From 13c6c62d8528872e07fb823522f859e3537db2e5 Mon Sep 17 00:00:00 2001
From: ieQu1 <99872536+ieQu1@users.noreply.github.com>
Date: Tue, 26 Mar 2024 13:32:14 +0100
Subject: [PATCH] fix(mria): Strictly check compatibility of the replicant with
 core

---
 apps/emqx/rebar.config                 |  2 +-
 apps/emqx_machine/src/emqx_machine.erl | 28 ++++++++++++++++++++++++++
 changes/ce/fix-12786.en.md             |  2 ++
 rebar.config                           |  2 +-
 4 files changed, 32 insertions(+), 2 deletions(-)
 create mode 100644 changes/ce/fix-12786.en.md

diff --git a/apps/emqx/rebar.config b/apps/emqx/rebar.config
index 3822078b76..14b041615b 100644
--- a/apps/emqx/rebar.config
+++ b/apps/emqx/rebar.config
@@ -28,7 +28,7 @@
     {gproc, {git, "https://github.com/emqx/gproc", {tag, "0.9.0.1"}}},
     {cowboy, {git, "https://github.com/emqx/cowboy", {tag, "2.9.2"}}},
     {esockd, {git, "https://github.com/emqx/esockd", {tag, "5.11.1"}}},
-    {ekka, {git, "https://github.com/emqx/ekka", {tag, "0.19.0"}}},
+    {ekka, {git, "https://github.com/emqx/ekka", {branch, "dev/custom-core-compat-cb"}}},
     {gen_rpc, {git, "https://github.com/emqx/gen_rpc", {tag, "3.3.1"}}},
     {hocon, {git, "https://github.com/emqx/hocon.git", {tag, "0.42.1"}}},
     {emqx_http_lib, {git, "https://github.com/emqx/emqx_http_lib.git", {tag, "0.5.3"}}},
diff --git a/apps/emqx_machine/src/emqx_machine.erl b/apps/emqx_machine/src/emqx_machine.erl
index d8858ff16a..e676f16cb2 100644
--- a/apps/emqx_machine/src/emqx_machine.erl
+++ b/apps/emqx_machine/src/emqx_machine.erl
@@ -51,6 +51,26 @@ start() ->
     configure_shard_transports(),
     set_mnesia_extra_diagnostic_checks(),
     emqx_otel_app:configure_otel_deps(),
+    %% Register mria callbacks that help to check compatibility of the
+    %% replicant with the core node. Currently they rely on the exact
+    %% match of the version of EMQX OTP application:
+    application:load(mria),
+    application:load(emqx),
+    mria_config:register_callback(
+        lb_custom_info,
+        fun() ->
+            get_emqx_vsn()
+        end
+    ),
+    mria_config:register_callback(
+        lb_custom_info_check,
+        fun
+            (undefined) ->
+                false;
+            (OtherVsn) ->
+                get_emqx_vsn() =:= OtherVsn
+        end
+    ),
     ekka:start(),
     ok.
 
@@ -227,3 +247,11 @@ resolve_dist_address_type() ->
         _ ->
             inet
     end.
+
+get_emqx_vsn() ->
+    case application:get_key(emqx, vsn) of
+        {ok, Vsn} ->
+            Vsn;
+        undefined ->
+            undefined
+    end.
diff --git a/changes/ce/fix-12786.en.md b/changes/ce/fix-12786.en.md
new file mode 100644
index 0000000000..49acef7da8
--- /dev/null
+++ b/changes/ce/fix-12786.en.md
@@ -0,0 +1,2 @@
+Add a strict check that prevents replicant nodes from connecting to the core nodes running with a different version of EMQX application.
+Effectively it means that during the rolling upgrades the replicant nodes can only work if there is at least one core node with the matching EMQX release.
diff --git a/rebar.config b/rebar.config
index 6713b5c9bd..6a9a85ef76 100644
--- a/rebar.config
+++ b/rebar.config
@@ -83,7 +83,7 @@
     {cowboy, {git, "https://github.com/emqx/cowboy", {tag, "2.9.2"}}},
     {esockd, {git, "https://github.com/emqx/esockd", {tag, "5.11.1"}}},
     {rocksdb, {git, "https://github.com/emqx/erlang-rocksdb", {tag, "1.8.0-emqx-2"}}},
-    {ekka, {git, "https://github.com/emqx/ekka", {tag, "0.19.0"}}},
+    {ekka, {git, "https://github.com/emqx/ekka", {branch, "dev/custom-core-compat-cb"}}},
     {gen_rpc, {git, "https://github.com/emqx/gen_rpc", {tag, "3.3.1"}}},
     {grpc, {git, "https://github.com/emqx/grpc-erl", {tag, "0.6.12"}}},
     {minirest, {git, "https://github.com/emqx/minirest", {tag, "1.4.0"}}},