On 14.12.2010 03:02, Steve Ebersole wrote: Thank you. That much I figured by now ;-) However, this does not explain (for me) why a) Why they are used at all in the situation I described - are they used for all bulk update/delete HQL operations regardless of context/mapping strategy? b) Why the update in my situation runs for more than an hour. That's an absolute disaster because all other transaction fail with lock timeout. -- Marcel Stör, http://www.frightanic.com Couchsurfing: http://www.couchsurfing.com/people/marcelstoer O< ascii ribbon campaign - stop html mail - www.asciiribbon.org

On 14.12.10 11:11, Emmanuel Bernard wrote: Thank you very much - both articles are very helpful, indeed! I claim to understand what Steve wrote BUT I fail to see how these concepts are applicable in my case. My inheritance hierarchy is as follows and the strategy is 'joined': Product / \ Offer Article So, in contrast to the example in the blog when I update a field of Article there's no need to update the Product table (it's not even a versioned update). For me the question remains: why does Hibernate resort to using something like 'update schema.article set flag=0 where (id) IN (select id from HT_article)'? Be it as it may, besides taking veeery long to complete (>1h for 130.000 articles) I noticed that the above statement causes the Java process to consume close to 100% CPU (i.e. 100% of a single core in multi-core environments). Similarly slow processing appears to affect all in-queries. With the help of a profiler singled out o.h.e.query.ParameterParser#parse(String, Recognizer) as being the culprit. The method is awfully long but it essentially iterates over the query string char-by-char and does some processing. In doing so it calls o.h.u.StringHelper#firstIndexOfChar(String, String, int) which in turn calls String.indexOf(int, int). When you have an in-query with a few hundred or thousand values the actual query string might have a million characters depending on how long your parameter name is. So, String.indexOf(int, int) is called maaany times and String processing is slow. Back to the original case...If Hibernate processes '...where (id) IN (select id from HT_article)' for 130.000 articles the query string might be a few million characters long... -- Marcel Stör, http://www.frightanic.com Couchsurfing: http://www.couchsurfing.com/people/marcelstoer O< ascii ribbon campaign - stop html mail - www.asciiribbon.org