Menu
▾
▴
bodington-developers — Mailing list for Bodington developers
You can subscribe to this list here.
| 2005 |
Jan
|
Feb
|
Mar
|
Apr
(41) |
May
(353) |
Jun
(133) |
Jul
(534) |
Aug
(401) |
Sep
(219) |
Oct
(86) |
Nov
(144) |
Dec
(61) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2006 |
Jan
(200) |
Feb
(130) |
Mar
(345) |
Apr
(153) |
May
(247) |
Jun
(338) |
Jul
(222) |
Aug
(70) |
Sep
(39) |
Oct
(27) |
Nov
(76) |
Dec
(30) |
| 2007 |
Jan
(81) |
Feb
(44) |
Mar
(9) |
Apr
|
May
(3) |
Jun
(2) |
Jul
(34) |
Aug
(2) |
Sep
(1) |
Oct
|
Nov
|
Dec
(6) |
| 2008 |
Jan
|
Feb
|
Mar
(5) |
Apr
|
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
| 2009 |
Jan
(1) |
Feb
|
Mar
|
Apr
|
May
(7) |
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
|
From: Matthew B. <mat...@co...> - 2005-06-13 11:14:26
|
Matthew Buckett wrote: > Currently org.bodington.database.ConnectionPool doesn't cope with a > database restart while Bodington is still running and rather than adding > support for this I though it might make more sense to replace it with a > "standard" open source equivilent. Correction Bodington does seem to survive a database restart. I get a nice stack trace from ConnectionPool when it tries to find any outstanding Jobs but otherwise is seems to work. -- +--Matthew Buckett-----------------------------------------+ | VLE Developer, Learning Technologies Group | | Tel: +44 (0) 1865 283660 http://www.oucs.ox.ac.uk/ | +------------Computing Services, University of Oxford------+ |
|
From: Matthew B. <mat...@co...> - 2005-06-13 11:03:45
|
I was going to have a look at moving from our custom database connection pooling to common-dhcp (http://jakarta.apache.org/commons/dbcp/). Initially this work would be against WebLearn HEAD but I could push it across to Bodington HEAD. Currently org.bodington.database.ConnectionPool doesn't cope with a database restart while Bodington is still running and rather than adding support for this I though it might make more sense to replace it with a "standard" open source equivilent. With the move to commons-dhcp we also get things like custom connection validation. One feature I might have to reimplement (or throw away?) is the restriction of the number of database connections per user which o.b.d.ConnectionPool currently supports. Any comments? -- +--Matthew Buckett-----------------------------------------+ | VLE Developer, Learning Technologies Group | | Tel: +44 (0) 1865 283660 http://www.oucs.ox.ac.uk/ | +------------Computing Services, University of Oxford------+ |
|
From: Colin T. <col...@co...> - 2005-06-10 09:25:54
|
5th, 6th OK Might be away 7th and 8th. Colin ____________________________________ Colin Tatham VLE Team Oxford University Computing Services www.oucs.ox.ac.uk/ltg/vle/ www.bodington.org > -----Original Message----- > From: bod...@li... > [mailto:bod...@li...]On Behalf Of > Jane Jotcham > Sent: 09 June 2005 11:37 > To: bod...@li... > Subject: [Bodington-developers] Next Developers meeting > > > We need to set a date for the next developers' meeting to take place > prior to release 2.6. Can people please advise when they are > available > in July from 5 July onwards. One thought I had was that if > most people > are in Oxford for Buzz on 4 July the meeting could be in > Oxford on the > 5th. However, I have not checked anyone's availability for this so > advice would be appreciated. > > Thanks > Jane > > > > > > ------------------------------------------------------- > This SF.Net email is sponsored by: NEC IT Guy Games. How far > can you shotput > a projector? How fast can you ride your desk chair down the > office luge track? > If you want to score the big prize, get to know the little guy. > Play to win an NEC 61" plasma display: http://www.necitguy.com/?r=20 > _______________________________________________ > Bodington-developers mailing list > Bod...@li... > https://lists.sourceforge.net/lists/listinfo/bodington-developers > |
|
From: Colin T. <col...@co...> - 2005-06-09 11:44:22
|
me Tarzan! > -----Original Message----- > From: bod...@li... > [mailto:bod...@li...]On Behalf Of > Alistair Young > Sent: 09 June 2005 12:23 > To: bod...@li... > Subject: Re: [Bodington-developers] Next Developers meeting > > > After the 11th would be better for me Jane. > > Alistair > > > On 9 Jun 2005, at 11:37, Jane Jotcham wrote: > > > We need to set a date for the next developers' meeting to take > > place prior to release 2.6. Can people please advise when they are > > available in July from 5 July onwards. One thought I had > was that > > if most people are in Oxford for Buzz on 4 July the meeting could > > be in Oxford on the 5th. However, I have not checked anyone's > > availability for this so advice would be appreciated. > > > > Thanks > > Jane > > > > > > > > > > > > ------------------------------------------------------- > > This SF.Net email is sponsored by: NEC IT Guy Games. How far can > > you shotput > > a projector? How fast can you ride your desk chair down the office > > luge track? > > If you want to score the big prize, get to know the little guy. > > Play to win an NEC 61" plasma display: http://www.necitguy.com/?r=20 > > _______________________________________________ > > Bodington-developers mailing list > > Bod...@li... > > https://lists.sourceforge.net/lists/listinfo/bodington-developers > > > > > > ------------------------------------------------------- > This SF.Net email is sponsored by: NEC IT Guy Games. How far > can you shotput > a projector? How fast can you ride your desk chair down the > office luge track? > If you want to score the big prize, get to know the little guy. > Play to win an NEC 61" plasma display: http://www.necitguy.com/?r=20 > _______________________________________________ > Bodington-developers mailing list > Bod...@li... > https://lists.sourceforge.net/lists/listinfo/bodington-developers > |
|
From: Alistair Y. <ali...@sm...> - 2005-06-09 11:22:54
|
After the 11th would be better for me Jane. Alistair On 9 Jun 2005, at 11:37, Jane Jotcham wrote: > We need to set a date for the next developers' meeting to take > place prior to release 2.6. Can people please advise when they are > available in July from 5 July onwards. One thought I had was that > if most people are in Oxford for Buzz on 4 July the meeting could > be in Oxford on the 5th. However, I have not checked anyone's > availability for this so advice would be appreciated. > > Thanks > Jane > > > > > > ------------------------------------------------------- > This SF.Net email is sponsored by: NEC IT Guy Games. How far can > you shotput > a projector? How fast can you ride your desk chair down the office > luge track? > If you want to score the big prize, get to know the little guy. > Play to win an NEC 61" plasma display: http://www.necitguy.com/?r=20 > _______________________________________________ > Bodington-developers mailing list > Bod...@li... > https://lists.sourceforge.net/lists/listinfo/bodington-developers > |
|
From: Andrew B. <a.g...@le...> - 2005-06-09 10:55:49
|
I'm away 23rd June until 11th July. Aggie -----Original Message----- From: bod...@li... [mailto:bod...@li...] On Behalf Of Jane Jotcham Sent: 09 June 2005 11:37 To: bod...@li... Subject: [Bodington-developers] Next Developers meeting We need to set a date for the next developers' meeting to take place prior to release 2.6. Can people please advise when they are available in July from 5 July onwards. One thought I had was that if most people are in Oxford for Buzz on 4 July the meeting could be in Oxford on the 5th. However, I have not checked anyone's availability for this so advice would be appreciated. Thanks Jane ------------------------------------------------------- This SF.Net email is sponsored by: NEC IT Guy Games. How far can you shotput a projector? How fast can you ride your desk chair down the office luge track? If you want to score the big prize, get to know the little guy. Play to win an NEC 61" plasma display: http://www.necitguy.com/?r=20 _______________________________________________ Bodington-developers mailing list Bod...@li... https://lists.sourceforge.net/lists/listinfo/bodington-developers |
|
From: Matthew B. <mat...@co...> - 2005-06-09 10:48:08
|
Jane Jotcham wrote: > We need to set a date for the next developers' meeting to take place > prior to release 2.6. Can people please advise when they are available > in July from 5 July onwards. One thought I had was that if most people > are in Oxford for Buzz on 4 July the meeting could be in Oxford on the > 5th. However, I have not checked anyone's availability for this so > advice would be appreciated. At the moment I don't have anything in my diary (apart from the Buzz) around that time. -- +--Matthew Buckett-----------------------------------------+ | VLE Developer, Learning Technologies Group | | Tel: +44 (0) 1865 283660 http://www.oucs.ox.ac.uk/ | +------------Computing Services, University of Oxford------+ |
|
From: Jane J. <jan...@bo...> - 2005-06-09 10:31:19
|
We need to set a date for the next developers' meeting to take place prior to release 2.6. Can people please advise when they are available in July from 5 July onwards. One thought I had was that if most people are in Oxford for Buzz on 4 July the meeting could be in Oxford on the 5th. However, I have not checked anyone's availability for this so advice would be appreciated. Thanks Jane |
|
From: Sean M. <se...@sm...> - 2005-06-08 14:32:29
|
<quote who=3D"Peter Crowther">
I know I haven't, being the
> {Sybase,Microsoft} SQL Server weenie that I am.
>
> - Peter
>
>
he said it!-)
s
|
|
From: Matthew B. <mat...@co...> - 2005-06-08 13:29:29
|
Andrew Booth wrote: > Has anyone tried IBM Cloudscape (now Apache Derby) with Bodington? Not me. > It is a Java-based SQL RDMS originally developed by Cloudscape, then > purchased by IBM and finally donated to Apache as the open-source Derby. http://incubator.apache.org/derby/ > At first sight it looks like it might be a reasonable candidate for bundling > with Bod. Possibly more suitable for a production Bod than hSQL? Dunno. My quick glance suggests that it doesn't have those bitwise operators (manual may be out of date) which Bodington currently needs. More than that I don't know anything about it. The reason I orginally picked HSQLDB was I'd heard of it and other products (uPortal) used it in a similar way (quickstart) to the way I was looking to. If derby looks better I'm all for updating but I can't see any reason to change. -- +--Matthew Buckett-----------------------------------------+ | VLE Developer, Learning Technologies Group | | Tel: +44 (0) 1865 283660 http://www.oucs.ox.ac.uk/ | +------------Computing Services, University of Oxford------+ |
|
From: Alexis O'C. <ale...@co...> - 2005-06-08 13:27:23
|
Andrew Booth wrote: > Has anyone tried IBM Cloudscape (now Apache Derby) with Bodington? > > It is a Java-based SQL RDMS originally developed by Cloudscape, then > purchased by IBM and finally donated to Apache as the open-source Derby. > > At first sight it looks like it might be a reasonable candidate for bundling > with Bod. Possibly more suitable for a production Bod than hSQL? > > Aggie > I think Derby is definitely worth a look, but I think for database choices the emphasis should still be HSQLDB for evaluation (single user), PostgreSQL / MS SQL / Oracle for production (possibly hundreds of concurrent users). I'm not convinced Derby would scale for the latter case and in the former case is not as lightweight! (I don't think it can be run in memory for instance). Alexis |
|
From: Peter C. <Pet...@me...> - 2005-06-08 13:24:20
|
> From: Andrew Booth [mailto:a.g...@le...]=20
> Has anyone tried IBM Cloudscape (now Apache Derby) with Bodington?
Atif mentioned it on the developer list a couple of months ago, I think,
and was told to take a look at whether it was a better candidate than
hsqldb :-). Other than that... I know I haven't, being the
{Sybase,Microsoft} SQL Server weenie that I am.
- Peter
|
|
From: Andrew B. <a.g...@le...> - 2005-06-08 13:12:40
|
Has anyone tried IBM Cloudscape (now Apache Derby) with Bodington? It is a Java-based SQL RDMS originally developed by Cloudscape, then purchased by IBM and finally donated to Apache as the open-source Derby. At first sight it looks like it might be a reasonable candidate for = bundling with Bod. Possibly more suitable for a production Bod than hSQL? Aggie |
|
From: Colin T. <col...@co...> - 2005-06-06 16:10:31
|
I can't believe this problem is still lurking! We had a fix from Antony, which somehow escaped from our repository... I think there are still issues with the fix, relating to loading the CSS correctly (static vs. dynamic stylesheet) -- anyone looked at it recently? Colin ____________________________________ Colin Tatham VLE Team Oxford University Computing Services www.oucs.ox.ac.uk/ltg/vle/ www.bodington.org |
|
From: Matthew B. <mat...@co...> - 2005-06-06 16:05:00
|
Matthew Buckett wrote:
> Andrew Booth wrote:
>
>> Sounds a good idea. While you are at it, could you make owners have MARK
>> privilege by default in pigeonholes and short answer papers?
>
>
> I think thats in our list as well :-) I'll have a look at it in the next
> few days hopefully.
I'm going mad. I'd actually fixed that in WebLearn HEAD already! This
was the method I added to PigeonHoleFacility:
> /**
> * Setup the PigeonHole resource to a sensible state. We add Record, Mark
> * and Review Permissions to the owner for this.
> */
> public boolean create(Request request, Connection connection,
> Resource resource)
> {
> try
> {
> Acl acl = resource.getAcl();
> if (acl != null)
> {
> AclEntry aclEntry = acl.getOwnerGroupAclEntry();
> if (aclEntry != null)
> {
> aclEntry.addPermission(Permission.RECORD);
> aclEntry.addPermission(Permission.MARK);
> aclEntry.addPermission(Permission.REVIEW);
> return true;
> }
> }
> log.warn("Resource doesn't have default ACLs");
> }
> catch (BuildingServerException bse)
> {
> log.warn("Could not get ACL details for: " + resource);
> }
> return false;
>
> }
--
+--Matthew Buckett-----------------------------------------+
| VLE Developer, Learning Technologies Group |
| Tel: +44 (0) 1865 283660 http://www.oucs.ox.ac.uk/ |
+------------Computing Services, University of Oxford------+
|
|
From: Matthew B. <mat...@co...> - 2005-06-06 15:32:00
|
Andrew Booth wrote: [..snipped..] >>>If the visitor's IdP doesn't release the eduPersonPrincipalName > > attribute, > >>>the visitor doesn't get in. Yes - the users get put into the pass_phrase >>>table with a null or dummy passphrase. If necessary, we could prevent > > them > >>>from logging in except via Shib. > > >>Would having a shibb_user table be a simpler/cleaner way to get this to >>work? > > > The problem there is that the pass_phrase table is a very important one. It > is used for much more than just password authentication. If we move to a > shib_user table, there's a lot of code that will get duplicated and/or > changed. I'm inclined to put the shib users in with the other users and live > with the schema change. Ok. Looking at the call hierarchy for PassPhrase.getUserName() it seems to be reasonable confined to the user management stuff: > getUserName() - org.bodington.server.realm.PassPhrase > resetusername(UserManagementSession, Request, PrintWriter) - org.bodington.servlet.facilities.UserDirectoryFacility (2 matches) > userdata(Request, PrintWriter, String) - org.bodington.servlet.facilities.UserDirectoryFacility > getUsername(Request) - org.bodington.servlet.facilities.Facility (2 matches) > listusers(Request, PrintWriter, boolean) - org.bodington.servlet.facilities.Facility (4 matches) > outputAclTable(PrintWriter, Request) - org.bodington.servlet.facilities.AclDisplayFacility > createUsers(BufferedReader, PrintWriter) - org.bodington.server.realm.UserManagementSessionImpl > resetUserName(PrimaryKey) - org.bodington.server.realm.UserManagementSessionImpl > userdata(Request, PrintWriter, String) - org.bodington.servlet.facilities.PasswordFacility > denyAccess(URL) - org.bodington.servlet.facilities.FeedFacility (2 matches) > chooseUserName(Zone, User, Vector) - org.bodington.server.realm.UserManagementSessionImpl Before I dig around in the code does anyone know how X509User works? Does a certificate based user have an entry in the pass_phrase table aswell? -- +--Matthew Buckett-----------------------------------------+ | VLE Developer, Learning Technologies Group | | Tel: +44 (0) 1865 283660 http://www.oucs.ox.ac.uk/ | +------------Computing Services, University of Oxford------+ |
|
From: Matthew B. <mat...@co...> - 2005-06-06 15:23:22
|
Andrew Booth wrote: > Sounds a good idea. While you are at it, could you make owners have MARK > privilege by default in pigeonholes and short answer papers? I think thats in our list as well :-) I'll have a look at it in the next few days hopefully. -- +--Matthew Buckett-----------------------------------------+ | VLE Developer, Learning Technologies Group | | Tel: +44 (0) 1865 283660 http://www.oucs.ox.ac.uk/ | +------------Computing Services, University of Oxford------+ |
|
From: Andrew B. <a.g...@le...> - 2005-06-06 15:18:43
|
Sounds a good idea. While you are at it, could you make owners have MARK privilege by default in pigeonholes and short answer papers? Aggie -----Original Message----- From: bod...@li... [mailto:bod...@li...] On Behalf Of Matthew Buckett Sent: 06 June 2005 16:12 To: Bodington Developers Subject: [Bodington-developers] Allowing UPLOAD rights to Admins/Sysadmin Bodington by default doesn't allow UPLOAD rights to admins although they can add themselves (as they have MANAGE rights). I have changed this in the WebLearn tree. Is there a reason why this was the case? Should I push this change across to Bodington? -- +--Matthew Buckett-----------------------------------------+ | VLE Developer, Learning Technologies Group | | Tel: +44 (0) 1865 283660 http://www.oucs.ox.ac.uk/ | +------------Computing Services, University of Oxford------+ |
|
From: Andrew B. <a.g...@le...> - 2005-06-06 15:15:31
|
>>Just one of the issues that came up when we were doing the WebAuth=20 >>integration here at oxford was that having two URLs depending on which = >>authentication method was going to be used would cause problems for = users. >It means you can't give out a definite URL in your lectures because it=20 >depends on who your audience are. The definite URL is the same as it always was. Only those students = coming in via Shibboleth, to a restricted set of URLs will need the alternative = URL. Presumably, they know who they are. In most cases, they won't need to = know about the dual URLs, as they will simply click on a link in their 'home' Bodington, which then takes them via Shib to the resources revealed in = the 'target' Bodington. If these students try the /shibsite URL, they will = get in. If they try the /site URL, they will be challenged for a username = and password at the target. =20 >> If the visitor's IdP doesn't release the eduPersonPrincipalName attribute, >> the visitor doesn't get in. Yes - the users get put into the = pass_phrase >> table with a null or dummy passphrase. If necessary, we could prevent them >> from logging in except via Shib. >Would having a shibb_user table be a simpler/cleaner way to get this to = >work? The problem there is that the pass_phrase table is a very important one. = It is used for much more than just password authentication. If we move to a shib_user table, there's a lot of code that will get duplicated and/or changed. I'm inclined to put the shib users in with the other users and = live with the schema change. Aggie |
|
From: Matthew B. <mat...@co...> - 2005-06-06 15:11:43
|
Bodington by default doesn't allow UPLOAD rights to admins although they can add themselves (as they have MANAGE rights). I have changed this in the WebLearn tree. Is there a reason why this was the case? Should I push this change across to Bodington? -- +--Matthew Buckett-----------------------------------------+ | VLE Developer, Learning Technologies Group | | Tel: +44 (0) 1865 283660 http://www.oucs.ox.ac.uk/ | +------------Computing Services, University of Oxford------+ |
|
From: Matthew B. <mat...@co...> - 2005-06-06 14:35:09
|
Andrew Booth wrote: > >>>At the Bodington behind the SP, we have mapped /site and /shibsite to the >>>same bodington servlet. The Shibboleth servlet filter is set to protect > > the > >>>/shibsite URLs but not the /site ones, so the same resources can be >>>shib-protected or not depending on the URL used. > > >>Would it be preferable to use one URL for all access to Bodington from a >>user support point of view? > > > Maybe, but we have to allow non-shib operation of Bodington so that the Shib > filter doesn't kick in during normal operation. Using the /site URL as > normal provides this. Just one of the issues that came up when we were doing the WebAuth integration here at oxford was that having two URLs depending on which authentication method was going to be used would cause problems for users. It means you can't give out a definite URL in your lectures because it depends on who your audience are. >>What happens if the visitor doesn't give out username information? >>So are you putting these users into the pass_phrase table? > > > If the visitor's IdP doesn't release the eduPersonPrincipalName attribute, > the visitor doesn't get in. Yes - the users get put into the pass_phrase > table with a null or dummy passphrase. If necessary, we could prevent them > from logging in except via Shib. Would having a shibb_user table be a simpler/cleaner way to get this to work? Then users could have a shibb login and a bodington login associated with the same user. It would also mean that then existing installs wouldn't have to alter tables. Also it makes it easy to tell which users are shibb ones and which are internal bod ones. -- +--Matthew Buckett-----------------------------------------+ | VLE Developer, Learning Technologies Group | | Tel: +44 (0) 1865 283660 http://www.oucs.ox.ac.uk/ | +------------Computing Services, University of Oxford------+ |
|
From: Andrew B. <a.g...@le...> - 2005-06-06 14:22:37
|
>> At the Bodington behind the SP, we have mapped /site and /shibsite to = the >> same bodington servlet. The Shibboleth servlet filter is set to = protect the >> /shibsite URLs but not the /site ones, so the same resources can be >>shib-protected or not depending on the URL used.=20 >Would it be preferable to use one URL for all access to Bodington from = a=20 >user support point of view? Maybe, but we have to allow non-shib operation of Bodington so that the = Shib filter doesn't kick in during normal operation. Using the /site URL as normal provides this.=20 >> One thing that we need to point out is that usernames created in the >> SP-protected bodington are of the form use...@my... = - we >> therefore need to increase the size of the username field in the database, >> which is currently 30 characters wide. We propose to increase it to = 128 >> characters. (The same had to be done with mvnForum to cope with = usernames of >> this kind.) >What happens if the visitor doesn't give out username information? >So are you putting these users into the pass_phrase table? If the visitor's IdP doesn't release the eduPersonPrincipalName = attribute, the visitor doesn't get in. Yes - the users get put into the = pass_phrase table with a null or dummy passphrase. If necessary, we could prevent = them from logging in except via Shib. Aggie |
|
From: Matthew B. <mat...@co...> - 2005-06-06 14:11:54
|
Peter Crowther wrote:
>>From: Matthew Buckett
>>SELECT * FROM objects WHERE id IS NULL
>>
>>This is the query that is run when a connection is validated from the
>>connection pool. On PostgreSQL you cannot have an index
>>containing NULLs and so this results in a full table scan:
>
> That's a pretty poor optimiser - the id column doesn't even *allow*
> nulls. Presumably this clause was chosen as static analysis can reveal
> that it will return no rows. The SQL Server optimiser doesn't even need
> to go to the index statistics; SQL Server 2000 presents it as a
> 'constant scan' in the optimiser, with zero CPU and zero page reads.
Is this a SQL Server optimization? If I were connected to an Oracle
database I would do:
SELECT 1 FROM dual;
To test the connection (dual is a virtual table that always has one row).
If we used commons database pooling then it supports the idea that
different databases having difference queries for checking that the
connection is alive (rather than a hard coded one as we have at the moment).
The PostgreSQL people would probably argue it's a stupid query and
shouldn't be run in the first place ;-)
> Out of interest, why PostgreSQL? There are other RDBMSs these days that
> provide the required capabilities and are less... er... niche. The cost
> to Oxford of tuning this code to work well with PostgreSQL may well be
> higher than the cost of switching to an alternative RDBMS.
For an open source product working well with other open source products
is normally seen as a good thing (if we exclude Java for a sec). So I'd
say that even if it isn't the best thing for Oxford at the moment it
benefits the Bodington community in the long run. And it's never a bad
thing to have the ability to run on multiple databases.
Although buying an expensive database (like Oracle/MS SQL) and then not
using any of it's clever bits (functional indexes, etc) seems like a bit
of a waste.
>>The other big killer is the calculation of the size of the
>>"special" groups:
>>
>>SELECT count(*) FROM objects, users WHERE type = { } AND id = user_id
>>AND ( special_groups_a & ({ }) <> { } )
>>
>>One option would be to special case the ACL code of that it always
>>assumes the special groups are too big (easy and quick).
>>
>>The other option is to replace the special groups with normal ones.
>
>
> Given the lack of bitwise operators in HSQLDB and the inability of
I have a patch in the HSQLDB queue that fixes this.
> SoftCache to handle cached results where bitfields are involved, it may
> make sense to replace the special groups entirely. It rather depends on
> the queries involved - do we know what they are?
The special groups are allusers, allstaff, allstudents, and maybe more.
It's the groups that normally have 1000s of users and provides a very
fast way of checking if the user is a member. It does this check without
a database hit as the group information is stored in the user row. But
maybe this is premature optimization?
>>NB: Changing the connection check and making sure special groups are
>>never expanded changes the acl display page load time from around 4
>>seconds to less than half a second.
>
> Amending the connection check is a no-brainer, I agree; the trick is to
> amend it so that performance on other RDBMSs with different optimisers
> is not affected.
commons database pooling is probably the better answer as we also get
the ability to recover from database restarts.
> By the way... it's great to see someone profiling execution times and
> tracking down the performance bottlenecks. Assuming the tools in
> Bodington are going to use the same database schema for some time to
> come, it would be well worth the community doing this for the
> frequently-used or unacceptably slow tools in order to optimise them in
> a similar way. I don't believe that Bodington is inherently slow; I
> *do* believe there are a small number of hot-spots that would make a big
> difference if optimised.
This is where having a nice database with good support for profiling
makes things easier (Oracle, MS SQL). As turning on statement logging on
our live server would probably give an unacceptable performance hit (but
I'll look into this).
--
+--Matthew Buckett-----------------------------------------+
| VLE Developer, Learning Technologies Group |
| Tel: +44 (0) 1865 283660 http://www.oucs.ox.ac.uk/ |
+------------Computing Services, University of Oxford------+
|
|
From: Peter C. <Pet...@me...> - 2005-06-06 13:44:52
|
> From: Matthew Buckett=20
> SELECT * FROM objects WHERE id IS NULL
>=20
> This is the query that is run when a connection is validated from the=20
> connection pool. On PostgreSQL you cannot have an index=20
> containing NULLs and so this results in a full table scan:
That's a pretty poor optimiser - the id column doesn't even *allow*
nulls. Presumably this clause was chosen as static analysis can reveal
that it will return no rows. The SQL Server optimiser doesn't even need
to go to the index statistics; SQL Server 2000 presents it as a
'constant scan' in the optimiser, with zero CPU and zero page reads.
Out of interest, why PostgreSQL? There are other RDBMSs these days that
provide the required capabilities and are less... er... niche. The cost
to Oxford of tuning this code to work well with PostgreSQL may well be
higher than the cost of switching to an alternative RDBMS.
> The other big killer is the calculation of the size of the=20
> "special" groups:
>=20
> SELECT count(*) FROM objects, users WHERE type =3D { } AND id =3D =
user_id=20
> AND ( special_groups_a & ({ }) <> { } )
>=20
> One option would be to special case the ACL code of that it always=20
> assumes the special groups are too big (easy and quick).
>=20
> The other option is to replace the special groups with normal ones.
Given the lack of bitwise operators in HSQLDB and the inability of
SoftCache to handle cached results where bitfields are involved, it may
make sense to replace the special groups entirely. It rather depends on
the queries involved - do we know what they are?
> NB: Changing the connection check and making sure special groups are=20
> never expanded changes the acl display page load time from around 4=20
> seconds to less than half a second.
Amending the connection check is a no-brainer, I agree; the trick is to
amend it so that performance on other RDBMSs with different optimisers
is not affected.
By the way... it's great to see someone profiling execution times and
tracking down the performance bottlenecks. Assuming the tools in
Bodington are going to use the same database schema for some time to
come, it would be well worth the community doing this for the
frequently-used or unacceptably slow tools in order to optimise them in
a similar way. I don't believe that Bodington is inherently slow; I
*do* believe there are a small number of hot-spots that would make a big
difference if optimised.
- Peter
|
|
From: Matthew B. <mat...@co...> - 2005-06-06 12:38:48
|
Andrew Booth wrote: > Just to keep you up to date, we now have two bodingtons communicating IdP to > SP and we have automatic account creation working at the SP end. It > shouldn't take too long to get the groups sorted as well. Cool. Sounds very nice. > At the Bodington behind the SP, we have mapped /site and /shibsite to the > same bodington servlet. The Shibboleth servlet filter is set to protect the > /shibsite URLs but not the /site ones, so the same resources can be > shib-protected or not depending on the URL used. Would it be preferable to use one URL for all access to Bodington from a user support point of view? > One thing that we need to point out is that usernames created in the > SP-protected bodington are of the form use...@my... - we > therefore need to increase the size of the username field in the database, > which is currently 30 characters wide. We propose to increase it to 128 > characters. (The same had to be done with mvnForum to cope with usernames of > this kind.) What happens if the visitor doesn't give out username information? So are you putting these users into the pass_phrase table? > Do either of these changes (servlet mapping and field width) break anything > anyone else is doing? Doesn't sound like it will (famous last words). -- +--Matthew Buckett-----------------------------------------+ | VLE Developer, Learning Technologies Group | | Tel: +44 (0) 1865 283660 http://www.oucs.ox.ac.uk/ | +------------Computing Services, University of Oxford------+ |
19 messages has been excluded from this view by a project administrator.
