emqx/changes/e5.1.0.en.md

e5.1.0

Enhancements

• #10584 Add log level configuration to SSL communication

• #10598 Provide a callback method of Unary type in ExProto to avoid possible message disorder issues.

• #10676 Implement configuration and user data import/export CLI.

  The emqx ctl export and emqx ctl import commands allow to export configuration and built-in database data from a running EMQX cluster and later import it to the same or another running EMQX cluster.

• #10702 Introduce a more straightforward configuration option keepalive_multiplier and deprecate the old keepalive_backoff configuration. After this enhancement, EMQX checks the client's keepalive timeout status period by multiplying the "Client Requested Keepalive Interval" with keepalive_multiplier.

• #10782 Added a new deliver_rate option to the retainer configuration, which can limit the maximum delivery rate per session in the retainer.

• #10858 A new utility function timezone_to_offset_seconds/1 has been added to the rule engine SQL language. This function converts a timezone string (for example, "+02:00", "Z" and "local") to the corresponding offset in seconds.

• #10909 Remove the deprecated HTTP APIs for gateways

• #10910 The data bridge resource option auto_restart_interval was deprecated in favor of health_check_interval, and request_timeout was renamed to request_ttl. Also, the default request_ttl value went from 15 seconds to 45 seconds.

  The previous existence of both auto_restart_interval and health_check_interval was a source of confusion, as both parameters influenced the recovery of data bridges under failures. An inconsistent configuration of those two parameters could lead to messages being expired without a chance to retry. Now, health_check_interval is used both to control the periodicity of health checks that may transition the data bridge into disconnected or connecting states, as well as recovering from disconnected.

• #10926 Allow 'enable' as well as 'enabled' as the state flag for listeners.

  Prior to this change, listener can be enable/disabled by setting the 'true' or 'false' on the 'enabled' config. This is slightly different naming comparing to other state flags in the system. No the 'enable' flag is added as an aliase on listeners.

• #10929 Upgrade Erlang/OTP to 25.3.2-1

• #10933 Add support for configuring TCP keep-alive in MQTT/TCP and MQTT/SSL listeners

• #10948 Add live_connections field for some HTTP APIs, i.e:

  • /monitor_current, /monitor_current/nodes/{node}
  • /monitor/nodes/{node}, /monitor
  • /node/{node}, /nodes

• #10961 Adds support for unlimited max connections for gateway listeners by allowing infinity as a valid value for the max_connections field in the configuration and HTTP API

• #10985 Renamed emqx ctl command 'cluster_call' to 'conf cluster_sync'. The old command 'cluster_call' is still a valid command, but not included in usage info.

• #11034 Hide the broker and move the broker.shared_subscription_strategy to mqtt.shared_subscription_strategy as it belongs to mqtt.

• #11045 The listener's authentication and zone related apis have been officially removed in version 5.1.0.

• #11062 Rename log.file.to to log.file.path.

• #10778 Refactored Pulsar Producer bridge to avoid leaking resources during crashes at creation.

• #10813 Refactored Kafka Producer and Consumer bridges to avoid leaking resources during crashes at creation.

• #10841 Added a schema validation to ensure message key is not empty when dispatching by key in Kafka and Pulsar Producer bridges.

• #10877 Bump RocketMQ driver to enhance security for sensitive data.

• #10892 Require that SID or Service Name is set on Oracle Database bridge creation.

• #10895 Refactored most of the bridges to avoid resource leaks during crashes during creation.

• #10908 Refactored the RocketMQ bridge to avoid leaking resources during crashes at creation.

• #10924 Refactored influxdb bridge connector to avoid resource leaks during crashes at creation.

• #10944 Improved the GCP PubSub bridge to avoid a potential issue where messages could fail to be sent when restarting a node.

• #10970 A query_mode parameter has been added to the Kafka producer bridge. This parameter allows you to specify if the bridge should use the asynchronous or synchronous mode when sending data to Kafka. The default is asynchronous mode.

• #11003 Add an option to configure TCP keepalive in Kafka bridge.

• #11024 Added a small improvement to reduce the chance of seeing the connecting state when creating/updating a Pulsar Producer bridge.

• #10678 Optimized counter increment calls to avoid work if increment is zero.

• #10690 Added a retry mechanism to webhook bridge that attempts to improve throughput.

  This optimization retries request failures without blocking the buffering layer, which can improve throughput in situations of high messaging rate.

• #10698 Optimize memory usage when accessing the configuration during runtime.

• #10754 The MQTT bridge has been enhanced to utilize connection pooling and leverage available parallelism, substantially improving throughput.

  As a consequence, single MQTT bridge now uses a pool of clientids to connect to the remote broker.

• #10790 Reducing overhead of reading configs per zone.

• #10941 Improve the collection speed of Prometheus metrics when setting prometheus.vm_dist_collector=disabled and metric erlang_vm_statistics_run_queues_length_total is renamed to erlang_vm_statistics_run_queues_length

• #10988 Improve log security when data bridge creation fails to ensure sensitive data is always obfuscated.

• #11019 Improve log security for JWT, now it will be obfuscated before print.

• #11035 Bump Cassandra driver to avoid sensitive data leakages.

Bug Fixes

• #10563 Corrected an issue where the no_local flag was not functioning correctly.

• #10600 Deleted emqx_statsd application.

• #10653 Store gateway authentication TLS certificates and keys in the data directory.

• #10677 In Rule API, reapond with 404 HTTP error code when trying to delete a rule that does not exist.

• #10682 Fix the timestamp for the will message is incorrectly assigned at the session creation time, now this timestamp is the disconnected time of the session.

• #10701 RPM package for Amazon Linux 2 did not support TLS v1.3 as it was assembled with Erlang/OTP built with openssl 1.0.

• #10715 Postpone trimming the connection information structure until after client.connected hooks have been executed. These hooks once again have access to the client's peer certificate.

• #10737 Fix the issue where the HTTP API interface of Gateway cannot handle ClientIDs with special characters, such as: !@#$%^&*()_+{}:"<>?/.

• #10785 Ensure EMQX_LOG_DIR is set by Windows boot script.

  The environment variable EMQX_LOG_DIR was missing in v5.0.25, caused EMQX Windows package fail to boot unless set by sysadmin.

• #10809 Address ** ERROR ** Mnesia post_commit hook failed: error:badarg error messages happening during node shutdown or restart. Mria pull request: https://github.com/emqx/mria/pull/142

• #10818 Fixing emqx_ctl traces command.

• #10820 In case the cluster updated license before the new node join in. The new node will not apply the updated license. After this change, the new joined node will use the cluster's license key.

  Sometimes the new node must start with a outdated license. e.g. use emqx-operator deployed and needed to scale up after license expired. At the time the cluster's license key already updated by API/CLI, but the new node won't use it.

• #10833 Only include enabled authenticators and authorizers in telemetry report, not all of them.

• #10851 Obfuscated sensitive data in the bad API logging.

• #10871 Fixes for connection deletion and message publishing requests not taking effect issues once the connection has been created in a different UDP port first.

• #10880 Add a new HTTP API endpoint /clients/kickout/bulk for kicking out multiple clients in bulk.

• #10884 Fixes an issue where trying to get rule info or metrics could result in a crash when a node is joining a cluster.

• #10887 Fixed a potential issue where requests to bridges might take a long time to be retried.

  This only affected low throughput scenarios, where the buffering layer could take a long time to detect connectivity and driver problems.

• #10902 Avoid syncing cluser.hocon file from the nodes runing a newer version than self.

  During cluster rolling upgrade, if an older version node has to restart due to whatever reason, if it copies the cluster.hocon file from a newer version node, it may fail to start. After this fix, the older version node will not copy the cluster.hocon file from a newer, so it will use its own cluster.hocon file to start.

• #10911 The error message and log entry that appear when one tries to create a bridge with a name the exceeds 255 bytes is now easier to understand.

• #10923 Fix a race-condition in channel info registration.

  Prior to this fix, when system is under heavy load, it might happen that a client is disconnected (or has its session expired) but still can be found in the clients page in dashboard. One of the possible reasons is a race condition fixed in this PR: the connection is killed in the middle of channel data registration.

• #10930 Added a schema validation for values that might be used in timeouts to avoid invalid values.

  Before this fix, it was possible to use absurd values in the schema that would exceed the system limit, causing a crash.

• #10943 Deprecated UDP mcast mechanism for cluster discovery.

  This feature has been planed for deprecation since 5.0 mainly due to the lack of actual production use. This feature code is not yet removed in 5.1, but the document interface is demoted.

• #10950 Fix the issue where the enable_qos option does not take effect in the MQTT-SN gateway.

• #10951 Fix the issue in MQTT-SN gateway where the mountpoint does not take effect on message publishing.

• #10952 Disallow enabling fail_if_no_peer_cert in listener SSL options if verify_none is set.

  Setting fail_if_no_peer_cert = true and verify = verify_none caused connection errors due to incompatible options. This fix validates the options when creating or updating a listener to avoid these errors.

  Note: any old listener configuration with fail_if_no_peer_cert = true and verify = verify_none that was previously allowed will fail to load after applying this fix and must be manually fixed.

• #10955 Fix the issue in MQTT-SN gateway where deleting Predefined Topics configuration does not work.

• #10977 Fix delay in updating subscription count metric and correct configuration issues in Stomp gateway.

• #10983 Fix issue when mqtt clients could not connect over TLS if the listener was configured to use TLS v1.3 only.

  The problem was that TLS connection was trying to use options incompatible with TLS v1.3.

• #10994 Redact proxy-authorization headers as used by HTTP connector to not leak secrets into log-files.

• #10996 Default to 404 for any HTTP/API request, we don't know, rather than the dashboard's index.html.

• #11005 Fix the issue where the method field cannot be correctly printed in the trace logs of AuthN HTTP.

• #11018 Fixed multiple issues with the Stomp gateway, including:

  • Fixed an issue where is_superuser was not working correctly.
  • Fixed an issue where the mountpoint was not being removed in message delivery.
  • After a message or subscription request fails, the Stomp client should be disconnected immediately after replying with an ERROR message.

• #11028 Disallow using multiple TLS versions in the listener config that include tlsv1.3 but exclude tlsv1.2.

  Using TLS configuration with such version gap caused connection errors. Additionally, drop and log TLS options that are incompatible with the selected TLS version(s).

  Note: any old listener configuration with the version gap described above will fail to load after applying this fix and must be manually fixed.

• #11030 Improved error messages when a validation error occurs while using the Listeners HTTP API.

• #11033 Deprecates the mountpoint field in AuthenticateRequest in ExProto gateway.

  This field was introduced in v4.x, but in fact, in 5.0 we have provided gateway.exproto.mountpoint for configuration, so there is no need to override it through the Authenticate request.

  Additionally, updates the default value of subscriptions_max, inflight_max, mqueue_max to infinity

• #11042 Fix crash on /api/listeners when listener's max_connections is set to a string.

• #11051 Add validation to ensure that certificate 'depth' (listener SSL option) is a non negative integer.

• #11056 - Fix the issue where newly created listeners do not start properly at times, when you delete a system default listener and add a new one named 'default', it will not start correctly.

  • Fix the bug where configuration failure on certain nodes can cause dashboard unavailability.

• #11070 Fix the problem that the cluster.autoclean configuration item does not take effect.

• #11092 Fix problem when replicants were unable to connect to the core node due to timeout in mria_lb:core_nodes() call. Relevant mria pull request: https://github.com/emqx/mria/pull/143

• #10807 Removed license check debug logs.

• #10878 A vulnerability in the RabbitMQ bridge, which could potentially expose passwords to log files, has been rectified

• #10913 Fixed an issue where a node that left the cluster would still report plugin status from other nodes.

• #10967 Fixed error message formatting in rebalance API: previously they could be displayed as unclear dumps of internal Erlang structures.

  Added wait_health_check option to node evacuation CLI and API. This is a time interval when the node reports "unhealthy status" without beginning actual evacuation. We need this to allow a Load Balancer (if any) to remove the evacuated node from balancing and not forward (re)connecting clients to the evacuated node.

• #10997 The ClickHouse bridge had a problem that could cause messages to be dropped when the ClickHouse server is closed while sending messages even when the request_ttl is set to infinity. This has been fixed by treating errors due to a closed connection as recoverable errors.

• #10998 Do not allow batch_size option for MongoDB bridge resource. MongoDB connector currently does not support batching, the bath_size config value is forced to be 1 if provided.

• #10999 Changed schema validation for Kafka fields 'Partition Count Refresh Interval' and 'Offset Commit Interval' to avoid accepting values larger then maximum allowed.

• #11006 Fix QUIC listeners's default cert file paths.

  Prior to this change, the default cert file paths are prefixed with environment variable ${EMQX_ETC_DIR} which were not interpolated before used in QUIC listeners.

• #11025 Fixed a case_clause error that could arise in race conditions in Pulsar Producer bridge.

• #11031 Fixed credential validation when creating bridge and checking status for InfluxDB Bridges.

• #11038 Fixed a health check issue for Pulsar Producer that could lead to loss of messages when the connection to Pulsar's brokers were down.

• #11040 Fixed a health check issue for Kafka Producer that could lead to loss of messages when the connection to Kafka's brokers were down.