From 98571db537cea1f0a8cccfd92c21568d9206185a Mon Sep 17 00:00:00 2001
From: Alexander Indenbaum <aindenba@redhat.com>
Date: Mon, 7 Jul 2025 10:42:37 +0300
Subject: [PATCH] src/nvmeof/NVMeofGwMonitorClient: connect panic

- add NVMe-oF Monitor Client configuration reference

Tracker: https://tracker.ceph.com/issues/72509

Signed-off-by: Alexander Indenbaum <aindenba@redhat.com>
(cherry picked from commit babd3c00341c59cb82e66f93f41c687ab4fc6928)
---
 doc/rados/configuration/mon-config-ref.rst |  7 +++++++
 src/common/options/mon.yaml.in             |  9 ++++++++-
 src/nvmeof/NVMeofGwMonitorClient.cc        | 20 +++++++++++++++++++-
 src/nvmeof/NVMeofGwMonitorClient.h         |  6 ++++++
 4 files changed, 40 insertions(+), 2 deletions(-)

diff --git a/doc/rados/configuration/mon-config-ref.rst b/doc/rados/configuration/mon-config-ref.rst
index 880fb608090..9f4cf6ea1f2 100644
--- a/doc/rados/configuration/mon-config-ref.rst
+++ b/doc/rados/configuration/mon-config-ref.rst
@@ -629,6 +629,13 @@ Miscellaneous
 .. confval:: mon_memory_autotune
 .. confval:: enable_availability_tracking
 
+NVMe-oF Monitor Client
+=====================
+
+.. confval:: nvmeof_mon_client_disconnect_panic
+.. confval:: nvmeof_mon_client_connect_panic
+.. confval:: nvmeof_mon_client_tick_period
+
 .. _Paxos: https://en.wikipedia.org/wiki/Paxos_(computer_science)
 .. _Monitor Keyrings: ../../../dev/mon-bootstrap#secret-keys
 .. _Ceph configuration file: ../ceph-conf/#monitors
diff --git a/src/common/options/mon.yaml.in b/src/common/options/mon.yaml.in
index 2c337fcea70..4ac1b2f05a4 100644
--- a/src/common/options/mon.yaml.in
+++ b/src/common/options/mon.yaml.in
@@ -1412,4 +1412,11 @@ options:
   - mon
   flags:
   - runtime
- 
\ No newline at end of file
+- name: nvmeof_mon_client_connect_panic
+  type: secs
+  level: advanced
+  desc: The duration, expressed in seconds, after which the nvmeof gateway
+    should trigger a panic if it does not receive the initial map from the monitor
+  default: 30
+  services:
+  - mon
diff --git a/src/nvmeof/NVMeofGwMonitorClient.cc b/src/nvmeof/NVMeofGwMonitorClient.cc
index 0b798c370a2..2000f718a7d 100644
--- a/src/nvmeof/NVMeofGwMonitorClient.cc
+++ b/src/nvmeof/NVMeofGwMonitorClient.cc
@@ -40,6 +40,7 @@ NVMeofGwMonitorClient::NVMeofGwMonitorClient(int argc, const char **argv) :
   gwmap_epoch(0),
   last_map_time(std::chrono::steady_clock::now()),
   reset_timestamp(std::chrono::steady_clock::now()),
+  start_time(last_map_time),
   monc{g_ceph_context, poolctx},
   client_messenger(Messenger::create(g_ceph_context, "async", entity_name_t::CLIENT(-1), "client", getpid())),
   objecter{g_ceph_context, client_messenger.get(), &monc, poolctx},
@@ -263,10 +264,27 @@ void NVMeofGwMonitorClient::disconnect_panic()
   }
 }
 
+void NVMeofGwMonitorClient::connect_panic()
+{
+  // Return immediately if the gateway was assigned group ID by the monitor
+  if (set_group_id) {
+    return;
+  }
+  // If the gateway has not been assigned a group ID, panic after timeout
+  auto connect_panic_duration = g_conf().get_val<std::chrono::seconds>("nvmeof_mon_client_connect_panic").count();
+  auto now = std::chrono::steady_clock::now();
+  auto elapsed_seconds = std::chrono::duration_cast<std::chrono::seconds>(now - start_time).count();
+  if (elapsed_seconds > connect_panic_duration) {
+    dout(4) << "Triggering a panic: did not receive initial map from monitor, elapsed " << elapsed_seconds << ", configured connect panic duration " << connect_panic_duration << " seconds." << dendl;
+    throw std::runtime_error("Did not receive initial map from monitor (connect panic).");
+  }
+}
+
 void NVMeofGwMonitorClient::tick()
 {
   dout(10) << dendl;
 
+  connect_panic();
   disconnect_panic();
   send_beacon();
   first_beacon = false;
@@ -339,7 +357,7 @@ void NVMeofGwMonitorClient::handle_nvmeof_gw_map(ceph::ref_t<MNVMeofGwMap> nmap)
       dout(10) << "Can not find new gw state" << dendl;
       return;
     }
-    bool set_group_id = false;
+    ceph_assert(!set_group_id);
     while (!set_group_id) {
       NVMeofGwMonitorGroupClient monitor_group_client(
           grpc::CreateChannel(monitor_address, gw_creds()));
diff --git a/src/nvmeof/NVMeofGwMonitorClient.h b/src/nvmeof/NVMeofGwMonitorClient.h
index 546fff27db7..f187a47b481 100644
--- a/src/nvmeof/NVMeofGwMonitorClient.h
+++ b/src/nvmeof/NVMeofGwMonitorClient.h
@@ -47,8 +47,12 @@ private:
               last_map_time; // used to panic on disconnect
   std::chrono::time_point<std::chrono::steady_clock>
                 reset_timestamp; // used to bypass some validations
+  std::chrono::time_point<std::chrono::steady_clock>
+                start_time; // used to panic on connect
 
   bool first_beacon = true;
+  bool set_group_id = false;
+
   // init gw ssl opts
   void init_gw_ssl_opts();
 
@@ -96,6 +100,8 @@ public:
   void disconnect_panic();
 
   void handle_nvmeof_gw_map(ceph::ref_t<MNVMeofGwMap> m);
+
+  void connect_panic();
 };
 
 #endif
-- 
2.39.5