Hi Sebi,
Sebi Petreanu wrote:
Hello guys,
I read your messages and I'm sorry you couldn't find a solution. I had the
same problems when I tried to index my documents. you can see some of these
errors here:
http://www.nabble.com/Zend_Search_Lucene-errors-tf3205213s16154.html#a8900427.
I had this problem only on WinXP system. I want to make the following
observations:
- When AntiVirus is running the errors appears more often.
- When I place a wait time between consecutive functions calls that
access the index (delete(), find(), addDocument()) the probability to index
all documents is increased. I actually succeeded to index 8000 documents
using a wait time between consecutive functions calls.
Thanks for your comments!
One additional question. Is indexing performed in "exclusive" mode (or
do you have any parallel search/index updating tasks during indexing
your documents)?
Parallel searching/indexing/updating must work correctly, but it
important to find where the bug is.
Alexander, I have a little question regard the search optimizations. Why the
second query so bad search times compared with the first one?
1. +(((titleSrch:arte)) ((descriptionSrch:arte)) ((tagsSrch:arte))) is made
in 0.077 sec
2. +(((titleSrch:arte)) ((descriptionSrch:arte)) ((tagsSrch:arte)))
+(countryID:1) is made in 0.50 sec
The difference is huge. What do you think?
First query is optimized to one multi-term query. Second can't be
reduced to a primitive query:
1. => "titleSrch:arte descriptionSrch:arte tagsSrch:arte"
or
"titleSrch:arte OR descriptionSrch:arte OR tagsSrch:arte"
2. => "+(titleSrch:arte descriptionSrch:arte tagsSrch:arte) +countryID:1"
or
"(titleSrch:arte OR descriptionSrch:arte OR tagsSrch:arte) AND countryID:1"
Other thing which may affect performance is searched term selectivity.
Please check execution time of "countryID:1" query.
PS I have an idea how to improve performance of boolean queries, which
has clauses with large result sets.
But I am not sure I'll have time for this before 0.9
That would be good if you will have possibility to send me your index,
which gives these results, or make an Jira issue to have it in mind.
With best regards,
Alexander Veremyev.
