mon/MgrMonitor: allow disabling always-on MGR modules

Add a new command ("ceph mgr module force disable <module>") that allows
forcibly disabling an always-on module. This command should ideally only
be used to for cluster recovery.

Fixes: https://tracker.ceph.com/issues/66005
Signed-off-by: Rishabh Dave <[email protected]>

Author: rishabh-d-dave

src/mgr/PyModuleRegistry.cc

@@ -151,7 +151,8 @@ bool PyModuleRegistry::handle_mgr_map(const MgrMap &mgr_map_)
     return false;
   } else {
     bool modules_changed = mgr_map_.modules != mgr_map.modules ||
-      mgr_map_.always_on_modules != mgr_map.always_on_modules;
+      mgr_map_.always_on_modules != mgr_map.always_on_modules ||
+      mgr_map_.force_disabled_modules != mgr_map.force_disabled_modules;
     mgr_map = mgr_map_;
 
     if (standby_modules != nullptr) {
@@ -240,6 +241,16 @@ void PyModuleRegistry::active_start(
     // Anything we're skipping because of !can_run will be flagged
     // to the user separately via get_health_checks
     if (!(i.second->is_enabled() && i.second->is_loaded())) {
+      dout(8) << __func__ << " Not starting module '" << i.first << "', it is "
+	      << "not enabled and loaded"  << dendl;
+      continue;
+    }
+
+    // These are always-on modules but user force-disabled them.
+    if (mgr_map.force_disabled_modules.find(i.first) !=
+	mgr_map.force_disabled_modules.end()) {
+      dout(8) << __func__ << " Not starting module '" << i.first << "', it is "
+	      << "force-disabled" << dendl;
       continue;
     }
 

src/mon/MgrMap.h

@@ -297,6 +297,9 @@ class MgrMap
   // active version.
   std::map<uint32_t, std::set<std::string>> always_on_modules;
 
+  // Modules which are always-on but have been force-disabled by user.
+  std::set<std::string> force_disabled_modules;
+
   // Modules which are reported to exist
   std::vector<ModuleInfo> available_modules;
 
@@ -448,7 +451,7 @@ class MgrMap
       ENCODE_FINISH(bl);
       return;
     }
-    ENCODE_START(13, 6, bl);
+    ENCODE_START(14, 6, bl);
     encode(epoch, bl);
     encode(active_addrs, bl, features);
     encode(active_gid, bl);
@@ -473,13 +476,14 @@ class MgrMap
     encode(clients_addrs, bl, features);
     encode(clients_names, bl, features);
     encode(flags, bl);
+    encode(force_disabled_modules, bl);
     ENCODE_FINISH(bl);
     return;
   }
 
   void decode(ceph::buffer::list::const_iterator& p)
   {
-    DECODE_START(13, p);
+    DECODE_START(14, p);
     decode(epoch, p);
     decode(active_addrs, p);
     decode(active_gid, p);
@@ -549,6 +553,11 @@ class MgrMap
     if (struct_v >= 13) {
       decode(flags, p);
     }
+
+    if (struct_v >= 14) {
+      decode(force_disabled_modules, p);
+    }
+
     DECODE_FINISH(p);
   }
 
@@ -603,6 +612,13 @@ class MgrMap
       f->close_section();
     }
     f->close_section(); // always_on_modules
+
+    f->open_object_section("force_disabled_modules");
+    for (auto& m : force_disabled_modules) {
+        f->dump_string("module", m);
+      }
+    f->close_section();
+
     f->dump_int("last_failure_osd_epoch", last_failure_osd_epoch);
     f->open_array_section("active_clients");
     for (const auto& i : clients) {

src/mon/MgrMonitor.cc

@@ -1021,6 +1021,13 @@ bool MgrMonitor::preprocess_command(MonOpRequestRef op)
           f->dump_string("module", p);
         }
         f->close_section();
+
+        f->open_array_section("force_disabled_modules");
+        for (auto& p : map.force_disabled_modules) {
+          f->dump_string("module", p);
+        }
+        f->close_section();
+
         f->open_array_section("enabled_modules");
         for (auto& p : map.modules) {
           if (map.get_always_on_modules().count(p) > 0)
@@ -1050,7 +1057,11 @@ bool MgrMonitor::preprocess_command(MonOpRequestRef op)
 
       for (auto& p : map.get_always_on_modules()) {
         tbl << p;
-        tbl << "on (always on)";
+	if (map.force_disabled_modules.find(p) == map.force_disabled_modules.end()) {
+	  tbl << "on (always on)";
+	} else  {
+	  tbl << "off (always on but force-disabled)";
+	}
         tbl << TextTable::endrow;
       }
       for (auto& p : map.modules) {
@@ -1271,10 +1282,13 @@ bool MgrMonitor::prepare_command(MonOpRequestRef op)
       r = -EINVAL;
       goto out;
     }
-    if (pending_map.get_always_on_modules().count(module) > 0) {
+
+    if (pending_map.get_always_on_modules().count(module) > 0 &&
+        !pending_map.force_disabled_modules.contains(module)) {
       ss << "module '" << module << "' is already enabled (always-on)";
       goto out;
     }
+
     bool force = false;
     cmd_getval_compat_cephbool(cmdmap, "force", force);
     if (!pending_map.all_support_module(module) &&
@@ -1298,7 +1312,12 @@ bool MgrMonitor::prepare_command(MonOpRequestRef op)
       ss << "module '" << module << "' is already enabled";
       r = 0;
       goto out;
+    } else if (pending_map.force_disabled_modules.contains(module)) {
+      pending_map.force_disabled_modules.erase(module);
+      r = 0;
+      goto out;
     }
+
     pending_map.modules.insert(module);
   } else if (prefix == "mgr module disable") {
     string module;
@@ -1308,8 +1327,9 @@ bool MgrMonitor::prepare_command(MonOpRequestRef op)
       goto out;
     }
     if (pending_map.get_always_on_modules().count(module) > 0) {
-      ss << "module '" << module << "' cannot be disabled (always-on)";
-      r = -EINVAL;
+      ss << "module '" << module << "' cannot be disabled (always-on), use " <<
+	 "'ceph mgr module force disable' command to disable an always-on module";
+      r = -EPERM;
       goto out;
     }
     if (!pending_map.module_enabled(module)) {
@@ -1322,6 +1342,50 @@ bool MgrMonitor::prepare_command(MonOpRequestRef op)
     }
     dout(8) << __func__ << " disabling module " << module << " from new " << dendl;
     pending_map.modules.erase(module);
+  } else if (prefix == "mgr module force disable") {
+    string mod;
+    cmd_getval(cmdmap, "module", mod);
+
+    bool confirmation_flag = false;
+    cmd_getval(cmdmap, "yes_i_really_mean_it", confirmation_flag);
+
+    if (mod.empty()) {
+      ss << "Module name wasn't passed!";
+      r = -EINVAL;
+      goto out;
+    }
+
+    if (!pending_map.get_always_on_modules().contains(mod)) {
+      ss << "Always-on module named \"" << mod << "\" does not exist";
+      r = -EINVAL;
+      goto out;
+    } else if (pending_map.modules.contains(mod)) {
+      ss << "Module '" << mod << "' is not an always-on module, only always-on " <<
+	 "modules can be disabled through this command.";
+      r = -EINVAL;
+      goto out;
+    }
+
+    if (pending_map.force_disabled_modules.contains(mod)) {
+      ss << "Module \"" << mod << "\"is already disabled";
+      r = 0;
+      goto out;
+    }
+
+    if (!confirmation_flag) {
+      ss << "This command will disable operations and remove commands that "
+	 << "other Ceph utilities expect to be available. Do not continue "
+	 << "unless your cluster is already experiencing an event due to "
+	 << "which it is advised to disable this module as part of "
+	 << "troubleshooting. If you are sure that you wish to continue, "
+	 << "run again with --yes-i-really-mean-it";
+      r = -EPERM;
+      goto out;
+    }
+
+    dout(8) << __func__ << " force-disabling module '" << mod << "'" << dendl;
+    pending_map.force_disabled_modules.insert(mod);
+    pending_map.modules.erase(mod);
   } else {
     ss << "Command '" << prefix << "' not implemented!";
     r = -ENOSYS;

src/mon/MonCommands.h

@@ -1357,6 +1357,10 @@ COMMAND("mgr module enable "
 COMMAND("mgr module disable "
 	"name=module,type=CephString",
 	"disable mgr module", "mgr", "rw")
+COMMAND("mgr module force disable "
+	"name=module,type=CephString "
+	"name=yes_i_really_mean_it,type=CephBool,req=false",
+	"force disable a always-on mgr module", "mgr", "rw")
 COMMAND("mgr metadata name=who,type=CephString,req=false",
 	"dump metadata for all daemons or a specific daemon",
 	"mgr", "r")