From 5b420ededadfb5efe0bebdf2333b7b8bbd957156 Mon Sep 17 00:00:00 2001
From: Michael Martin <michael.martin@konghq.com>
Date: Fri, 3 Nov 2023 17:25:06 -0700
Subject: [PATCH] fix(clustering): ensure data plane config hash is never nil

The previous logic defaulted the config_hash to nil when it was
detected to be an empty string. This can cause update_sync_status() to
fail, because config_hash is a required attribute:

> 2023/11/03 17:13:30 [debug] 4052224#0: *150 [lua] connector.lua:560: execute(): SQL query throw error: ERROR: null value in column "config_hash" of relation "clustering_data_planes" violates not-null constraint
> Failing row contains (4fb29006-8db1-48bb-b68c-34b582e1d91a, soup, 127.0.0.1, 2023-11-04 00:13:30+00, null, 2023-11-18 00:13:30.799+00, 3.6.0, filter_set_incompatible, 2023-11-04 00:13:30+00, {})., close connection
> 2023/11/03 17:13:30 [notice] 4052224#0: *150 [lua] init.lua:275: upsert(): ERROR: null value in column "config_hash" of relation "clustering_data_planes" violates not-null constraint

This change addresses the problem from two angles:

1. when empty, config_hash is set to the default DECLARATIVE_EMPTY_CONFIG_HASH
   constant instead of nil
2. an additional guard was added to the dp reader thread, which checks the
   length of ping frame data and returns an error if it is not a proper
   config hash
---
 .../kong/clustering-empty-data-plane-hash-fix.yml         | 3 +++
 kong/clustering/control_plane.lua                         | 8 +++++++-
 kong/clustering/data_plane.lua                            | 2 +-
 3 files changed, 11 insertions(+), 2 deletions(-)
 create mode 100644 changelog/unreleased/kong/clustering-empty-data-plane-hash-fix.yml

diff --git a/changelog/unreleased/kong/clustering-empty-data-plane-hash-fix.yml b/changelog/unreleased/kong/clustering-empty-data-plane-hash-fix.yml
new file mode 100644
index 00000000000..1c405ecd53f
--- /dev/null
+++ b/changelog/unreleased/kong/clustering-empty-data-plane-hash-fix.yml
@@ -0,0 +1,3 @@
+message: Fix a bug causing data-plane status updates to fail when an empty PING frame is received from a data-plane
+type: bugfix
+scope: Clustering
diff --git a/kong/clustering/control_plane.lua b/kong/clustering/control_plane.lua
index 6939d7a78a5..7efa3d40d25 100644
--- a/kong/clustering/control_plane.lua
+++ b/kong/clustering/control_plane.lua
@@ -234,7 +234,9 @@ function _M:handle_cp_websocket()
     local ok
     ok, err = kong.db.clustering_data_planes:upsert({ id = dp_id, }, {
       last_seen = last_seen,
-      config_hash = config_hash ~= "" and config_hash or nil,
+      config_hash = config_hash ~= ""
+                and config_hash
+                 or DECLARATIVE_EMPTY_CONFIG_HASH,
       hostname = dp_hostname,
       ip = dp_ip,
       version = dp_version,
@@ -341,6 +343,10 @@ function _M:handle_cp_websocket()
 
       if not data then
         return nil, "did not receive ping frame from data plane"
+
+      elseif #data ~= 32 then
+        return nil, "received a ping frame from the data plane with an invalid"
+                 .. " hash: '" .. tostring(data) .. "'"
       end
 
       -- dps only send pings
diff --git a/kong/clustering/data_plane.lua b/kong/clustering/data_plane.lua
index 4030b3174b0..f82dda86bfc 100644
--- a/kong/clustering/data_plane.lua
+++ b/kong/clustering/data_plane.lua
@@ -91,7 +91,7 @@ local function send_ping(c, log_suffix)
 
   local hash = declarative.get_current_hash()
 
-  if hash == true then
+  if hash == "" or type(hash) ~= "string" then
     hash = DECLARATIVE_EMPTY_CONFIG_HASH
   end