On 12/11/2013 08:08 PM, Arne Vajh?j wrote:
On 12/11/2013 6:13 PM, Eric Sosman wrote:
On 12/11/2013 5:55 PM, RVic wrote:
Is anyone aware of a public domain package that can email you on an
error, include the stack trace or log, whenever an exception occurs?
I think you'd have to modify the JVM itself to do this. But
you wouldn't want to: If you got an E-mail for every exception
thrown, you'd drown in E-mail.
Presumably, what you really want is E-mail about "important"
exceptions, and the hard part is clarifying what "important" means.
Could you describe your use case in more detail?
In theory exceptions should be for exceptional cases meaning
no drowning in emails.
I have seen many apps where your concerns are valid though.
Arne
It's true that exceptions are for exceptional cases. The Java tutorial
says that:
"An exception is an event, which occurs during the execution of a
program, that disrupts the normal flow of the program's instructions."
However, as you yourself admit, Arne, there are many apps where there
are frequent exceptions. And no small number of these apps are actually
well-designed.
As an example (and this is based on a setup I am familiar with),
consider an ESB app that mediates between 2 systems: A and B. System A
is a source of data for system B, but both are unaware of each other,
and the data must be transformed before it goes into B. The hitch is,
the ESB app has no ability to do more than very simple validation, and
System A has a fair bit of bad data, because of human data entry and
configuration reasons [1]. Hence, System B must validate (because it
knows what it can accept), and this works OK.
The validation errors must come back to the ESB app, since the ESB app
is the controlling intelligence. They come back as SOAP faults (Java
runtime exceptions), which is appropriate.
Furthermore, the ESB app polls System A, and communicates with System B
using SOAP, every few minutes (this interval being configurable but
dictated by business requirements).
So the first few months we were getting a few dozen SOAP faults every
few minutes, so say about 500-1000 every hour, or 10-20 thousand a day.
Thing is, 99 percent of those 10-20 thousand SOAP faults were for the
same few dozen bad data entries in System A. Sending an email for each
of those would have overwhelmed the human recipients. We recognized that
a separate ESB job that ran every 24 hours (and effectively called the
same intermediary logic) would capture all the problems, and it could
send out a *single* email that provided a summary of the bad data in
system A (and why it was bad).
Point being, email notification on errors (as with logging errors to
file etc) are almost always going to a human being [2]. As such, the
urgency of an error dictates whether a notification should be sent at
all, how often an email should be sent, who gets it (human or system),
the nature of the email content (summary or single event), etc etc.
To wit, for a large percentage of apps I think an "email per exception"
strategy will lead to drowning in emails. And you can Splunk or
grep|uniq|cut|cat|paste all that to your hearts's content...essentially
wasting your time to do what the reporting systems could have done in
the first place.