I've been trying to resolve this problem for the past 2 weeks.
I've come to the conclusion, maybe erroneous, that some of our
jpgs can't be converted to thumbnails by thumbnail.createGraphics.
The same thing happens on 2 different machines.
linux kernel 2.4.27, jdk1.5.0.
Our assetstore contains some 1800 items and 54 GB.
There are a bunch of images (probably jpg) that when processed cause
jvm memory usage to hit ~200MB. Then the process just sits there.
I've let it sit for >12 hours.
When I cancel the job none of the processed bitstreams seem to
have been recorded in the database. The next time I run filter-media
all images are processed again. Also 'select count(bitstream_id) from
bitstream;' increases each time - lots of rows with null names. So,
while debugging I've been dropping and restoring dspace to get fresh starts.
I've been trying to work around this so that I can at least get some
thumbnails to show up on our test site (our dev site is working but
has only 661 items and 75MB of data). I copied our production dspace
to our test machine to test the real thing before I upgrade the
I go into MediaFilterManager and exclude the files that fail.
Then, I run filter-media until it fails and then exclude more files.
This is a slow way to go because each time all of the files are
processed again. I may write a wrapper that will allow me to process
a single image and put it in a shell script to identify the bad ones.
At first java is using about 60% of the CPU and postmaster ~40%.
As the process progresses java usage grows until postmaster is very quiet.
I made a few changes to MediaFilterManager to free up memory because
the jvm memory usage grows from ~40MB to 190+MB when things are working.
But this doesn't fix the main problem. When a bad file is hit memory goes
to ~200MB. If I start off processing a bad file memory usage immediately
hits ~200MB and then the process will sit there until I kill it.
I can send a few files that are causing the problem if you'd like.
I track them down by getting the item_id, finding it's bundle_id and
then it's bitstream_id and name. I can view those images but they don't
want to be thumbnails. The size of an image doesn't appear to be a factor.
Thanks, Joe Edwards