[
https://issues.apache.org/jira/browse/PDFBOX-845?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12916554#action_12916554
]
Larry West commented on PDFBOX-845:
-----------------------------------
(Responding to Adam):
My apologies, I worded that badly. The same tests have run *serially*
hundreds of times without problems on many Linux, Mac, Windows machines. Now
that I parallelized the tests, it is easy to reproduce the hang on all of those
systems. Roughly every 5-10 runs of the (parallelized) test suite it will
hang.
So if it is a JVM issue (still a possibility), it is Sun's 1.6.0_20 on Linux,
Windows, and OSX. (I can't bring myself to say "Oracle's JVM").
PS: yes, each of these threads is accessing its own file (in fact, they are all
stuck inside the static PDDocument.load(File) method).
I'll try to get set up with 1.5 in the next week.
Sure wouldn't hurt performance to switch to StringBuilder, so I just might do
that, time permitting.
> Lockup in PDDocument.load() --> PDFParser.parseObject() with 6 threads
> ----------------------------------------------------------------------
>
> Key: PDFBOX-845
> URL: https://issues.apache.org/jira/browse/PDFBOX-845
> Project: PDFBox
> Issue Type: Bug
> Components: Parsing
> Affects Versions: 1.2.1
> Environment: Linux lxdev01 2.6.32-24-generic #42-Ubuntu SMP Fri Aug
> 20 14:24:04 UTC 2010 i686 GNU/Linux
> java version "1.6.0_20"
> Java(TM) SE Runtime Environment (build 1.6.0_20-b02)
> Java HotSpot(TM) Server VM (build 16.3-b01, mixed mode)
> testng 5.11
> running under maven-surefire plugin v2.5 with parallel=methods and
> threadCount=6
> Reporter: Larry West
> Priority: Critical
> Attachments: pddocument-load-lockup.stack
>
> Original Estimate: 32h
> Remaining Estimate: 32h
>
> This is a TestNG unit test suite, with each test loading a different PDF via
> PDDocument.load(). I just switched TestNG to parallel=methods (had been
> serial) and it locked up first time. "jstack -l" output will be attached,
> but I'm putting the pdfbox portions here (just below).
> In looking at the code, it's not clear what's being waited on, except that
> four of the threads are stuck in BaseParser.parseDirObject(), apparently
> waiting on the (synchronized) toString() method of a [local] StringBuffer.
> (StringBuffer is used in BaseParser.java where a StringBuilder is clearly
> preferable -- this is probably true for every local instance of a
> StringBuffer.)
> I don't know why that would cause the thread to sit waiting, though (JVM
> problem? 1.6.0_20 on Linux), and the other two threads appear to be waiting
> on COSObjectKey.getNumber() [perhaps], and I see no synchronized objects or
> methods there.
> Finding the cause of the lockup would be preferable, but replacing
> StringBuffer with StringBuilder whereever they are used locally (possibly
> including any private non-static members) would be an improvement in
> performance if nothing else.
> Threads 1, 2, 4, & 6:
> at
> org.apache.pdfbox.pdfparser.BaseParser.parseDirObject(BaseParser.java:1013)
> at
> org.apache.pdfbox.pdfparser.BaseParser.parseCOSDictionaryValue(BaseParser.java:157)
> at
> org.apache.pdfbox.pdfparser.BaseParser.parseCOSDictionary(BaseParser.java:233)
> at
> org.apache.pdfbox.pdfparser.BaseParser.parseDirObject(BaseParser.java:929)
> at org.apache.pdfbox.pdfparser.PDFParser.parseObject(PDFParser.java:519)
> at org.apache.pdfbox.pdfparser.PDFParser.parse(PDFParser.java:179)
> Threads 3 & 5:
> at
> org.apache.pdfbox.cos.COSDocument.getObjectFromPool(COSDocument.java:481)
> at org.apache.pdfbox.pdfparser.PDFParser.parseObject(PDFParser.java:540)
> at org.apache.pdfbox.pdfparser.PDFParser.parse(PDFParser.java:179)
> I should add: these same tests, using the same files, have run literally
> hundreds of times on various machines (Windows, Mac, Linux) using PDFBox
> 1.2.1 without any lockup.
> Clearly I can back off parallelizing my tests for now, but there is no
> obvious reason why PDDocument.load() can't be called in parallel, and so it
> concerns me that this will be a real problem.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
