v5.8.4

Make sure to check the breaking changes and known issues before upgrading to EMQX 5.8.4.

Enhancements

#13739 Support clear monitor (statistics) data for the whole cluster.

Send DELETE request to endpoint api/v5/monitor to clear all collected monitoring metrics.
#14247 Write client attribute named tns to log messages if such client attribute exists.
#14269 Added etc/base.hocon config file.

In this release, we introduced a new configuration file, etc/base.hocon, to enhance configuration management and clarity.

Previously, since emqx.conf was the only place for manually crafted configurations, and because it sits at the top-most layer of the configuration override system, it caused some confusion. While mutable (not read-only) configurations set in emqx.conf could be changed through the UI, API, or CLI and take effect immediately, those changes would not persist after a node restart — leading to inconsistent behavior.

To address this, we’ve added etc/base.hocon as a foundational configuration layer. The updated configuration precedence order, from top to bottom, is as follows:
1. Environment variables
2. etc/emqx.conf
3. data/configs/cluster.hocon
4. etc/base.hocon
The etc/base.hocon file serves as the base layer for configurations. While settings defined here can still be modified after the node starts, this layer ensures consistent override behavior.
#14353 Make session rebalance and evacuation more robust. Previously, session evacuation could enter a dead loop after some clustering errors.
#14360 Added listener shutdown counts labeled by shutdown reason to Prometheus metrics, under the emqx_client_disconnected_reason counters.

Example output:
```
emqx_client_disconnected_reason{node="emqx@127.0.0.1",reason="takenover"} 1
emqx_client_disconnected_reason{node="emqx@127.0.0.1",reason="kicked"} 1
```
Currently, this feature is limited to TCP and TLS listeners only.
#14369 Introduced two size-related functions in the rule engine:
- is_empty : return true if the map or array is empty
- map_size : return the size of a map

Bug Fixes

#14248 Fixed a class of issues where intermittent connectivity issues between cluster nodes could potentially cause partial loss of cluster-wide routing table state.
#14267 Do not redact secrets in logs and HTTP responses when the secret string is a file path (file:///path/to/the/secret).
#14272 auto_subscribe configuration loaded via CLI shows success but fails to take effect.
#14317 Prevent potential issues where APIs involving paging may return empty pages, in case the internal APIs will be subtly misused in the future.
#14318 Fixed the initialization of the HTTP connector state. When there was incoming traffic being handled by an HTTP action and its underlying connector restarted, cryptic crashes could be seen in the logs mentioning function_clause.

Example:
```
20:42:36.850 [error] msg: "resource_exception", info: #{error => {error, function_clause}, id => <<"action:http:a:connector:http:a">>, name => call_query, ...
```
#14319 Refactored resource management internal state machine. As consequence, some race condition bugs have been eliminated. One such example is the HTTP action, which, when under incoming traffic and when its health check flap, may produce errors like the following:
```
2024-11-29T14:58:17.994119+00:00 [error] msg: action_not_found, connector: <<"connector:http:a">>, action_id: <<"action:http:a:connector:http:a">>
```
#14357 Fix bin/emqx help command.
#14362 Refactored data integration resource manager state machine, which should eliminate a few race conditions that could lead to inconsistent states.
#14371 Fix clientid override expression to render undefined as empty string.
#14376 Made config import to tolerate non-existing log file dir.

If log file handler is configured with a non-existing dir, the default value it will fallback to the default value "${EMQX_LOG_DIR}".
#14424 Eliminated a false warning level "unexpected_info" log message.
#14429 Fixed handling of rule action metrics when an underlying connector is disabled. Before the fix, the failed counter would be bumped twice for each message the action received (one categorized under unknown and another under out_of_service). Now, only out_of_service is bumped.

Breaking Changes

#14360 When requesting Prometheus metrics in JSON format, now the client top-level key will always be an array of JSON objects instead of a single JSON object.