Re: [WebMacro-user] Tomcat can't unload webmacro... logging issue?

[<Web...@St...>]

On Mon, 19 Jul 2004, Sebastian Kanthak wrote:

| On Monday 19 July 2004 18:39, Endre St=F8lsvik wrote:
| > How about getting rid of the Threads, ALL of them?! BE GONE!!!
|
| hey, calm down :)

;)

|
| > I mean, WebMacro is a -library-. If I want it to use any cache manage=
r
| > that use threads, I'll start those myself, and kill them afterwards t=
oo.
| > Although I cannot understand what the good point about this would be:=
 why
| > not use a WeakHashMap or other weak-style references for the cache? T=
he
| > default config and impl of the -library- WebMacro is, should most
| > definately be -lean- and -mean-, not fire up any unhandled -threads-!=
!
| >   ( Not to mention that thread that updates a "time" long, I think it=
 was
| > for logging: that must be the most .. useless .. piece of "performanc=
e
| > optimization" ever devised: in a production environment, how many
| > log-lines are one to expect from WebMacro? The potential synchronizat=
ion
| > costs must be thousand-folds worse than the millis-getting, -if- ther=
e was
| > an impact in any way.. )
|
| well. As you've perhaps noticed, Brian removed the clock thread in 2.0b=
1, so
| no need to worry about it anymore. The only thread remaining is the one
| responsible for removing templates from the cache after a given period =
of
| time and I'm afraid there is no other possibility than to do this with =
a
| thread.

MAJOR POINT: Why do you have to remove the cached templates?

See, this is my entire question. WHY do we have a Thread that removes the
templates, this in particular in a production environment?

Memory issues?

If the templates were put into the cache using "soft references" ..

  http://java.sun.com/developer/technicalArticles/ALT/RefObj/index.html
  http://saloon.javaranch.com/cgi-bin/ubb/ultimatebb.cgi?ubb=3Dget_topic&=
f=3D15&t=3D000179

.. then this would be the best for a production environment, by far. And
you'd not need the Thread. The -even better- thing would be to have som
"call-in" method into the Broker (or the thingy holding the cache) that I
could invoke that says "clear all caches, please, completely".
  We wouldn't need the Thread!!

The Thread is good in development setting, due to how the macro(-files)
apparently are implemented. This because a macro-file won't be
"reload-checked" at all, and thus will be stale forever - so I set the
reload-time to 3 seconds while developing. But I'd prefer to switch cache=
s
on deployment, and use a Soft or Weak reference cache, that just loaded
and loaded and loaded and used up -all- memory.

See the comment #3 by "Member #862" in the above-mentioned javaranch-link=
:
"free memory is _bad_". The logic is that if you have 2 GBs of RAM (which
isn't that uncommon), then why have to load -any- templates of disk? Why
not just have them precompiled and ready to go -always-? The memory -will=
-
get freed when there isn't enough memory for the working set of normal
activities. And

Reading this:
  http://www.axlrosen.net/stuff/softreferences.html
I realize that what I want is a -soft- reference cache, not a weak. Weak
is .. weaker .. than soft! So soft-cache it is.

|
| BTW, the thread is started by a class in the concurrent package, anothe=
r
| _library_... ;)

Well, another bad library! ;)

Really, I don't like libraries that fire up Threads w/o me asking in
particular for it. A program with millions of Threads, where each have on=
e
specific stupid little chore.
  OT, There should be some "cron"- and dispatch-based "TaskRunning" syste=
m
that was embedded "in the JVM" (read: standard java), in addition to a
event-queue that one could inject events into, and subscribe listeners to=
.
(I have made these features as services within the portal-server our
company makes, and developers actually use it and like it a lot.)

And, btw, I do know that the concurrent package is very nice, but firing
up Threads here and there and everywhere is really bad in my opinion,
anywhichway - one immediately get such problems as we now face.

|
| However, one could add logic to only start this thread, if the function=
ality
| to remove templates from the cache after some time is actually used (e.=
g.
| ExpireTime is set to a positive value).

YES! YES! YES!!!

Ehh.. I'd love to see this functionality implemented.. ;)
(As a first step)

To solve my tomcat-reload-problem (if it really is WebMacro that doesn't
clean its Threads on reload/redeploy) in development sceniarios, I could
actually just use NO cache, if it is impossible to get the macro-thingys
to "reload-check" also.

|
| > So WebMacro isn't cleaning up after itself? I've always had a problem=
 with
| > my webapplication "leaking memory" when i restart it dozens of times,
| > eating up my 64 MB "developer sized memory" for Tomcat after 10-12
| > restarts. So WebMacro is the one to blame? And it won't get better in=
 the
| > next version either?
|
| 1.1final is stopping all threads and garbage collection should take car=
e of
| the rest, I think, but apparently something does not work (see Mark's
| problem) completely.

I have huge memory-leaking when reloading, as mentioned, but I have reall=
y
no idea to where this might be, but -have- been suspicious towards
WebMacro due to the annoyance-Threads being fired up and .. hangin' out.

|
| > On a tangent: How's the next version coming along?! ;)
|
| well, at least we have some traffic on the list again. :)
|
| Regarding the next version: Do we have some functionality to make globa=
l
| "macro-libraries" available via WebMacro.properties, so that template a=
uthors
| do not need to include them in every template? If not, I'd like to add =
this
| functionality.

I think that 2.0 should be shipped a year ago, approx. So, do not add mor=
e
functionality to it now, just get it out the door. Now, please, now! Add
this kind of functionality to a 2.1 release.

Thanks,
Endre