feat: implement direct message passing on a cluster by kenkalang · Pull Request #99 · gojek/vernemq

kenkalang · 2026-04-17T08:48:55Z

Proposed Changes

Implement direct message passing node to node to reduce latency

Types of Changes

What types of changes does your code introduce to this project?
Put an x in the boxes that apply

Bugfix (non-breaking change which fixes issue #XXXX)
New feature (non-breaking change which adds functionality)
Breaking change (fix or feature that would cause existing functionality to not work as expected)
Documentation (correction or otherwise)
Cosmetics (whitespace, styles...)
DevOps (Build scripts, pipelines...)

Checklist

I have read the CODE_OF_CONDUCT.md document
All tests pass locally with my changes
I have added tests that prove my fix is effective or that my feature works
I have added necessary documentation (if needed)
Any dependent changes have been merged and published in related repositories
I have updated changelog (At the bottom of the release version)
I have squashed all my commits into one before merging

VivekPipaliya23 · 2026-05-04T10:10:44Z

    shared_subs_prefer_local_policy_test_with_local_caching/1,
    shared_subs_local_only_policy_test/1,
    shared_subs_local_only_policy_test_with_local_caching/1,
-    shared_subs_random_policy_dead_node_message_reaper_test/1,


why this test was removed, put this behind config while running if it requires redis.

Updated, please check latest commit

VivekPipaliya23 · 2026-05-04T10:19:51Z

+            _ -> false
+        end,
    vmq_state_store_backend:del_reaper(Node),
    ets:insert(?VMQ_CLUSTER_STATUS, {Node, true, 0}),


this insert is always getting overridden below, remove if not needed.

This is deleted

VivekPipaliya23 · 2026-06-02T10:21:47Z

+-spec delete(_) -> 'true' | {'error', 'redis_error'}.
 delete(SubscriberId) ->
-    case vmq_state_store_backend:msg_store_delete(SubscriberId) of
+    case


Why are you adding this back? If you need to change the Redis call, then please change the parent call from the vmq_state_store_backend > vmq_redis_store & vmq_noop_store.

VivekPipaliya23 · 2026-06-02T10:22:16Z

-    case vmq_state_store_backend:msg_store_pop(SubscriberId, MsgRef) of
+-spec delete(subscriber_id(), _) -> 'true' | {'error', 'redis_error'}.
+delete(SubscriberId, _MsgRef) ->
+    case


Why are you adding this back? If you need to change the Redis call, then please change the parent call from the vmq_state_store_backend > vmq_redis_store & vmq_noop_store.

VivekPipaliya23 · 2026-06-02T10:24:29Z

            ets:insert(?OFFLINE_MESSAGES, {count, binary_to_integer(OfflineMsgCount)});
        {error, _} ->
-            {error, not_supported}
+            {error, redis_error}


Can you share the reason behind the error message change?

Will revert this

…for toggling

…st on shared

dhruvjain99 · 2026-07-10T19:02:10Z


-stop(_State) ->
+stop(State) ->
+    %% Catch and log the termination event using VerneMQ's native Lager logger


Is it only for debugging and testing in staging env? Do we need this in prod?

dhruvjain99 · 2026-07-10T19:24:34Z

  {validators, ["buffer_size_validator"]}
 ]}.

+{mapping, "listener.vmq.$name.buffer_sizes", "vmq_server.listeners",


How is this config used?

This is used on both receiver and sender sides with the values being [send buf, rec buf, erl buffer]. Currently we're using default value on receiver side and default OS value on sender side

dhruvjain99 · 2026-07-13T01:44:15Z

+                                                                    {datatype, string},
+                                                                    hidden
+                                                                   ]}.
+{mapping, "listener.vmqs.mountpoint", "vmq_server.listeners", [


I think ssl connection configurations are not complete. Let's skip ssl for now.

We can add support for SSL later when required.

Ok let's put it as improvement later

dhruvjain99 · 2026-07-13T01:56:42Z

-            "cache_shared_subscriptions_locally"
+            "cache_shared_subscriptions_locally",
+            "direct_message_passing",
+            "cluster_node_liveness_rpc_timeout"


Do we need this as runtime configurable?

for cluster_node_liveness_rpc_timeout yes we don't need this, will remove this as part as clean up

dhruvjain99 · 2026-07-13T01:56:48Z

            "graphite_interval",
            "graphite_include_labels",
            "shared_subscription_policy",
+            "remote_enqueue_timeout",


Do we need this as runtime configurable?

dhruvjain99 · 2026-07-13T02:03:03Z

 {cluster_default_port, 44053}.
 {http_default_ip, "127.0.0.1"}.
 {http_default_port, 8888}.
 {metadata_plugin, vmq_plumtree}.


We are not using this as well. Can you just verify and remove it if it is unused?

kenkalang closed this Apr 17, 2026

kenkalang reopened this Apr 27, 2026

kenkalang force-pushed the feature/direct-message-passing branch 5 times, most recently from 3532e68 to e87d5c3 Compare April 29, 2026 05:59