Piyush Khandelwal created FOP-2937:
--------------------------------------

             Summary: Post PDF generation, Soft reference of PDFObject in 
PDFReference are not immediately garbage collected leading to excessive memory 
usage.
                 Key: FOP-2937
                 URL: https://issues.apache.org/jira/browse/FOP-2937
             Project: FOP
          Issue Type: Improvement
    Affects Versions: 2.4, 2.3
            Reporter: Piyush Khandelwal


PDFReference object holds a SoftReference of PDFObject (PDFPage, PDFLabel, 
PDFName etc.).
If we generate a huge PDF ; *I tried with a PDF having around 150 thousand 
pages with 12 GB of RAM;* lots of these references linger around waiting for 
the garbage collector to collect them. 
But GC wont collect them as long as JVM is able to recover enough memory 
without throwing out of memory.

Here are few metadata from my testing for further understanding of the issue - 
Stats for generating 1 PDF - 
*FO size:* 2.03GB
*Generated PDF No. of Pages:* Around 150 K
RAM: 12 GB
Peak memory that reached while generation - 11.3GB
Residual memory after forced GC: 9 GB

The FO mainly contains tabular data with each pages sequence having max of 500 
rows.

On analyzing the memory dump; found lots of reference for PDFPage, PDFName etc.

*Question - * Is there any specific reason for using SoftReference in 
PDFReference class  instead of WeakReference.

Testing by changing SoftReference  to WeakReference in PDFReference shows 
following improvements without any issue in the generation whatsoever - 

Stats for Generating 5 PDF in parallel - 
*FO size:* 2.03GB
*Generated PDF No. of Pages:* Around 150 K
RAM: 12 GB
Peak memory that reached while generation - 4GB
Residual memory after forced GC: 300 MB

So, by changing SoftReference to WeakReference, I was able to generate 5 PDF 
having 150K pages in parallel with max  4GB Ram; without any generation issues.

*Question- * Is there any specific reason for using SoftReference in 
PDFReference class  instead of WeakReference?

You can clearly see the performance benefits of changing to WeakReference. 
But as I dont understand the complete internal details of FOP works, I would 
like to understand  if we can target this change and if not what is the reason 
of using SoftReference?





--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to