I would suggest that in order to even bring up "thread local issues"
in the future that the submitter supplies a pure Java NON-LUCENE test
case that demonstrates the problem (just as you would if reporting a
bug to Sun).
All of the "guessing" about what is going on is counter productive.
You can review the JDK source code, and you can run test cases.
There is nothing inherently broken in ThreadLocals, and people that
keep claiming that are only doing a disservice to the people that
make sure Java is a robust and reliable platform.
On Dec 19, 2006, at 1:41 PM, Yonik Seeley wrote:
On 12/19/06, Otis Gospodnetic <[EMAIL PROTECTED]> wrote:
I _think_ Robert is right and ThreadLocals are not the problem (I
tried getting rid of them, and replacing them with an instance var
last week, but run into problems with multi-threaded unit tests).
If you want to try getting rid of the ThreadLocals, you can't replace
with instance vars because the stuff put into the Threadlocals isn't
thread safe.
Replace the ThreadLocal stuff with the code that populated it in the
first place.
So for instance, in TermInfosReader, replace this
private SegmentTermEnum getEnum() {
SegmentTermEnum termEnum = (SegmentTermEnum)enumerators.get();
if (termEnum == null) {
termEnum = terms();
enumerators.set(termEnum);
}
return termEnum;
}
With this:
private SegmentTermEnum getEnum() {
return terms()
}
What I'm seeing while profiling (and in production) is the
accumulation of these:
org.apache.lucene.search.FieldCacheImpl$Entry
org.apache.lucene.search.FieldCacheImpl$CreationPlaceholder
This is related to http://issues.apache.org/jira/browse/LUCENE-651
(the commit for that patch also happens to coincide with when I
started seeing the leak).
I'll take a look into that. Solr hasn't sync'd to that version of
Lucene yet, so we haven't seen this problem of course :-)
-Yonik
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
