Initially this was added to make development using Docker Compose easier.
However, in my experience Docker Compose is not great for the reason
that there might be environmental differences between Rust running in
Compose and Rust running outside Compose (if using VSCode with Rust
plugin, then rust-analyzer will perform checks in the background). This
and 'cross' not working properly in Compose was the reason to migrate
to nix, as it provides a reproducible environment for development that
can also be used by VSCode (using the Nix environment selector
extension) and GitHub Actions (for the CI pipeline).
As the chirpstack-docker repository provides its own configuration files,
these files are not included in the Dockerfile and because we are
rewriting these env variables anyway to 'localhost' in the .deb and .rpm
post-installation script, it is better to set these to localhost by
default.
The post-installation rewrite to 'localhost' is causing issues on
upgrade (#295). If we only do the rewrite on initial installation, then
we run in an other issue; the package-manager will prompt that the
config file has changed (from 'localhost' to '$MQTT_BROKER_HOST' for
example) and will ask if you would like to overwrite or not. If the
end-user would accept the config changes without looking at the diff
this would break the installation as most likely, these environment are
not set (which is why we were rewriting these to 'localhost' in the first
place).
If enabled, metrics will be automatically detected and created based on
the decoded payload. The key will be set, but the name is left blank for
the user to configure. This can be confusing as it leads to validation
errors in the device-profile form if the name is not filled in.
The UI does show that one of the fields failed to validate, but even
with such notification, it is not clear that users have to check one
of the form tabs which they didn't edit on save.
This makes the name optional and uses the key as fallback such that
saving the form no longer fails in case the name is still blank.
Closes#313.
This removes the aws-sdk-sns crate (+ dependencies) and refactors the
AWS SNS integration to use reqwest for the API call + aws-sign-v4 for
creating the AWS request signature.
Even with the messages being handled async (tokio::spawn(...)), during
high throughput or a burst of messages, the channel might still fill up
causing messages to be discarded without any error being printed.
This makes it possible to use a range of JoinEUIs per Join Server.
Use-cases are either Join Servers using a JoinEUI range or the
configuration of a "catch-all" Join Server prefix ("0000000000000000/0").
This improves the log output on the following points:
In case when .context(...) is used, the error printed in the logs would
only show the outer error. This has been improved to show the complete
chain of errors.
In many cases error! and warn! logs were used for logging related to
user input. This is very inconvenient when using log output for
monitoring as it can trigger alerts based on end-user mistakes. While
refactoring the logs, the following criteria has been used:
* error = Messages that need attention and that are within the domain
of the ChirpStack administrator.
* warn = Messages that should stand out of the other log messages, but
that might be end-user (or end-device) related. Depending on if you
have control of the full chain or not, you might or might not want to
be alerted based on these messages.
Tracing spans have been improved to make it easier to correlate between
log messages and events.
Note that the integration events will contain the application +
device-profile + device tags. Integration events will NOT contain the
tenant tags. Most likely tenant tags will be used to store information
about the tenant, data that is unrelated to the integration events.
Fixes#211.
Using sub-queries, the amount of potentially selected records is
reduced. As well the joins on api_key.tenant_id causes (or might
cause) seq scans, because api_key.tenant_id is nullable.
The gw_time defines the RX time by the gateway, the ns_time defines when
it was received by the NS. The latter could for example help to debug
latency between the GW <> NS.
The JSON encoding should only be used for debugging purposes! However
this change avoids showing errors in case there are unknown fields in the
JSON payload. This would happen when for example the MQTT Forwarder
and ChirpStack uses a different API version (which in case of Protobuf
would be fine, as long as the major version remains the same).
