- Distingush more explictly between mobile and web flow and simplify logic
- Allow state parameter to be passed in mobile flow
- Additional checks for correct parameters
- Remove unused id_token code
- Enforce S256 and don't allow plain PKCE
Logger.dev is kind of in a weird spot where it doesn't fit into the
standard log level. It is called directly by some code and it only
checks whether a property is set (which comes from an env var) before
deciding to print out.
This standardizes on `debug` by changing the dev calls to debug. Also
removes the now unused code.
This PR adds `tini` to the container image and uses it as PID 1 when
starting the container. This ensures that proper PID 1 signal-handling
is implemented and passed to the underlying node.js process, thereby
ensuring that the ABS process has a chance to receive and handle signals
other than `SIGKILL`, such as the important `SIGINT`.
This is somewhat related to #2445 . Without this, the signal handled by
2445 won't be received when running in a container.
Some background:
In linux, PID 1 has special duties involving signal handling that are
different than other processes. Node doesn't properly handle these
signals, which can lead to a number of problems ranging from annoying to
disruptive. PID 1 also has reaping duties that can lead to resource
exhaustion if not properly handled.
For example, the container ignores `SIGINT` (Ctrl+C) as well as `docker stop`,
which can be annoying in development as you have to kill or wait for
the timeout to be reached. In a production environment (such as Kubernetes)
this can lead to signal escalation and unnecessarily adds delays to
deployments and restarts as K8s has to wait for the timeout to be reached
before sending `SIGKILL`.
At best this is annoying and unnecessarily adds
delays. At worst this can lead to file/data corruption as the process
doesn't get a chance to clean anything up when it is sent `SIGKILL`.
Without a proper PID 1 to forward signals, only SIGKILL can be used to
terminate the running process.
