyConfig(object):
 
         # Initialize custom handlers.
         cherrypy.tools.authenticate = AuthManagerTool()
-        configure_cors()
+        config = {'/': {}}
+        configure_cors(config)
         cherrypy.tools.plugin_hooks_filter_request = cherrypy.Tool(
             'before_handler',
             lambda: PLUGIN_MANAGER.hook.filter_request_before_handler(request=cherrypy.request),
@@ -129,31 +154,8 @@ class CherryPyConfig(object):
         cherrypy.tools.dashboard_exception_handler = HandlerWrapperTool(dashboard_exception_handler,
                                                                         priority=31)
 
-        cherrypy.log.access_log.propagate = False
-        cherrypy.log.error_log.propagate = False
-
-        # Apply the 'global' CherryPy configuration.
-        config = {
-            'engine.autoreload.on': False,
-            'server.socket_host': server_addr,
-            'server.socket_port': int(server_port),
-            'error_page.default': json_error_page,
-            'tools.request_logging.on': True,
-            'tools.gzip.on': True,
-            'tools.gzip.mime_types': [
-                # text/html and text/plain are the default types to compress
-                'text/html', 'text/plain',
-                # We also want JSON and JavaScript to be compressed
-                'application/json',
-                'application/*+json',
-                'application/javascript',
-                'text/css',
-            ],
-            'tools.json_in.on': True,
-            'tools.json_in.force': True,
-            'tools.plugin_hooks_filter_request.on': True,
-        }
 
+        ssl_info = None
         if use_ssl:
             # SSL initialization
             cert = self.get_localized_store("crt")  # type: ignore
@@ -184,10 +186,11 @@ class CherryPyConfig(object):
             else:
                 context.options |= ssl.OP_NO_TLSv1 | ssl.OP_NO_TLSv1_1 | ssl.OP_NO_TLSv1_2
 
-            config['server.ssl_module'] = 'builtin'
-            config['server.ssl_certificate'] = cert_fname
-            config['server.ssl_private_key'] = pkey_fname
-            config['server.ssl_context'] = context
+            ssl_info = {
+                'cert': cert_fname,
+                'key': pkey_fname,
+                'context': context
+            }
 
         self.update_cherrypy_config(config)
 
@@ -202,7 +205,7 @@ class CherryPyConfig(object):
             port=server_port,
         )
         uri = f'{base_url}{self.url_prefix}/'
-        return uri
+        return uri, (server_addr, server_port), ssl_info, config
 
     def await_configuration(self):
         """
@@ -213,7 +216,7 @@ class CherryPyConfig(object):
         """
         while not self._stopping.is_set():
             try:
-                uri = self._configure()
+                uri, bind_addr, ssl_info, config = self._configure()
             except ServerConfigException as e:
                 self.log.info(  # type: ignore
                     "Config not ready to serve, waiting: {0}".format(e)
@@ -222,7 +225,7 @@ class CherryPyConfig(object):
                 self._stopping.wait(5)
             else:
                 self.log.info("Configured CherryPy, starting engine...")  # type: ignore
-                return uri
+                return uri, bind_addr, ssl_info, config
 
 
 if TYPE_CHECKING:
@@ -291,6 +294,9 @@ class Module(MgrModule, CherryPyConfig):
     def __init__(self, *args, **kwargs):
         super(Module, self).__init__(*args, **kwargs)
         CherryPyConfig.__init__(self)
+        self.shutdown_event = threading.Event()
+        self.server_adapter = None
+
         # configure the dashboard's crypto caller. by default it will
         # use the remote caller to avoid pyo3 conflicts
         choose_crypto_caller(str(self.get_module_option('crypto_caller', '')))
@@ -341,10 +347,10 @@ class Module(MgrModule, CherryPyConfig):
         AuthManager.initialize()
         load_sso_db()
 
-        uri = self.await_configuration()
-        if uri is None:
-            # We were shut down while waiting
+        conf_result = self.await_configuration()
+        if conf_result is None:
             return
+        uri, bind_addr, ssl_info, app_config = conf_result
 
         # Publish the URI that others may use to access the service we're
         # about to start serving
@@ -353,16 +359,26 @@ class Module(MgrModule, CherryPyConfig):
         mapper, parent_urls = Router.generate_routes(self.url_prefix)
 
         config = {}
+        self.update_cherrypy_config(config)
         for purl in parent_urls:
-            config[purl] = {
-                'request.dispatch': mapper
-            }
-
-        cherrypy.tree.mount(None, config=config)
+            # Ensure the key exists
+            if purl not in config:
+                config[purl] = {}
+            config[purl]['request.dispatch'] = mapper
+        
+        logger.info('Starting ceph dashboard server at %s', uri)
+
+        tree = _cptree.Tree()
+        tree.mount(None, f"{self.url_prefix}/", config=config)
+        self.server_adapter, _ = CherryPyMgr.mount(
+            tree, 
+            bind_addr, 
+            ssl_info=ssl_info, 
+            conf=config
+        )
 
         PLUGIN_MANAGER.hook.setup()
 
-        cherrypy.engine.start()
         NotificationQueue.start_queue()
         TaskManager.init()
         logger.info('Engine started.')
@@ -379,15 +395,20 @@ class Module(MgrModule, CherryPyConfig):
         # wait for the shutdown event
         self.shutdown_event.wait()
         self.shutdown_event.clear()
+        if hasattr(self, 'server_adapter'):
+            self.server_adapter.stop()
+            self.server_adapter.unsubscribe()
         NotificationQueue.stop()
-        cherrypy.engine.stop()
         logger.info('Engine stopped')
 
     def shutdown(self):
         super(Module, self).shutdown()
+        self.shutdown_event.set()
         CherryPyConfig.shutdown(self)
         logger.info('Stopping engine...')
-        self.shutdown_event.set()
+        if hasattr(self, 'server_adapter'):
+            self.server_adapter.stop()
+            self.server_adapter.unsubscribe()
 
     def _set_ssl_item(self, item_label: str, item_key: 'SslConfigKey' = 'crt',
                       mgr_id: Optional[str] = None, inbuf: Optional[str] = None):
@@ -569,6 +590,7 @@ class StandbyModule(MgrStandbyModule, CherryPyConfig):
         super(StandbyModule, self).__init__(*args, **kwargs)
         CherryPyConfig.__init__(self)
         self.shutdown_event = threading.Event()
+        self.server_adapter = None
         # configure the dashboard's crypto caller. by default it will
         # use the remote caller to avoid pyo3 conflicts
         choose_crypto_caller(str(self.get_module_option('crypto_caller', '')))
@@ -578,10 +600,10 @@ class StandbyModule(MgrStandbyModule, CherryPyConfig):
         mgr.init(self)
 
     def serve(self):
-        uri = self.await_configuration()
-        if uri is None:
-            # We were shut down while waiting
+        conf_result = self.await_configuration()
+        if conf_result is None:
             return
+        uri, bind_addr, ssl_info, app_config = conf_result
 
         module = self
 
@@ -629,19 +651,31 @@ class StandbyModule(MgrStandbyModule, CherryPyConfig):
                     status = module.get_module_option('standby_error_status_code', 500)
                     raise cherrypy.HTTPError(status, message="Keep on looking")
 
-        cherrypy.tree.mount(Root(), "{}/".format(self.url_prefix), {})
+        standby_conf = {}
+        self.update_cherrypy_config(standby_conf)
+
+        standby_tree = _cptree.Tree()
+        standby_tree.mount(Root(), f"{self.url_prefix}/", config=standby_conf)
         self.log.info("Starting engine...")
-        cherrypy.engine.start()
+        self.server_adapter, _ = CherryPyMgr.mount(
+            standby_tree,
+            bind_addr,
+            ssl_info=ssl_info
+        )
         self.log.info("Engine started...")
         # Wait for shutdown event
         self.shutdown_event.wait()
         self.shutdown_event.clear()
-        cherrypy.engine.stop()
+        if hasattr(self, 'server_adapter'):
+            self.server_adapter.stop()
+            self.server_adapter.unsubscribe()
         self.log.info("Engine stopped.")
 
     def shutdown(self):
-        CherryPyConfig.shutdown(self)
-
         self.log.info("Stopping engine...")
         self.shutdown_event.set()
+        CherryPyConfig.shutdown(self)
+        if hasattr(self, 'server_adapter'):
+            self.server_adapter.stop()
+            self.server_adapter.unsubscribe()
         self.log.info("Stopped engine...")
index 7f1cdb5887c390408349b917a4d8e7e2677c2d14..b3f966ecffe4097a8af47008c39db50ff076d878 100644 (file)
@@ -22,13 +22,6 @@ from ..access_control import LocalAuthenticator, UserDoesNotExist
 if TYPE_CHECKING:
     from dashboard.services.sso import SsoDB
 
-cherrypy.config.update({
-    'response.headers.server': 'Ceph-Dashboard',
-    'response.headers.content-security-policy': "frame-ancestors 'self';",
-    'response.headers.x-content-type-options': 'nosniff',
-    'response.headers.strict-transport-security': 'max-age=63072000; includeSubDomains; preload'
-})
-
 
 class AuthType(str, Enum):
     LOCAL = 'local'
index fbf1e26bf41e968bdbfd7d683c6fadf8e7199d43..53d3d840b83445fe8ee0dc07170254de684adf1e 100644 (file)
@@ -840,7 +840,7 @@ def merge_list_of_dicts_by_key(target_list: list, source_list: list, key: str):
     return target_list
 
 
-def configure_cors(url: str = ''):
+def configure_cors(config, url: str = ''):
     """
     Allow CORS requests if the cross_origin_url option is set.
     """
@@ -850,11 +850,9 @@ def configure_cors(url: str = ''):
     else:
         cross_origin_url = mgr.get_localized_module_option('cross_origin_url', '')
     if cross_origin_url:
-        cherrypy.tools.CORS = cherrypy.Tool('before_handler', cors_tool)
-        config = {
-            'tools.CORS.on': True,
-        }
-        cherrypy.config.update(config)
+        if not hasattr(cherrypy.tools, 'CORS'):
+            cherrypy.tools.CORS = cherrypy.Tool('before_handler', cors_tool)
+        config['/']['tools.CORS.on'] = True
 
 
 def cors_tool():
index 836f64aa2b76238d55ef435105edbe9faaeee3c0..2cd6d237b380ca7ed2ba42a929e4221aa912c5d0 100644 (file)
@@ -10,6 +10,8 @@ import enum
 from collections import namedtuple
 from collections import OrderedDict
 from tempfile import NamedTemporaryFile
+from cherrypy_mgr import CherryPyMgr
+from cherrypy import _cptree
 
 from mgr_module import CLIReadCommand, MgrModule, MgrStandbyModule, PG_STATES, Option, ServiceInfoT, HandleCommandResult, CLIWriteCommand
 from mgr_util import get_default_addr, profile_method, build_url, test_port_allocation, PortAlreadyInUse
@@ -64,11 +66,6 @@ def _wait_for_port_available(
     return False
 
 
-cherrypy.config.update({
-    'response.headers.server': 'Ceph-Prometheus'
-})
-
-
 def health_status_to_number(status: str) -> int:
     if status == 'HEALTH_OK':
         return 0
@@ -1959,7 +1956,7 @@ class Module(MgrModule, OrchestratorClientMixin):
         self.collect()
         self.get_file_sd_config()
 
-    def configure(self, server_addr: str, server_port: int) -> None:
+    def configure(self) -> None:
         cmd = {'prefix': 'orch get-security-config'}
         ret, out, _ = self.mon_command(cmd)
 
@@ -1967,7 +1964,7 @@ class Module(MgrModule, OrchestratorClientMixin):
             try:
                 security_config = json.loads(out)
                 if security_config.get('security_enabled', False):
-                    self.setup_tls_config(server_addr, server_port)
+                    self.setup_tls_config()
                     return
             except Exception as e:
                 self.log.exception(
@@ -1977,29 +1974,27 @@ class Module(MgrModule, OrchestratorClientMixin):
                 )
 
         # In any error fallback to plain http mode
-        self.setup_default_config(server_addr, server_port)
-
-    def setup_default_config(self, server_addr: str, server_port: int) -> None:
-        cherrypy.config.update({
-            'server.socket_host': server_addr,
-            'server.socket_port': server_port,
-            'engine.autoreload.on': False,
-            'server.ssl_module': None,
-            'server.ssl_certificate': None,
-            'server.ssl_private_key': None,
-            'tools.gzip.on': True,
-            'tools.gzip.mime_types': [
-                'text/plain',
-                'text/html',
-                'application/json',
-            ],
-            'tools.gzip.compress_level': 6,
-        })
-        # Publish the URI that others may use to access the service we're about to start serving
-        self.set_uri(build_url(scheme='http', host=self.get_server_addr(),
-                     port=server_port, path='/'))
+        return self.setup_default_config()
+    
+    def get_cherrypy_config(self):
+        config = {
+            '/': {
+                'response.headers.server': 'Ceph-Prometheus',
+                'tools.gzip.on': True,
+                'tools.gzip.mime_types': [
+                    'text/plain',
+                    'text/html',
+                    'application/json',
+                ],
+                'tools.gzip.compress_level': 6,
+            }
+        }
+        return config
+
+    def setup_default_config(self) -> None:
+        return self.get_cherrypy_config(), None, 'http'
 
-    def setup_tls_config(self, server_addr: str, server_port: int) -> None:
+    def setup_tls_config(self) -> None:
         # Temporarily disabling the verify function due to issues.
         # Please check verify_tls_files below to more information.
         # from mgr_util import verify_tls_files
@@ -2027,25 +2022,12 @@ class Module(MgrModule, OrchestratorClientMixin):
         # Re-enable once the issue is resolved.
         # verify_tls_files(self.cert_file.name, self.key_file.name)
         cert_file_path, key_file_path = self.cert_file.name, self.key_file.name
+        ssl_info = {
+            'cert': cert_file_path,
+            'key': key_file_path
+        }
 
-        cherrypy.config.update({
-            'server.socket_host': server_addr,
-            'server.socket_port': server_port,
-            'engine.autoreload.on': False,
-            'server.ssl_module': 'builtin',
-            'server.ssl_certificate': cert_file_path,
-            'server.ssl_private_key': key_file_path,
-            'tools.gzip.on': True,
-            'tools.gzip.mime_types': [
-                'text/plain',
-                'text/html',
-                'application/json',
-            ],
-            'tools.gzip.compress_level': 6,
-        })
-        # Publish the URI that others may use to access the service we're about to start serving
-        self.set_uri(build_url(scheme='https', host=self.get_server_addr(),
-                     port=server_port, path='/'))
+        return self.get_cherrypy_config(), ssl_info, 'https'
 
     def serve(self) -> None:
 
@@ -2124,36 +2106,39 @@ class Module(MgrModule, OrchestratorClientMixin):
         if self.stale_cache_strategy not in [self.STALE_CACHE_FAIL,
                                              self.STALE_CACHE_RETURN]:
             self.stale_cache_strategy = self.STALE_CACHE_FAIL
-
-        server_addr = cast(str, self.get_localized_module_option('server_addr', get_default_addr()))
-        server_port = cast(int, self.get_localized_module_option('server_port', DEFAULT_PORT))
-        self.log.info(
-            "server_addr: %s server_port: %s" %
-            (server_addr, server_port)
-        )
-
+        
         self.cache = cast(bool, self.get_localized_module_option('cache', True))
         if self.cache:
             self.log.info('Cache enabled')
             self.metrics_thread.start()
         else:
             self.log.info('Cache disabled')
+        def start_server():
+            server_addr = cast(str, self.get_localized_module_option('server_addr', get_default_addr()))
+            server_port = cast(int, self.get_localized_module_option('server_port', DEFAULT_PORT))
+
+            config, ssl_info, scheme = self.configure()
+            tree = _cptree.Tree()
+            tree.mount(Root(), "/", config=config)
+
+            # Wait for port to be available before starting (handles standby->active transition)
+            if not _wait_for_port_available(self.log, server_addr, server_port):
+                self.log.warning(f'Port {server_port} still in use after waiting, attempting to start anyway')
+
+            self.log.info(f'Starting prometheus server on {server_addr}:{server_port}')
+            adapter, _ = CherryPyMgr.mount(
+                tree,
+                (server_addr, int(server_port)),
+                ssl_info=ssl_info
+            )
+            self.set_uri(build_url(scheme=scheme, host=self.get_server_addr(), port=server_port, path='/'))
+            return adapter
 
-        self.configure(server_addr, server_port)
-
-        cherrypy.tree.mount(Root(), "/")
-
-        # Wait for port to be available before starting (handles standby->active transition)
-        if not _wait_for_port_available(self.log, server_addr, server_port):
-            self.log.warning(f'Port {server_port} still in use after waiting, attempting to start anyway')
-        self.log.info('Starting engine...')
         try:
-            cherrypy.engine.start()
+            self.server_adapter = start_server()
         except Exception as e:
-            self.log.error(f'Failed to start engine: {e}')
+            self.log.error(f'Failed to start Prometheus: {e}')
             return
-        self.log.info('Engine started.')
-
         # Main event loop: handle both shutdown and config change events
         while True:
             # Wait for either shutdown or config change event (check every 0.5s)
@@ -2169,33 +2154,23 @@ class Module(MgrModule, OrchestratorClientMixin):
                 self.config_change_event.clear()
                 self.log.info('Restarting engine due to config change...')
 
-                # https://stackoverflow.com/questions/7254845/change-cherrypy-port-and-restart-web-server
-                # if we omit the line: cherrypy.server.httpserver = None
-                # then the cherrypy server is not restarted correctly
-                cherrypy.engine.stop()
-                cherrypy.server.httpserver = None
-
-                # Re-read configuration
-                server_addr = cast(str, self.get_localized_module_option('server_addr', get_default_addr()))
-                server_port = cast(int, self.get_localized_module_option('server_port', DEFAULT_PORT))
-                self.configure(server_addr, server_port)
-
-                # Wait for port to be available before starting
-                if not _wait_for_port_available(self.log, server_addr, server_port):
-                    self.log.warning(f'Port {server_port} still in use after waiting, attempting to start anyway')
+                if hasattr(self, 'server_adapter'):
+                    self.server_adapter.stop()
+                    self.server_adapter.unsubscribe()
 
                 try:
-                    cherrypy.engine.start()
-                    self.log.info('Engine restarted.')
+                    self.server_adapter = start_server()
+                    self.log.debug('Prometheus restarted successfully.')
                 except Exception as e:
-                    self.log.error(f'Failed to restart engine: {e}')
+                    self.log.error(f'Failed to restart Prometheus: {e}')
 
         # Cleanup on shutdown
         self.shutdown_event.clear()
         # tell metrics collection thread to stop collecting new metrics
         self.metrics_thread.stop()
-        cherrypy.engine.stop()
-        cherrypy.server.httpserver = None
+        if hasattr(self, 'server_adapter'):
+            self.server_adapter.stop()
+            self.server_adapter.unsubscribe()
         self.log.info('Engine stopped.')
         self.shutdown_rbd_stats()
         # wait for the metrics collection thread to stop
@@ -2252,12 +2227,6 @@ class StandbyModule(MgrStandbyModule):
             'server_port', DEFAULT_PORT))
         self.log.info("server_addr: %s server_port: %s" %
                       (server_addr, server_port))
-        cherrypy.config.update({
-            'server.socket_host': server_addr,
-            'server.socket_port': server_port,
-            'engine.autoreload.on': False,
-            'request.show_tracebacks': False
-        })
 
         module = self
 
@@ -2283,23 +2252,38 @@ class StandbyModule(MgrStandbyModule):
             def metrics(self) -> str:
                 return ''
 
-        cherrypy.tree.mount(Root(), '/', {})
+        config = {
+            '/': {
+                'response.headers.server': 'Ceph-Prometheus',
+                'engine.autoreload.on': False,
+            }
+        }
+        tree = _cptree.Tree()
+        tree.mount(Root(), '/', config=config)
 
         # Wait for port to be available before starting
         if not _wait_for_port_available(self.log, server_addr, server_port):
             self.log.warning(f'Port {server_port} still in use after waiting, attempting to start anyway')
         self.log.info('Starting engine...')
-        cherrypy.engine.start()
+        self.server_adapter, _ = CherryPyMgr.mount(
+            tree,
+            (server_addr, int(server_port)),
+            conf=config,
+        )
         self.log.info('Engine started.')
 
         # Wait for shutdown event
         self.shutdown_event.wait()
         self.shutdown_event.clear()
-        cherrypy.engine.stop()
-        cherrypy.server.httpserver = None
+        if hasattr(self, 'server_adapter'):
+            self.server_adapter.stop()
+            self.server_adapter.unsubscribe()
         self.log.info('Engine stopped.')
 
     def shutdown(self) -> None:
         self.log.info("Stopping engine...")
-        self.shutdown_event.set()
-        self.log.info("Stopped engine")
+        if hasattr(self, 'shutdown_event'):
+            self.shutdown_event.set()
+        if hasattr(self, 'server_adapter'):
+            self.server_adapter.stop()
+            self.server_adapter.unsubscribe()
index c2deb627261ecb3f50da391ff5b68e90205ff88b..5cb6fffe11e791b7788b457f94dd0dfa1f2ab127 100644 (file)
@@ -150,6 +150,7 @@ modules =
     localpool \
     mgr_module.py \
     mgr_util.py \
+    cherrypy_mgr.py \
     nfs \
     object_format.py \
     orchestrator \
@@ -198,7 +199,7 @@ modules = smb
 [isort]
 profile = black
 line_length = 78
-known_first_party = ceph,rados,rbd,cephfs,mgr,mgr_module,mgr_util,object_format
+known_first_party = ceph,rados,rbd,cephfs,mgr,mgr_module,mgr_util,object_format,cherrypy_mgr
 known_typing = typing
 sections = FUTURE,TYPING,STDLIB,THIRDPARTY,FIRSTPARTY,LOCALFOLDER