Fwd: [pylucene-dev] ParallelMultiSearcher and Hits.doc(i)
vajda at osafoundation.org
Wed Apr 20 09:47:52 PDT 2005
The reason this bug was happening was because of a deadlock situation between
python threads who call into java and back into python and threads created
from java (ParallelMultiSearcher does that) that call into python.
This was fixed in 0.9.8 where all calls from python to java release the python
GIL (global interpreter lock) and reacquire it when calling back from java
into python. This is not OS specific.
On Wed, 20 Apr 2005, Yura Smolsky wrote:
> Hello, Andi Vajda.
> AV> Great, this would get us one step closer to a 1.0 release !
> Andi, I have that problem appeared again. It works great on Windows
> platform, and it worked good on RedHat Linux.
> But when we migrated to Debian with kernel 2.6.11 the problem has appeared again.
> Can you suggest us how to detect the real reason of this bug with hits
> on Debian Linux?
> ---------- Forward Message ----------
> From: Yura Smolsky <info at altervision.biz>
> To: pylucene-dev at osafoundation.org
> Time: Mon, 28 Feb 2005 09:28:04 -0600
> Subj: [pylucene-dev] ParallelMultiSearcher and Hits.doc(i)
> Attach: RemoteSearcherBad.py, ?????.txt
> I have noticed that ParallelMultiSearcher does not return correct Hits
> object when we use custom implemented Searchable, but MultiSearcher return correct one.
> "Correct" object Hits object means that it can return documents more
> than 100th.
> So, I cannot get documents more than 100th from Hits object returned
> by ParallelMultiSearcher (with custom implemented Searchable object).
> When It cannot return needed object it loops forever, it does not
> return control outside Hits.doc(i) method.
> I provide attach file which can easily reproduce the situation.
> Yura Smolsky.
> ---------- End of Forward ----------
> Yura Smolsky,
More information about the pylucene-dev