From 23b2ae112ebfc12a279fb5625d12570f029e71d8 Mon Sep 17 00:00:00 2001
From: Chrono <chrono_cpp@me.com>
Date: Wed, 27 Nov 2024 10:12:49 +0800
Subject: [PATCH] fix(clustering/rpc): try sync after rpc is ready (#13901)

KAG-5569, KAG-5724
---
 kong/clustering/services/sync/init.lua |  3 +--
 kong/clustering/services/sync/rpc.lua  | 22 ++++++++++++++++++++--
 2 files changed, 21 insertions(+), 4 deletions(-)

diff --git a/kong/clustering/services/sync/init.lua b/kong/clustering/services/sync/init.lua
index 40f1b836241b..06802f6088db 100644
--- a/kong/clustering/services/sync/init.lua
+++ b/kong/clustering/services/sync/init.lua
@@ -7,8 +7,7 @@ local strategy = require("kong.clustering.services.sync.strategies.postgres")
 local rpc = require("kong.clustering.services.sync.rpc")
 
 
--- TODO: what is the proper value?
-local FIRST_SYNC_DELAY = 0.5  -- seconds
+local FIRST_SYNC_DELAY = 0.2  -- seconds
 local EACH_SYNC_DELAY  = 30   -- seconds
 
 
diff --git a/kong/clustering/services/sync/rpc.lua b/kong/clustering/services/sync/rpc.lua
index 1e80906b4649..5783107f0caf 100644
--- a/kong/clustering/services/sync/rpc.lua
+++ b/kong/clustering/services/sync/rpc.lua
@@ -181,7 +181,27 @@ function _M:init(manager, is_cp)
 end
 
 
+-- check if rpc connection is ready
+local function is_rpc_ready()
+  for i = 1, 5 do
+    local res = kong.rpc:get_peers()
+
+    -- control_plane is ready
+    if res["control_plane"] then
+      return true
+    end
+
+    -- retry later
+    ngx.sleep(0.1 * i)
+  end
+end
+
+
 local function do_sync()
+  if not is_rpc_ready() then
+    return nil, "rpc is not ready"
+  end
+
   local ns_deltas, err = kong.rpc:call("control_plane", "kong.sync.v2.get_delta",
                                        { default =
                                          { version =
@@ -394,8 +414,6 @@ end
 
 
 function _M:sync_once(delay)
-  --- XXX TODO: check rpc connection is ready
-
   return start_sync_timer(ngx.timer.at, delay or 0)
 end