Menu
▾
▴
gusdev-gusdev
|
From: Arnaud K. <ax...@sa...> - 2003-10-01 16:18:41
|
Hi Jonathan We've found an error with the boolean queries. It happens when a query goes over timeout. The statement is reclaimed (see code in reclaimLongRunningStatements() method of the Connection class) and the cache table is not created. But this cache table is expected to exist and we get a "table or view does not exist" exception. So every time someone is trying to do the same query again the code tries to query the cache table and as a consequence, it doesn't return any result. I don't know if I've made myself clear but I can give you the errput if you want. Do you know how to fix this ? Would it be possible to insert a rollback statement in the code when the statement is reclaimed ? cheers Arnaud |
|
From: Steve F. <st...@pc...> - 2003-10-01 16:24:08
|
Arnaud- Jonathan has handed responsibility for the WDK over to me and Angel, since he has moved to D.C. I'll look into it. Feel free to delve into the code and find out whatever you can. Steve Arnaud Kerhornou wrote: >Hi Jonathan > >We've found an error with the boolean queries. >It happens when a query goes over timeout. The statement is reclaimed >(see code in reclaimLongRunningStatements() method of the Connection >class) and the cache table is not created. >But this cache table is expected to exist and we get a "table or view >does not exist" exception. >So every time someone is trying to do the same query again the code >tries to query the cache table and as a consequence, it doesn't return >any result. > >I don't know if I've made myself clear but I can give you the errput if >you want. > >Do you know how to fix this ? Would it be possible to insert a rollback >statement in the code when the statement is reclaimed ? > >cheers >Arnaud > > > >------------------------------------------------------- >This sf.net email is sponsored by:ThinkGeek >Welcome to geek heaven. >http://thinkgeek.com/sf >_______________________________________________ >Gusdev-gusdev mailing list >Gus...@li... >https://lists.sourceforge.net/lists/listinfo/gusdev-gusdev > > |
|
From: Adrian R. T. <ar...@sa...> - 2003-10-01 17:08:59
|
> Arnaud- > > Jonathan has handed responsibility for the WDK over to me and Angel, > since he has moved to D.C. > > I'll look into it. Feel free to delve into the code and find out > whatever you can. > > Steve Does anybody know how reliable Statement.setQueryTimeout is on Oracle? If it's robust we could easily switch to that. Then we get the timeout report in (more or less) the right bit of the code to handle it. Adrian > > Arnaud Kerhornou wrote: > > >Hi Jonathan > > > >We've found an error with the boolean queries. > >It happens when a query goes over timeout. The statement is reclaimed > >(see code in reclaimLongRunningStatements() method of the Connection > >class) and the cache table is not created. > >But this cache table is expected to exist and we get a "table or view > >does not exist" exception. > >So every time someone is trying to do the same query again the code > >tries to query the cache table and as a consequence, it doesn't return > >any result. > > > >I don't know if I've made myself clear but I can give you the errput if > >you want. > > > >Do you know how to fix this ? Would it be possible to insert a rollback > >statement in the code when the statement is reclaimed ? > > > >cheers > >Arnaud > > |
|
From: Pjm <pj...@sa...> - 2003-10-01 17:28:04
|
Adrian Roy Tivey wrote: >>Arnaud- >> >>Jonathan has handed responsibility for the WDK over to me and Angel, >>since he has moved to D.C. >> >>I'll look into it. Feel free to delve into the code and find out >>whatever you can. >> >>Steve > > > Does anybody know how reliable Statement.setQueryTimeout is on Oracle? If it's > robust we could easily switch to that. Then we get the timeout report in (more > or less) the right bit of the code to handle it. The statement timeout code works fine. Its more the fact that if one of the sub queries fails (for whatever reason) then the cacheXXXX table does not get created. Thus the query that intersects with it will also fail. The code does not handle this and the results page has a table with no rows. The next time that query runs it will look for the cacheXXXX table and fail. The user will think the query found nothing. Paul. |
|
From: Steve F. <st...@pc...> - 2003-10-01 18:00:38
|
if i understand paul correctly, then the problem is this: 1. query 1 produces SQLIdResultTable_1 2. query 2 times out, leaving as an artifact a broken SQLIdResultTable, ie one without a valid cached result. 3. the boolean intersect operator tries to intersect these and bombs out, producing an empty result? 4. is it also the case that if the user tries to redo query 2, outside of the boolean context, then it again produces no result because it is trying to reference the non-existent cache table? steve Pjm wrote: > Adrian Roy Tivey wrote: > >>> Arnaud- >>> >>> Jonathan has handed responsibility for the WDK over to me and Angel, >>> since he has moved to D.C. >>> >>> I'll look into it. Feel free to delve into the code and find out >>> whatever you can. >>> >>> Steve >> >> >> >> Does anybody know how reliable Statement.setQueryTimeout is on >> Oracle? If it's >> robust we could easily switch to that. Then we get the timeout report >> in (more >> or less) the right bit of the code to handle it. > > > The statement timeout code works fine. Its more the fact that if one > of the sub queries fails (for whatever reason) then the cacheXXXX > table does not get created. Thus the query that intersects with it > will also fail. The code does not handle this and the results page has > a table with no rows. > > The next time that query runs it will look for the cacheXXXX table and > fail. The user will think the query found nothing. > > Paul. > |
|
From: Arnaud K. <ax...@sa...> - 2003-10-01 21:17:14
|
That's all correct. I haven't investigated further but I reckon the reference to the non-existent cache table in an entry in the queries table (correct me if I'm wrong). So we need a way to delete this entry or roll it back if it is possible. Quoting Steve Fischer <st...@pc...>: > if i understand paul correctly, then the problem is this: > 1. query 1 produces SQLIdResultTable_1 > 2. query 2 times out, leaving as an artifact a broken > SQLIdResultTable, ie one without a valid cached result. > 3. the boolean intersect operator tries to intersect these and bombs > out, producing an empty result? > 4. is it also the case that if the user tries to redo query 2, outside > of the boolean context, then it again produces no result because it is > trying to reference the non-existent cache table? > > steve > > > Pjm wrote: > > > Adrian Roy Tivey wrote: > > > >>> Arnaud- > >>> > >>> Jonathan has handed responsibility for the WDK over to me and Angel, > >>> since he has moved to D.C. > >>> > >>> I'll look into it. Feel free to delve into the code and find out > >>> whatever you can. > >>> > >>> Steve > >> > >> > >> > >> Does anybody know how reliable Statement.setQueryTimeout is on > >> Oracle? If it's > >> robust we could easily switch to that. Then we get the timeout report > >> in (more > >> or less) the right bit of the code to handle it. > > > > > > The statement timeout code works fine. Its more the fact that if one > > of the sub queries fails (for whatever reason) then the cacheXXXX > > table does not get created. Thus the query that intersects with it > > will also fail. The code does not handle this and the results page has > > a table with no rows. > > > > The next time that query runs it will look for the cacheXXXX table and > > fail. The user will think the query found nothing. > > > > Paul. > > > |
|
From: Steve F. <st...@pc...> - 2003-10-02 01:21:46
|
ok. then just to clarify a dead horse, the Boolean query error is a secondary effect. the primary problem, as i think you guys have been exploring, is that we are not handling timed out queries correctly. in particular, as arnaud speculates, we are getting an entry in Queries which points to a non-existent cache table. steve Arnaud Kerhornou wrote: >That's all correct. > >I haven't investigated further but I reckon the reference to the non-existent >cache table in an entry in the queries table (correct me if I'm wrong). So we >need a way to delete this entry or roll it back if it is possible. > >Quoting Steve Fischer <st...@pc...>: > > > >>if i understand paul correctly, then the problem is this: >> 1. query 1 produces SQLIdResultTable_1 >> 2. query 2 times out, leaving as an artifact a broken >>SQLIdResultTable, ie one without a valid cached result. >> 3. the boolean intersect operator tries to intersect these and bombs >>out, producing an empty result? >> 4. is it also the case that if the user tries to redo query 2, outside >>of the boolean context, then it again produces no result because it is >>trying to reference the non-existent cache table? >> >>steve >> >> >>Pjm wrote: >> >> >> >>>Adrian Roy Tivey wrote: >>> >>> >>> >>>>>Arnaud- >>>>> >>>>>Jonathan has handed responsibility for the WDK over to me and Angel, >>>>>since he has moved to D.C. >>>>> >>>>>I'll look into it. Feel free to delve into the code and find out >>>>>whatever you can. >>>>> >>>>>Steve >>>>> >>>>> >>>> >>>>Does anybody know how reliable Statement.setQueryTimeout is on >>>>Oracle? If it's >>>>robust we could easily switch to that. Then we get the timeout report >>>>in (more >>>>or less) the right bit of the code to handle it. >>>> >>>> >>>The statement timeout code works fine. Its more the fact that if one >>>of the sub queries fails (for whatever reason) then the cacheXXXX >>>table does not get created. Thus the query that intersects with it >>>will also fail. The code does not handle this and the results page has >>>a table with no rows. >>> >>>The next time that query runs it will look for the cacheXXXX table and >>>fail. The user will think the query found nothing. >>> >>>Paul. >>> >>> >>> > > > >------------------------------------------------------- >This sf.net email is sponsored by:ThinkGeek >Welcome to geek heaven. >http://thinkgeek.com/sf >_______________________________________________ >Gusdev-gusdev mailing list >Gus...@li... >https://lists.sourceforge.net/lists/listinfo/gusdev-gusdev > > |
|
From: Arnaud K. <ax...@sa...> - 2003-10-02 13:17:24
|
Hi
I think we found a way to fix this problem.
We've added
this line
=> stmt.setQueryTimeout (this.maxQueryTimeout);
in the getStatement() method in ConnectionPool class.
As a consequence the inner class Reaper in ConnectionPool is not in use
any longer.
cheers
Arnaud
public Statement getStatement() {
/ / ...
stmt.setQueryTimeout (this.maxQueryTimeout);
// ...
} catch (SQLException sqle) {
// ...
}
Steve Fischer wrote:
> ok. then just to clarify a dead horse, the Boolean query error is a
> secondary effect. the primary problem, as i think you guys have been
> exploring, is that we are not handling timed out queries correctly.
> in particular, as arnaud speculates, we are getting an entry in
> Queries which points to a non-existent cache table.
>
> steve
>
> Arnaud Kerhornou wrote:
>
>> That's all correct.
>>
>> I haven't investigated further but I reckon the reference to the
>> non-existent
>> cache table in an entry in the queries table (correct me if I'm
>> wrong). So we
>> need a way to delete this entry or roll it back if it is possible.
>>
>> Quoting Steve Fischer <st...@pc...>:
>>
>>
>> > if i understand paul correctly, then the problem is this:
>> > 1. query 1 produces SQLIdResultTable_1
>> > 2. query 2 times out, leaving as an artifact a broken
>> > SQLIdResultTable, ie one without a valid cached result.
>> > 3. the boolean intersect operator tries to intersect these and
>> > bombs
>> > out, producing an empty result?
>> > 4. is it also the case that if the user tries to redo query 2,
>> > outside
>> > of the boolean context, then it again produces no result because it
>> > is
>> > trying to reference the non-existent cache table?
>> >
>> > steve
>> >
>> >
>> > Pjm wrote:
>> >
>> >
>> >> Adrian Roy Tivey wrote:
>> >>
>> >>
>> >> >> Arnaud-
>> >> >>
>> >> >> Jonathan has handed responsibility for the WDK over to me and
>> >> >> Angel,
>> >> >> since he has moved to D.C.
>> >> >>
>> >> >> I'll look into it. Feel free to delve into the code and find
>> >> >> out
>> >> >> whatever you can.
>> >> >>
>> >> >> Steve
>> >> >>
>> >> > Does anybody know how reliable Statement.setQueryTimeout is on
>> >> > Oracle? If it's
>> >> > robust we could easily switch to that. Then we get the timeout
>> >> > report
>> >> > in (more
>> >> > or less) the right bit of the code to handle it.
>> >> >
>> >> The statement timeout code works fine. Its more the fact that if
>> >> one
>> >> of the sub queries fails (for whatever reason) then the
>> >> cacheXXXX
>> >> table does not get created. Thus the query that intersects with
>> >> it
>> >> will also fail. The code does not handle this and the results
>> >> page has
>> >> a table with no rows.
>> >>
>> >> The next time that query runs it will look for the cacheXXXX
>> >> table and
>> >> fail. The user will think the query found nothing.
>> >>
>> >> Paul.
>> >>
>> >>
|
|
From: Steve F. <st...@pc...> - 2003-10-01 16:41:23
|
Arnaud- let me try to better understand what you mean. your subject line is 'boolean query error'. is this problem limited to boolean queries? i don't know exactly how they work, but i would guess that each query is run individually and then the set operations are applied to the results which reside in cache tables. do you know if that is true? if so, then the problem resolves down to how we handle timed out queries, ie, that we should expunge them from the table that points to the cached results. steve Arnaud Kerhornou wrote: >Hi Jonathan > >We've found an error with the boolean queries. >It happens when a query goes over timeout. The statement is reclaimed >(see code in reclaimLongRunningStatements() method of the Connection >class) and the cache table is not created. >But this cache table is expected to exist and we get a "table or view >does not exist" exception. >So every time someone is trying to do the same query again the code >tries to query the cache table and as a consequence, it doesn't return >any result. > >I don't know if I've made myself clear but I can give you the errput if >you want. > >Do you know how to fix this ? Would it be possible to insert a rollback >statement in the code when the statement is reclaimed ? > >cheers >Arnaud > > > >------------------------------------------------------- >This sf.net email is sponsored by:ThinkGeek >Welcome to geek heaven. >http://thinkgeek.com/sf >_______________________________________________ >Gusdev-gusdev mailing list >Gus...@li... >https://lists.sourceforge.net/lists/listinfo/gusdev-gusdev > > |
×
Want the latest updates on software, tech news, and AI?
Get latest updates about software, tech news, and AI from SourceForge directly in your inbox once a month.
Thanks for helping keep SourceForge clean.
X