John,
First, thanks for the detailed response! I'll have to try out some of
the utilities that you mention, reproduce the problems, and get back
with the details. But I'd like to say how much I appreciate all of the
advice you've provided. More responses in-line.
John Sisson wrote:
Joe Bohn wrote:
I've been encountering a number of problems building on Windows which
seem to be steadily getting worse and I was wondering if anybody else
is experiencing this (and hopefully has some work-arounds).
- Out of Memory Errors. These seem to be related to two different
things:
1) Long file names. If I remember correctly Windows has issues if
the file names exceed 256 bytes. I've been working with multiple
levels of geronimo concurrently and so have had to name the root
something other than simply geronimo. It seems like if I start to get
longer than 12 characters or so I begin to hit these problems. I
think we may be getting into trouble here with the depth of our
packages and embedded classes.
Yes, it is concerning..
http://mail-archives.apache.org/mod_mbox/geronimo-dev/200601.mbox/[EMAIL PROTECTED]
I also have had to shorten my directory path I am building in. The long
file name issue should go way when we move to JDK 1.5_06 but a number of
other programs on Windows that deal with files also have the issue (e.g.
Windows File Explorer, WinZip, Visual SourceSafe) so this issue could
bite us with users complaining that other tools they are using can't
work with Geronimo's long directory paths.
Do we have a JIRA for this issue?
I'm not aware of a JIRA for this. I'll search again and if I can't find
one I'll create one. Another tool to add to the list is xcopy.
2) Extra garbage hanging around in %temp% from previous builds. The
geronimo build leaves a lot of trash in %temp% and when that seems to
cause problems with out of memory errors and subsequent builds.
See http://issues.apache.org/jira/browse/GERONIMO-777
Hmmm ... I had seen this JIRA and all the discussion but somehow I
always only focused on the config-store part of it and I haven't noticed
that problem recently.
Are you seeing these problems on trunk, 1.0 or 1.1 branches or all?
I'm seeing them on 1.0 and trunk for sure. I haven't been keeping
careful tabs on 1.1 but I'd be surprised if it was any different there.
What JDK are you using?
Sun JDK 1.4.2_08 (build 1.4.2_08-b03)
For a long time I have been using the following (from memory I think
there was something in maven or one of the libraries it uses that had a
leak - not sure if this still exists)
SET MAVEN_OPTS=-Xmx512m -XX:MaxPermSize=128m
What command are you using to build? (this could possibly affect what
code gets invoked and how much memory is used). Do you have MAVEN_OPTS
environment variables set?
No, I don't have MAVEN_OPTS set but I'll certainly give that a try. I
build with many different commands and from many different location
depending upon what I'm doing. This problem seems more pronounced when
I do a full clean build with itests .... "maven m:clean new" from
geronimo root.
I will try to reproduce with the same JDK level and commands once you
provide the info.
- File IO errors when running the CRLF plugin. These may be related to
the %temp% storage but I seem to get them at times even when I've just
cleaned out my %temp%. Running back to back I get different results.
One thing to try is disabling XP's file indexing (which I have done).
It looks like this has caused problems for Java elsewhere (see the
workaround in this bug)..
http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=5006328
I also found that
http://www.sysinternals.com/Utilities/ProcessExplorer.html utility is
good for tracking down processes that have file handles open. Use the
Find-->Find Handle menu option and specify the file name.
I had some similar problems for a while and ended up tracking it down my
incorrectly configured Diskeeper disk defragmenter that had paused
defragmentation and held file handles open.
Have you tried with real-time virus checking disabled to check it isn't
a virus checker causing issues?
What files are you getting the IO errors on? Can you include the error
output?
Thanks for the pointers to utilities. I'll give them a try (including
disabling the real-time virus checker but it seems to happen with too
great a frequency to be this). The FileIO errors are always on files
deep within the config-store when it fails during the CRLF plugin ...
perhaps related to the long file names again.
It makes me wonder if we need to have some kind of retry logic for file
operations if we are possibly competing with file indexers etc so we can
run on a Windows box without problems. I noticed some of the Ant tasks
have retry logic for some file operations:
http://svn.apache.org/viewcvs.cgi/ant/core/trunk/src/main/org/apache/tools/ant/taskdefs/Mkdir.java?view=markup
http://svn.apache.org/viewcvs.cgi/ant/core/trunk/src/main/org/apache/tools/ant/taskdefs/Delete.java?view=markup
Regards,
John
Thanks for the help,
Joe
--
Joe Bohn
joe.bohn at earthlink.net
"He is no fool who gives what he cannot keep, to gain what he cannot
lose." -- Jim Elliot
