We've encountered a problem with the generate_statistics configuration in production.
The ConcurrentStatisticsImpl uses a ConcurrentHashMap that saves the statistics for all the HQL queries that are executed. However from our application we generate a lot of unique query HQL queries because we use the JPA query builder, and something like this:
{ { code:java}
public void query(int id, int userId) {
// ...
CriteriaBuilder builder = session.getCriteriaBuilder();
CriteriaQuery<InboxMessage> query = builder.createQuery(InboxMessage.class);
Root<InboxMessage> root = query.from(InboxMessage.class);
query.where(builder.and(
builder.equal(root.get(InboxMessage_.id), id)),
builder.equal(root.get(InboxMessage_.inboxUser).get("id"), userId));
// ...
}
{code } }
creates a new query string for each new input.
This is problematic because the map only grows, and after a couple of days our application consumes too much memory.
It seems possible to clear the stats periodically, however:
- the default behavior 'leaks' memory
- it's not documented
So perhaps it'd be better to have a default {{max_size}} option or something to limit the size of the queryStatistics. |
|
