From aa87fecce2ea5beb34d8c2776f7e2161e6b3ecba Mon Sep 17 00:00:00 2001
From: Gleb Natapov <gleb@scylladb.com>
Date: Wed, 30 Oct 2024 15:51:21 +0200
Subject: [PATCH] gossiper: add is_alive that works on host_id

The function checks if a node with provided id is alive. If it fails to map id to ip
or there is no state for the ip found the node is considered to be dead.
---
 gms/gossiper.cc | 24 ++++++++++++++++++++++++
 gms/gossiper.hh |  3 +++
 2 files changed, 27 insertions(+)

diff --git a/gms/gossiper.cc b/gms/gossiper.cc
index 9671b50a95..ec642f3f89 100644
--- a/gms/gossiper.cc
+++ b/gms/gossiper.cc
@@ -2410,6 +2410,30 @@ bool gossiper::is_alive(inet_address ep) const {
     return is_alive;
 }
 
+bool gossiper::is_alive(locator::host_id id) const {
+    auto ip_opt = _address_map.find(id);
+
+    if (!ip_opt) {
+        // if host ID is not in the gossiper state (and hence not in the address map) it is dead
+        return false;
+    }
+
+    auto ep = get_endpoint_state_ptr(*ip_opt);
+    if (!ep) {
+        // _address_map may have stale entry since we rely on gc to remove entries there
+        // FIXME: add function to address_map to remove immediately
+        return false;
+    }
+
+    if (id != ep->get_host_id()) {
+        // If IDs do not match it means that the node with provided ID was replaced
+        // with a new node with same IP address and hence it is dead
+        return false;
+    }
+
+    return is_alive(*ip_opt);
+}
+
 future<> gossiper::wait_alive(std::vector<gms::inet_address> nodes, std::chrono::milliseconds timeout) {
     return wait_alive([nodes = std::move(nodes)] { return nodes; }, timeout);
 }
diff --git a/gms/gossiper.hh b/gms/gossiper.hh
index 5c5c44fd9b..a7cee90935 100644
--- a/gms/gossiper.hh
+++ b/gms/gossiper.hh
@@ -13,6 +13,7 @@
 #include <seastar/core/shared_ptr.hh>
 #include <seastar/core/gate.hh>
 #include <seastar/rpc/rpc_types.hh>
+#include "locator/host_id.hh"
 #include "utils/atomic_vector.hh"
 #include "utils/UUID.hh"
 #include "gms/generation-number.hh"
@@ -516,6 +517,8 @@ private:
 
 public:
     bool is_alive(inet_address ep) const;
+    bool is_alive(locator::host_id id) const;
+
     bool is_dead_state(const endpoint_state& eps) const;
     // Wait for nodes to be alive on all shards
     future<> wait_alive(std::vector<gms::inet_address> nodes, std::chrono::milliseconds timeout);