From f01b65c5e5426c48c53f27a13f8c8a8cb86dd1b2 Mon Sep 17 00:00:00 2001
From: Alexey Zatelepin <ztlpn@redpanda.com>
Date: Thu, 21 Sep 2023 23:04:44 +0200
Subject: [PATCH] admin: add raft follower_recovery_state to debug partition
 json

---
 src/v/cluster/cluster_utils.cc          | 10 +++++++++
 src/v/cluster/types.h                   | 27 +++++++++++++++++++++++--
 src/v/raft/consensus.h                  |  5 +++++
 src/v/redpanda/admin/api-doc/debug.json | 18 +++++++++++++++++
 src/v/redpanda/admin_server.cc          |  6 ++++++
 5 files changed, 64 insertions(+), 2 deletions(-)

diff --git a/src/v/cluster/cluster_utils.cc b/src/v/cluster/cluster_utils.cc
index 158eb07ccf648..67f536fbdc3c4 100644
--- a/src/v/cluster/cluster_utils.cc
+++ b/src/v/cluster/cluster_utils.cc
@@ -350,6 +350,16 @@ partition_raft_state get_partition_raft_state(consensus_ptr ptr) {
         raft_state.followers = std::move(followers);
     }
     raft_state.stms = get_partition_stm_state(ptr);
+
+    const auto& frs = ptr->get_follower_recovery_state();
+    if (frs) {
+        raft_state.recovery_state
+          = partition_raft_state::follower_recovery_state{
+            .is_active = frs->is_active(),
+            .pending_offset_count = frs->pending_offset_count(),
+          };
+    }
+
     return raft_state;
 }
 
diff --git a/src/v/cluster/types.h b/src/v/cluster/types.h
index 833ca521b8352..1bb55f3efd82f 100644
--- a/src/v/cluster/types.h
+++ b/src/v/cluster/types.h
@@ -3557,7 +3557,7 @@ struct partition_stm_state
 struct partition_raft_state
   : serde::envelope<
       partition_raft_state,
-      serde::version<1>,
+      serde::version<2>,
       serde::compat_version<0>> {
     using rpc_adl_exempt = std::true_type;
 
@@ -3619,10 +3619,32 @@ struct partition_raft_state
               suppress_heartbeats,
               is_recovering);
         }
+
+        friend bool operator==(const follower_state&, const follower_state&)
+          = default;
+    };
+
+    struct follower_recovery_state
+      : serde::envelope<
+          follower_recovery_state,
+          serde::version<0>,
+          serde::compat_version<0>> {
+        bool is_active = false;
+        int64_t pending_offset_count = 0;
+
+        auto serde_fields() {
+            return std::tie(is_active, pending_offset_count);
+        }
+
+        friend bool operator==(
+          const follower_recovery_state&, const follower_recovery_state&)
+          = default;
     };
 
     // Set only on leaders.
     std::optional<std::vector<follower_state>> followers;
+    // Set only on recovering followers.
+    std::optional<follower_recovery_state> recovery_state;
 
     auto serde_fields() {
         return std::tie(
@@ -3644,7 +3666,8 @@ struct partition_raft_state
           is_leader,
           is_elected_leader,
           followers,
-          stms);
+          stms,
+          recovery_state);
     }
 
     friend bool
diff --git a/src/v/raft/consensus.h b/src/v/raft/consensus.h
index 20d98bc5f6c38..23616164ddd79 100644
--- a/src/v/raft/consensus.h
+++ b/src/v/raft/consensus.h
@@ -496,6 +496,11 @@ class consensus {
 
     void reset_last_sent_protocol_meta(const vnode&);
 
+    const std::optional<follower_recovery_state>&
+    get_follower_recovery_state() const {
+        return _follower_recovery_state;
+    }
+
 private:
     friend replicate_entries_stm;
     friend vote_stm;
diff --git a/src/v/redpanda/admin/api-doc/debug.json b/src/v/redpanda/admin/api-doc/debug.json
index b4183f70868ad..45c329571bf01 100644
--- a/src/v/redpanda/admin/api-doc/debug.json
+++ b/src/v/redpanda/admin/api-doc/debug.json
@@ -913,6 +913,20 @@
                 }
             }
         },
+        "follower_recovery_state": {
+            "id": "follower_recovery_state",
+            "description": "Follower-side Raft recovery state",
+            "properties": {
+                "is_active": {
+                    "type": "boolean",
+                    "description": "True if recovery is currently allowed by the scheduler"
+                },
+                "pending_offset_count": {
+                    "type": "long",
+                    "description": "Difference between leader and our last offsets"
+                }
+            }
+        },
         "raft_replica_state": {
             "id": "raft_replica_state",
             "description": "Raft level state for a single replica of a partition",
@@ -998,6 +1012,10 @@
                         "type": "stm_state"
                     },
                     "description": "All snapshottable stms attached to this replica"
+                },
+                "follower_recovery_state": {
+                    "type": "follower_recovery_state",
+                    "description": "Raft recovery state if this replica is a follower in recovery"
                 }
             }
         },
diff --git a/src/v/redpanda/admin_server.cc b/src/v/redpanda/admin_server.cc
index 8fdae7d1c7fd1..89b96519909d2 100644
--- a/src/v/redpanda/admin_server.cc
+++ b/src/v/redpanda/admin_server.cc
@@ -4107,6 +4107,12 @@ void fill_raft_state(
         state.max_collectible_offset = stm.last_applied_offset;
         raft_state.stms.push(std::move(state));
     }
+    if (src.recovery_state) {
+        ss::httpd::debug_json::follower_recovery_state frs;
+        frs.is_active = src.recovery_state->is_active;
+        frs.pending_offset_count = src.recovery_state->pending_offset_count;
+        raft_state.follower_recovery_state = std::move(frs);
+    }
     replica.raft_state = std::move(raft_state);
 }
 ss::future<result<std::vector<cluster::partition_state>>>