From: Leonid Chernin <leonidc@il.ibm.com>
Date: Thu, 5 Jun 2025 07:59:03 +0000 (+0300)
Subject: nvmeofgw: fix sending acks during upgrade
X-Git-Tag: v20.1.0~126^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=58d66e48f24b8ef4c8caae5d0360069cd7a85fe7;p=ceph.git

nvmeofgw: fix sending acks during upgrade
          fix gw-epoch during upgrade
          1.fixed issue  when during upgrade before feature NVMEOFHAMAP is
          applyed, monitor did not send acks to beacon - this cause
          expiration GW mon-client beacon ack timeout and assert of the GW.
          2. fixed issue that during upgrade new GW epoch, generated
	  after applying of new feature NVMEOFHAMAP was less than
	  global epoch. So GW mon-client asserted on receiving new epoch
          that is less than its last received epoch

Signed-off-by: Leonid Chernin <leonidc@il.ibm.com>
(cherry picked from commit 3042c3695cbdbc75bad1d3564e887f1b1ad77ef6)
---

diff --git a/src/mon/NVMeofGwMap.cc b/src/mon/NVMeofGwMap.cc
index 12bd93cef746..d5479b1ffce2 100755
--- a/src/mon/NVMeofGwMap.cc
+++ b/src/mon/NVMeofGwMap.cc
@@ -267,6 +267,11 @@ void  NVMeofGwMap::gw_performed_startup(const NvmeGwId &gw_id,
 void NVMeofGwMap::increment_gw_epoch(const NvmeGroupKey& group_key)
 {
   if (HAVE_FEATURE(mon->get_quorum_con_features(), NVMEOFHAMAP)) {
+    if (gw_epoch.find(group_key) == gw_epoch.end()) {
+      gw_epoch[group_key] = epoch;
+      dout(4) << "recreated GW epoch of " << group_key
+           << " " << gw_epoch[group_key] << dendl;
+    }
     gw_epoch[group_key] ++;
     dout(4) << "incremented epoch of " << group_key
          << " " << gw_epoch[group_key] << dendl;
diff --git a/src/mon/NVMeofGwMon.cc b/src/mon/NVMeofGwMon.cc
index e41060084eae..96611d08750b 100644
--- a/src/mon/NVMeofGwMon.cc
+++ b/src/mon/NVMeofGwMon.cc
@@ -158,7 +158,8 @@ version_t NVMeofGwMon::get_trim_to() const
  * function called during new paxos epochs
  * function called to restore in pending map all data that is not serialized
  * to paxos peons. Othervise it would be overriden in "pending_map = map"
- * currently "allow_failovers_ts" and "last_gw_down_ts" variables restored
+ * currently "allow_failovers_ts", "last_gw_down_ts",
+ * "last_gw_map_epoch_valid" variables are restored
  */
 void NVMeofGwMon::restore_pending_map_info(NVMeofGwMap & tmp_map) {
   std::chrono::system_clock::time_point now = std::chrono::system_clock::now();
@@ -175,6 +176,8 @@ void NVMeofGwMon::restore_pending_map_info(NVMeofGwMap & tmp_map) {
       }
       pending_map.created_gws[group_key][gw_id].last_gw_down_ts =
           gw_created_pair.second.last_gw_down_ts;
+      pending_map.created_gws[group_key][gw_id].last_gw_map_epoch_valid =
+	  gw_created_pair.second.last_gw_map_epoch_valid;
     }
   }
 }
@@ -806,12 +809,13 @@ bool NVMeofGwMon::prepare_beacon(MonOpRequestRef op)
     send_ack = true;
     if (apply_ack_logic) {
       dout(20) << "ack sent: beacon index "
-       << pending_map.created_gws[group_key][gw_id].beacon_index
-       << " gw " << gw_id <<dendl;
+      << pending_map.created_gws[group_key][gw_id].beacon_index
+      << " gw " << gw_id << " epoch-filter  " << epoch_filter_enabled
+      << " propose " << propose << " gw-propose " << gw_propose << dendl;
     }
   }
   if (send_ack && ((!gw_propose && epoch_filter_enabled) ||
-                    (propose && !epoch_filter_enabled) ||
+                    (!propose && !epoch_filter_enabled) ||
                     (avail == gw_availability_t::GW_CREATED)) ) {
           /* always send beacon ack to gw in Created state,
            * it should be temporary state