http://www.tfd.co.uk/kernel/sakai-kernel-impl/cobertura/index.html

However this is coverage without integration tests.

The base of the site is at http://www.tfd.co.uk/kernel

Please dont take any notice of the checkstyle reports as they have not been configured.

Sound like an edge case. Well not really, Flex network connections dont share the same connection path as the browser, and so will not share the session cookie with the browser. This sounds like an easy thing in sakai, just use the ?sakai.session parameter to bind the session to the sakai session, and thats all fine until you get into a load balancer.

mod_jk needs a session cookie to make a request sticky and send it to the correct app server. Fortunately the load balancer understands 2 forms. The JSESSIONID cookie and in the request URI. eg


http://localhost/servlets-examples/servlet/SessionExample;jsessionid=E0C23E770473D20640C4907F75304137.tomcat2?xyz=1


Will translate as

• URI:http://localhost/servlets-examples/servlet/SessionExample
• Query String:xyz=1
• SessionID:E0C23E770473D20640C4907F75304137.tomcat2

Looking at the mod_jk source code (1.2.26) the session ID is parsed to locate the ‘.’ and generate the app server node name of ‘tomcat2′. The list of workers is then checked to see if a node name matches the name from the session cookie.

If you set up the workers.properties file with JKLogLevel=debug you can see what is going on inside the loadbalancer


[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] map_uri_to_worker::jk_uri_worker_map.c (678): Found session identifier ‘;jsessionid=E0C23E770473D20640C4907F75304137.tomcat2′ in url ‘/servlets-examples/servlet/SessionExample;jsessionid=E0C23E770473D20640C4907F75304137.tomcat2′
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] map_uri_to_worker::jk_uri_worker_map.c (682): Attempting to map URI ‘/servlets-examples/servlet/SessionExample’ from 3 maps
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] find_match::jk_uri_worker_map.c (503): Attempting to map context URI ‘/*=ieb_lb_test’ source ‘JkMount’
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] find_match::jk_uri_worker_map.c (516): Found a wildchar match ‘/*=ieb_lb_test’
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] jk_handler::mod_jk.c (2222): Into handler jakarta-servlet worker=ieb_lb_test r->proxyreq=0
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] wc_get_worker_for_name::jk_worker.c (115): found a worker ieb_lb_test
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] wc_get_name_for_type::jk_worker.c (292): Found worker type ‘lb’
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] init_ws_service::mod_jk.c (888): Service protocol=HTTP/1.1 method=GET host=(null) addr=172.16.1.1 name=bort.caret.local port=80 auth=(null) user=(null) laddr=10.0.0.197 raddr=172.16.1.1 uri=/servlets-examples/servlet/SessionExample;jsessionid=E0C23E770473D20640C4907F75304137.tomcat2
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] service::jk_lb_worker.c (940): service sticky_session=1 id=’E0C23E770473D20640C4907F75304137.tomcat2′
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] get_most_suitable_worker::jk_lb_worker.c (778): searching worker for partial sessionid E0C23E770473D20640C4907F75304137.tomcat2
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] get_most_suitable_worker::jk_lb_worker.c (786): searching worker for session route tomcat2

[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] service::jk_lb_worker.c (962): service worker=tomcat2 route=tomcat2
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] ajp_get_endpoint::jk_ajp_common.c (2579): acquired connection pool slot=0
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] ajp_marshal_into_msgb::jk_ajp_common.c (553): ajp marshaling done
[Tue Jul 22 21:57:52 2008] [19198:1251589008] [debug] ajp_service::jk_ajp_common.c (2050): processing tomcat2 with 2 retries


You can see where the session is routed to the appserver node. There is a trick at this point that a little bit of experimentation reveals. The ID on the end of the session should match the name of the worker and should match the jvmRoute parameter and the sakai session ID.

If I have 2 nodes called tomat1 and tomcat2,

• I must have 2 workers in the load balancer called tomcat1 and tomcat2
• I must have 2 app servers with jvmRoute= tomcat1 and tomcat2 in the Engine element of the server.xml on each app server node
• And for Sakai, I must have a the serverId set to tomcat1 and tomcat2 on each appserver node

Sadly, anything else results in the request being sent to a random app server node.

• Install SVN 1.4.3 from the DMG at Tigris http://www.collab.net/downloads/community/
• Make certain you have a gcc compiler, probably from xcode on the OSX install disks
• Build Git 1.5.5 form source with a ./configure; sudo make install
• Symbolic link the subersion perl libraries into the perl build library
  
eg

ln -s /opt/subversion/lib/svn-perl/SVN /System/Library/Perl/Extras/5.8.6/SVN
ln -s /opt/subversion/lib/svn-perl/auto/SVN /System/Library/Perl/Extras/5.8.6/auto/SVN


Thats Git installed, now to sync the repo you are after using git-svn, now go an read Git for SVN Users

Having done that to mirror an SVN repo do


git-svn clone http://svn.foo.org/project/trunk
cd trunk


And start working. git commit . will commit locally, git svn dcommit will commit all changes to the remote SVN and rebase your local gir repo to the remote svn

There are man pages on all of this and more … worth reading

• The standards compliance of Safari is good, and gives devlopers a second browser platform with good compliance
• You thought Safari had got fast in v3.0, but it just got faster again, SquirelFish the core Javascript engine of WebKit just got 1.6 times faster than the version in v3.0 of Safari. see http://webkit.org/blog/189/announcing-squirrelfish/
• iPhone and Google Android is using WebKit behind the browser, so if it works on Safari it will more than likely work on iPhone, iTouch and any other mobile running Google Android in the future

Three reasons to support Safari

“Don’t send 6 experts into a room for 6 months to write a spec for an API, start with spec on 1 sheet of paper.”

So I can use the patch generations capabilities of git to generate a stream of patches.
cd sdata; git log . tells me the stream of commits in the subdir.
Having located the commit I want to start from
git-format-patch -o patches --relative d649f550a43d8f94cbe91a7c735a4a68cab4ba14 .
generates a number sequence of patch emails in a directory gitrepobase/patches. The --relative is vital since it makes the patch set relative to working directory. Unfortunately there is a bug in the next git command that means you must have the correct path for the patch to work -p4 used to strip path elements from a patch in git-am does not work for new files.

Once the patch set is created, I dont need to email it, I can just run the patch set against the target repo from files on disk. git-am -i ../camtools-trunk/patches/*.patch will interactively apply patches to the target repo, asking me for each patch.

Once the patch operation is complete a git svn dcommit commits the changes back to the target svn repo

With a bit of scripting, it should be possible to build a complete patch set with moves and transfer it between detached repos, just as if the modifications where made to the target repo by hand.

• I am not forced to commit when I restructure, so when I go out of network range I am not forced to stop 2 minutes after the network is gone.
• I can run multiple branches in the same code base, a quick git commit -a, git checkout otherbranch changes the active files on disk
• I dont have to persuade anyone to change the central repo, if there is one
• I can batch up all my svn commits into a single big block, and make a decision to commit when I am done

Ok the sync takes a while for a very large repo, like the whole of sakai, but once synced everything becomes much easier. And you will notice the I,I,I above, thats because its distributed so it doesn’t effect you (unless you want to use it). All of this goes for others like Hg.

I am only wondering why I didnt use it earlier…. lots of people said it was a good idea. Lesson there, listen more, think less.

Building Git from source is easy on OSX, just download one of the tarballs from linux.org, configure make install, but getting it to work with SVN is not quite so easy because by default it binds to the OSX perl which may or may not have Alien::SVN

MacPorts appears to work: sudo port install git-core +svn

After fixing a bug, and getting the database built by starting tomcat, (2h), I popped up Cayenne modeler and reverse engineered the DB. (10s ), Garbage in Garbage out I thought. Well not exactly, the modeler identified a number of tables with no proper PK’s and generated all the classes, and generated a significant number of the relationships between the tables. Its going to be interesting to see if these classes can replace the data storage area. Especially if its going to give us enough control in the AuthZGroups service to get any performance.

Oh, and I now have core SQL DDL scripts for DB2, Derby, Ingres, Postgres, Sybase, SQLServer and naturally HSQLDB, Mysql and Oracle…. all thanks to the cayenne plugin.

On other reason for thinking that this is worth doing is that Cayenne considers the request cycle to be the transaction boundary, and has built in caching. A test query I didn in the OpenSocial mode took 49ms first time round and 3ms second time round, indicating that the caching is working (mysql Query cache disabled to highlight DB load)