On Wed, 30 Sep 2020 09:51:27 +0100, Ahem A Rivet's Shot wrote:

> On Wed, 30 Sep 2020 07:13:30 -0000 (UTC)
> Martin Gregorie  wrote:
>
>> I much prefer using a tracing system, i.e. I put trace events in my
>> code that are enabled from the command line, because often the
>> triggering event is a few steps back along the timeline from the code
>> that crashed.
>> A trace log shows you that clearly while a stack dump often can't.
>
>  The last time I used Java heavily for work I was seriously
> considering using aspect weaving to hook a call trace log into the code
> dumping into a circular buffer.

Been there, done that as well. It words equally well in COBOL, C and Java.

FWIW the first time I saw a circular buffer in use was when I was sysadmin
for George 3, the ICL mainframe OS. Its design philosophy was to throw a
GEOERR, dump diagnostics and quit if it hit a major problem. It used two
circular buffers: one very fine-grained to show exactly what happened,
the other coarse enough to show how the system had arrived at the GEOERR.
IOW the fine-grained buffer covered the last 3-4 entries in the coarse
buffer. Made trouble-shooting a breeze, so I've tended to leave switchable
tracing statements in live systems ever since. Its paid off: diagnosing
faults in production code becomes much easier. I make using a circular
buffer configurable too, mainly because I've found its most useful for
diagnosing rare errors in long-running non-interactive processes without
drowning people in megabytes if trace log.

IME a good approach in interactive processes is to make tracing
controllable by more experienced users. If they have a problem they can
talk to support and, if that doesn't resolve the problem, switch tracing
on, repeat what they did and turn it off again. That keeps everybody
happy: users because they know the support team will be on the case once
they've retrieved the log and support because they've got an accurate
record of what happened and why.


--
Martin    | martin at
Gregorie  | gregorie dot org

--- SoupGate-Win32 v1.05