| Age | Commit message (Collapse) | Author |
|---|
|
If you use ezau on Windows, you might prefer to use the built-in zipping
functionality from arcDPS instead of relying on ezau to do this job.
However, that would lead to weird interactions because arcDPS would
still create the temporary file in the watched folder, and
powershell would race with ezau to zip and delete this temporary file.
To prevent this from breaking existing (& working) configurations - and
to stick true to the name - zipping is enabled by default if not given
otherwise in the configuration.
|
|
If we don't do this, we end up trying to decode the error page as the
expected JSON response, which usually ends up with a "missing permalink"
field, which doesn't tell us a lot.
This way, we get a proper 404/500 error.
|
|
As it turns out, uploading is often the reason why the process
crashes/exits. This is bad because it means that 1) we lose links to
logs (as they are not being uploaded), leading to incomplete reporting
and 2) we rely on an external watchdog to keep the service alive (and
I'd rather just not have ezau crashing, especially on Windows where we
usually don't supervise it with systemd).
Therefore, a configuration setting has been added that lets ezau retry
the upload process. This is not 100% good and failsafe, because
1) it always waits a hardcoded amount of seconds (instead of e.g. using
a proper backoff timer)
2) it blocks the rest of the process, so no logs will be compressed
while it is retrying a single log
3) after those retries, the process will still exit
But it is a good first approximation, and the aforementioned issues can
be fixed "relatively easily" (e.g. by moving the whole per-log logic
into a separate thread(pool) and handling failures even better).
|
|
Some error messages currently look very weird. For example, if the given
configuration file does not exist, it gives you an error about a missing
file - which could also be the file to upload though, as the error
doesn't specify. Therefore, some more context for the error messages is
nice.
The "sourceback" could still use some work.
|
|
This is important to signal e.g. systemd that there was an error and the
process should be restarted.
|
|
This might help with identifying in the logs when/if ezau was started,
and when the control flow returned from the Discord client to ezau.
|
|
|
|
ezau having the watching functionality is nice, but sometimes for
scripts you might want to have the old "upload this single log and post
it to discord" functionality. As such, ezau has now been split into two
subcommands (which use the same core):
ezau watch runs the inotify-based directory watcher to zip and upload
new logs. Additionally, it now respects the "upload = ..." config
settings, which means you can also use it as a zipper only, without
having every log uploaded.
ezau upload performs a single-shot upload with the discord notification.
Furthermore, the discord auth token/channel id have been moved to a
configuration file. Switches to override this for single runs might be
provided in the future, but for now, it seems more sensible to have it
in a persistent configuration.
|
|
evtc-watch consists of three parts at the moment: watch the files, zip
them up and call ezau to upload them. We can now just do all of those
inside of ezau, which saves us the extra script, makes it more
platform-independent (as notify also works on Windows) and makes
configuration and everything easier, as all the data will be inside of
one program and doesn't need to be passed around.
A flag (or subcommand!) to upload a single file might be added later to
retain the previous behaviour of ezau.
|
|
|
