grouper-dev - RE: [grouper-dev] grouper 1.4 performance
Subject: Grouper Developers Forum
List archive
- From: "GW Brown, Information Systems and Computing" <>
- To: Chris Hyzer <>,
- Subject: RE: [grouper-dev] grouper 1.4 performance
- Date: Wed, 08 Apr 2009 13:50:37 +0100
--On 07 April 2009 12:33 -0400 Chris Hyzer
<>
wrote:
I took a pass at implementing the sorting/paging for listing large groupsI've checked in two changes to MembershipFinder - the methods which page and return the 'count' of results were all returning the number of immediate memberships - even when displaying effective/all memberships.
(e.g. when assigning privileges or browsing memberships).
When I tried to sort by subjectId and page with mysql, it was slow (3
minutes for a 40k group). But if I page without sorting, it is very
fast. I wonder how oracle would do. Anyways, the concept of the sort
limit will still be needed (though maybe up to 10k once we get everything
working in 1.5).
I made a tag to where we were before, since this was a lot of changes:
GROUPER_1_4_20090407
Gary, if you are still here, can you review the changes and test the UI?
Or anyone can test the API/UI if you don't mind so we can make sure the
1.4 branch is still stable. I shouldn't have affected existing API
methods, though I will run the unit tests...
There is another issue which is best dealt with in the API - filtering results by source. If you have a group with members from different sources (and there are < 500 results - members.filter.limit) the UI offers the option of only showing results from one source (it does 'all' by default). Currently the code below is used:
if(!isEmpty(selectedSource) && sources.get(selectedSource)!=null && membershipCount<membersFilterLimit) {
Iterator it = uniqueMemberships.iterator();
Membership mShip=null;
while(it.hasNext()) {
mShip=(Membership)it.next();
if(!mShip.getMember().getSubjectSourceId().equals(selectedSource)) {
it.remove();
}
}
}
If the paging methods were overloaded with ones which take a Source as an argument they could do the filtering so the count returned would reflect the number of members from the particular source.
I've only had a quick look but didn't see anything else.
Attached is a (not very interesting) screenshot.
Here is an example of the API. Here is getting the resultSize of an
operation (immediate members), without listing all the members:
QueryOptions queryOptions = new
QueryOptions().retrieveCount(true).retrieveResults(false);
group.getImmediateMembers(field, queryOptions);
int totalSize = queryOptions.getCount().intValue();
Here is a paged query:
QueryPaging queryPaging = new QueryPaging();
queryPaging.setPageSize(pageSize);
queryPaging.setFirstIndexOnPage(start);
//.sortAsc("m.subjectIdDb") this kills performance
queryOptions = new QueryOptions().paging(queryPaging);
Set<Member> members = group.getImmediateMembers(field,
queryOptions);
Here is the HibernateSession underlying API (just pass the queryOptions
along):
return HibernateSession.byHqlStatic().options(queryOptions)
.createQuery(
query)
.setCacheable(false)
.setCacheRegion(KLASS + ".FindAllMembersByOwnerAndField")
.setString( "owner", ownerUUID )
.setString( "fieldId", f.getUuid() )
.setString( "type", type )
.listSet(Member.class);
For sorting, you just add that to the query options, but you have to know
the hibernate hql sort field [we should document these in the queries and
not change :) ]
queryOptions = new
QueryOptions().paging(queryPaging).sortAsc("m.subjectIdDb");
Set<Member> members = group.getImmediateMembers(field,
queryOptions);
Let me know how it does,
Thanks,
Chris
-----Original Message-----
From: Chris Hyzer
Sent: Monday, March 30, 2009 10:51 AM
To: 'GW Brown, Information Systems and Computing'; grouper-
Subject: RE: [grouper-dev] grouper 1.4 performance
> This can work - the UI has a notion of paging in some areas - but
> should still only kick in when we are over the configured limit for
> sorting. I look forward to having sorting through the API in 1.5, but
> sorting is useful in most cases so we shouldn't lose it in 1.4.
Agreed. If the display limit is 200, and the sort limit is 50, the
easiest thing is probably just to get the first page of 200. Then if
less than 200, sort it, and show 50. If greater than 200, just show
50. Alternatively, we can get the full count, and based on that either
get up to 200 and sort, or just 50. They are probably the same
performance, so whatever is easiest to code in 1.4
>
> There is an issue with displaying membership lists which include
> effective memberships. The UI processes the list of memberships so it
> can figure out how many 'routes' for a membership there are -
> see<https://bugs.internet2.edu/jira/browse/GRP-55>
No problem. I think once we get the members, we can get a list of
counts for them in one, or we can get all memberships for only the 50
members we need in one query. We can use an "in" clause in 1 query,
will be easy and good performance. Again, if you are already working
with memberships, and that is easiest for 1.4, would be fine for
performance.
>
> Group.getAdmins etc could also be paged - in case anyone adds privs
to
> a large group rather than GrouperAll
Sure. And we can do the AccessAdapter thing where the implementor
doesn't have to implement paging, but if they do, it would be better...
again, in 1.5 if we put more info in the members table, it will be
sorted. In 1.4, it will be seemingly unordered (e.g. by subjectId)
> Depending on the browse mode the UI does different things:
>
> MyMemberships: only groups you are a member of which you can VIEW
> Join groups: only groups where you have OPTIN
> Manage groups: only groups where you have UPDATE or ADMIN
> Explore: only groups where you have at least VIEW privilege
>
> so you would have to join on grouper_fields - or get the ids for the
> fields and do an 'in' clause - does Hibernate have a way of passing a
> collection for that or do you have to build dynamic SQL - or have
separate queries?
We can easily handle this is a well-coded way. I actually picture a
generic method that uses regex where you pass it an HQL, and the
security you want on it (granted, this is the AccessAdapter doing
this), and returns the HQL of the secured query. It would just add
some tables to the from clause, and some constraints on the where
clause. And it can do "in" to pick and choose the privileges. Then
that generic method would be used by a bunch of default implementations
we need (e.g. like the ones above) so it is easy to do easy things. Or
it could be in the query options. E.g. to get all groups of a member
in a certain stem, paged, sorted (in 1.5+), and with only two
privileges UPDATE and ADMIN:
List<Group> groups = member.getGroupsFromParentStem(stem,
new QueryOptions().paging(QueryPaging(1,3,true))
.sorting(QuerySort.asc("name"))
.groupSecurity(GrouperUtil.toSet(AccessPrivilege.UPDATE,
AccessPrivilege.ADMIN));
This would under the covers use the AccessAdapter in a smart and
generic way.
> > ...composite stuff...
> Sounds reasonable.
Good
> (from tom)
> Assuming that users don't typically want to look into the reasons why
> an effective membership is there, can we optimize things so that that
data is retrieved only when the user indicates they want to know why?
Again, for 1.4 I think the least changes the better especially when I
think we can easily do this efficiently (see above). But if Gary wants
to implement this suggestion, it sounds like a good idea.
> If we are paging I still think we need methods to return the
> membership count so we can show text such as 'Showing 51-98 of 98
items ', otherwise a user has no idea how many pages there might be.
My HibernateSession paging API exposes all this information. I have a
paging custom tag I could put into Grouper which has a readonly mode
and a button mode (both are shown in the attached image, one on top of
the other... page 7 is being displayed). If Gary wants to setup the
action(s) [if the paging bean is in session, I believe I just need to
pass in the page number and action from custom tag, and it will work] I
can make the custom tag work. Gary, is that something you want for 1.4
or 1.5?
Thanks!
Chris
----------------------
GW Brown, Information Systems and Computing
- RE: [grouper-dev] grouper 1.4 performance, Chris Hyzer, 04/07/2009
- RE: [grouper-dev] grouper 1.4 performance, GW Brown, Information Systems and Computing, 04/08/2009
Archive powered by MHonArc 2.6.16.