Fixes https://emqx.atlassian.net/browse/EMQX-12442
e.g.:
```
2024-05-23T08:52:39.811845+00:00 [error] Supervisor: {local,emqx_resource_manager_sup}. Context: shutdown_error. Reason: noproc. Offender: id=<<99, 101, 110, 115, 111, 114, 101, 100>>,pid=<0.7752.1030>.
```
It could be just a race condition, as it seems to be the case for resource manager: i) a call is made to the process to stop it; ii) the call times out; iii) the after clause ends up calling supervisor:terminate_child; iv) while the supervisor is finding the child to terminate, the process actually finishes terminating, and the supervisor receives a noproc reason back.
This commit fixes an issue found by CI test case
emqx_bridge_influxdb_SUITE:t_start_stop and others. While the channel
health check process is running, the channel could be removed or updated
which could cause a crash in the resource manager or non up-to-date
alarms being triggered.
Before this commit the following happened sometimes:
1. action status is connected
2. action config is updated to something that should change the status to
disconnected
3. action status is still connected and the old config is being used by
the connector even though the config has been correctly updated.
The reason for this bug is that the post_config_hook runs before the
global EMQX config is updated. The post config hook is adding the new
config to the connector. Since the new config causes the action to get
status disconnected, the adding of the action to the connector is
retried when the health check runs but this time the config will be
loaded from the global config which could cause the old config to be
loaded (this happens when the global config has not had time to get
updated).
The above problem has been fixed in this commit by only reading action
configs from the global config when the connector starts/restarts and
instead store the latest configs for the actions in the connector.
Fixes:
https://emqx.atlassian.net/browse/EMQX-12376
This commit renames trace fields rule_trigger_time and
rule_trigger_times to rule_trigger_ts and makes sure that the value for
rule_trigger_ts will always be a list of timestamps.
* Simpler handling of true and false in best effort JSON formatter
* inet:ntoa/1 to format IP addresses
* Made a record for lazy formatted trace values and formatter to improve
maintainability
* Added callback to format return value from connector
* Extended test case to check that the format return value callback
works
* Added handling of "lazy" trace entry data to the text formatter. Do we
need to handle this data in the normal log formatters as well?