errors: separate business, integration and bug -- three failures logged three ways
Everything was logged identically: ERROR, titled 'Internal Server Error', and in
web stack-traced twice (log4j2 plus printStackTrace, the second copy landing in
catalina.out). A partner mistyping an IMEI produced the same output as a
NullPointerException.
That makes the error stream unalertable. Measured over six hours across web and
fofo: 909 ERROR lines, of which 294 (32%) were ProfitMandiBusinessException --
HTTP 400s where the user is simply told what to fix. Any rule on ERROR rate
fires constantly, and an error tracker would rank 'insufficient balance' as the
top issue.
business WARN, no stack trace, 4xx -- expected, user-correctable
integration ERROR + dependency name -- ours is fine, theirs is not
anything else ERROR + stack trace, 500 -- a bug
New IntegrationException carries getDependency(), so two hundred failures of one
gateway group as one problem rather than two hundred unrelated traces. That
category did not exist: such failures were previously either a bare Exception
(indistinguishable from our own bug) or a business exception (which wrongly
blames the user).
printStackTrace removed from the web handler -- it was writing a second copy of
every trace to catalina.out.
Prerequisite for wiring the GlitchTip appender, which must not be attached until
ERROR means something.