e5.0.1-rc.2

Changes on top of e5.0.0

Enhancements

• #10019 Add low level tuning settings for QUIC listeners.

• #10059 Errors returned by rule engine API are formatted in a more human readable way rather than dumping the raw error including the stacktrace.

• #9213 Add pod disruption budget to helm chart

• #9949 QUIC transport Multistreams support and QUIC TLS cacert support.

• #10011 Add pod disruption budget to helm chart

• #9932 Integrate TDengine into bridges as a new backend.

• #9967 New common TLS option 'hibernate_after' to reduce memory footprint per idle connecion, default: 5s.

Bug Fixes

• #10009 Validate bytes param to GET /trace/:name/log to not exceed signed 32bit integer.

• #10015 To prevent errors caused by an incorrect EMQX node cookie provided from an environment variable, we have implemented a fail-fast mechanism. Previously, when an incorrect cookie was provided, the command would still attempt to ping the node, leading to the error message 'Node xxx not responding to pings'. With the new implementation, if a mismatched cookie is detected, a message will be logged to indicate that the cookie is incorrect, and the command will terminate with an error code of 1 without trying to ping the node.

• #10020 Fix bridge metrics when running in async mode with batching enabled (batch_size > 1).

• #10021 Fix error message when the target node of emqx_ctl cluster join command is not running.

• #10032 When the resource manager is busy trying to establish a connection with the remote, the resource might yet lack any metrics information. Prior to this fix, the bridges/ API handler crashed in such circumstances.

• #10037 Fix Swagger API doc rendering crash. In version 5.0.18, a bug was introduced that resulted in duplicated field names in the configuration schema. This, in turn, caused the Swagger schema generated to become invalid.

• #10041 For influxdb bridge, added integer value placeholder annotation hint to write_syntax documentation. Also supported setting a constant value for the timestamp field.

• #10042 Improve behavior of the replicant nodes when the core cluster becomes partitioned (for example when a core node leaves the cluster). Previously, the replicant nodes were unable to rebalance connections to the core nodes, until the core cluster became whole again. This was indicated by the error messages: [error] line: 182, mfa: mria_lb:list_core_nodes/1, msg: mria_lb_core_discovery divergent cluster.

Mria PR

• #10043 Fixed two bugs introduced in v5.0.18.
• The environment varialbe SSL_DIST_OPTFILE was not set correctly for non-boot commands.

• When cookie is overridden from environment variable, EMQX node is unable to start.

• #10044 Fix node information formatter for stopped nodes in the cluster.

• #10058 Deprecate unused QUIC TLS options. Only following TLS options are kept for the QUIC listeners:

• cacertfile

• certfile
• keyfile

• verify

• #10076 Fix webhook bridge error handling: connection timeout should be a retriable error. Prior to this fix, connection timeout was classified as unrecoverable error and led to request being dropped.

• #10078 Fix an issue that invalid QUIC listener setting could casue segfault.

• #10084 Fix problem when joining core nodes running different EMQX versions into a cluster. Mria PR

• #10086 Upgrade HTTP client ehttpc to 0.4.7. Prior to this upgrade, HTTP clients for authentication, authorization and webhook may crash if body is empty but content-type HTTP header is set. For more details see ehttpc PR#44.

• #9939 Allow 'emqx ctl cluster' command to be issued before Mnesia starts. Prior to this change, EMQX replicant could not use manual discovery strategy. Now it's possible to join cluster using 'manual' strategy.

• #9958 Fix bad http response format when client ID is not found in clients APIs

• #9961 Avoid parsing config files for node name and cookie when executing non-boot commands in bin/emqx.

• #9974 Report memory usage to statsd and prometheus using the same data source as dashboard. Prior to this fix, the memory usage data source was collected from an outdated source which did not work well in containers.

• #9978 Fixed configuration issue when choosing to use SSL for a Postgres connection (authn, authz and bridge). The connection could fail to complete with a previously working configuration after an upgrade from 5.0.13 to newer EMQX versions.

• #9997 Fix Swagger API schema generation. deprecated metadata field is now always boolean, as Swagger specification suggests.

• #10007 Change Kafka bridge's config memory_overload_protection default value from true to false. EMQX logs cases when messages get dropped due to overload protection, and this is also reflected in counters. However, since there is by default no alerting based on the logs and counters, setting it to true may cause messages being dropped without noticing. At the time being, the better option is to let sysadmin set it explicitly so they are fully aware of the benefits and risks.

• #10087 Use default template ${timestamp} if the timestamp config is empty (undefined) when inserting data in InfluxDB. Prior to this change, InfluxDB bridge inserted a wrong timestamp when template is not provided.