As much as I like the idea, I tried to find any real life use case for
it. Biggest issue for me is that such subsystem will spam my
email/jabber account with too much messages and that will make it
unusable.

* in case of application internal errors it's much better to use tools
like airbrake/errbit, not on I will be notified much less frequently,
I will alway have much more usefull information about the issue
* in case of 404 or other events that are not directly errors that
needs to be resolved, but rather pointers that suggest that there
might be an error (for example if number of 404 per second spikes up,
it might mean that there is invalid css or js file linked in my app)
IMHO it's better to use log analyzer tools like graylog2, I can create
a stream there and set some limits for number of logs that contains
some pattern

But one thing that I could use it for is tracking harakiri with
errbit, I could hookup log-alarm with some handler that would push
harakiri backtrace to errbit so that I could analyze if I'm have the
same issue all the time, or is it random view that is causing
harakiri.
But I'm not sure if I could do this with log-alarm since it works per
log line and backtrace is logged as multiple log lines, so the only
thing I can get using it is the url on which harakiri occured. To get
around this harakiri would need to push whole backtrace as single log
message or log-alarm would need to be able to extract more lines from
log, not only one, for example:

--log-alarm handler,"some text that triggers alarm":"text that ends
alarm log collection"

Once alarm subsystem notices first pattern in logs it starts
collecting log messages until it finds second pattern (or some log
lines count limit). Once all log messages are collected than we
trigger alarm handler.

-- 
Łukasz Mierzwa
_______________________________________________
uWSGI mailing list
[email protected]
http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi

Reply via email to