Page History

...

Yinlin posted results from 3 tests: test 2, test 3, test 5
Esme posted results from test 5
Nick still running test #5 on janky system but it has been running for 3.5 weeks, over 7 million resources, still going
Esme raised real world performance issues
- Performance problems with nodes that link to thousands of other nodes, especially with proxies
- Testing/exploring ongoing
- Caching, improving performance of converting node reference to URI
- Currently need to look up node by UUID and convert to path to generate URI
- Esme: We could reverse relationship so each object links to its parent (rather than the parent linking to all children)
  - Works well for collections but not objects with lots of parts
Next meeting is August 15 at 11am ET

Supporting gzip compression of responses via Jersey filters

Aaron C: A lot of web servers allow you to request compressed resources via gzip so there is less network bandwidth
Fedora does not currently support this
There is a fairly standard way to do this with Jersey filters
Are there any objections?
Adam: Can we do this at the container level?
Aaron: Maybe - hasn’t done this before
Adam: Is there a downside to this?
Aaron: Not aware of any
Adam: Let’s file a ticket and do it
Adam: Should have a switch to disable for debugging

Aaron C: Jena 3.x has been out for more than a year
We currently use 2.13
There is no support/development for 2.x
Aaron C has a pull request into the core codebase
- Lots of files are changed due to package name changes and differences between RDF 1.0 and 1.1
- Pretty straightforward in terms of substantive changes
- fcrepo-transform uses ldpath-backend-jena, but there was a ticket to swap out that backend for ldpath-backend-linkeddata
- Unfortunately, this change is currently blocked by the Maven Enforcer module
Andrew does not want to move forward with the Jena 3.x pull request since fcrepo-transform can’t move forward
- Andrew says he will do the rebasing
Should we let the core codebase wait on certain extension modules with idiosyncratic dependencies?
- OSGi would help but this is a long term solution
- Adam: Can we relax the enforcer rules?
- Aaron: Yes
This caching functionality would be hugely beneficial to the community
Proposal: ~~Relax Maven enforcer on fcrepo-transform and fcrepo-webapp-plus~~, keep fcrepo-transform, swap out backend for ldpath-backend-linkeddata
- Proposal approved by attendees on the call
- Update: it is not necessary to relax the maven enforcer. The same thing can be accomplished with maven dependency exclusions.

Lots of help from the community
Still need more Windows testing
Jared volunteered to do Vagrant testing
Need volunteers to do testing with various databases
- Esme can test MySQL and PostgreSQL
We are 1 week into 2 week testing period
Aaron C: Do we need a new release candidate for new issues?
Adam: Yes, we will want an RC2RC3

Aaron C: Current policy is that master branch is frozen when RC is in flight
Changes go into dev branch and dev branch is merged to master
This is too complex and inhibits changes to master
Should have a master branch pinned to next major version
Proposal: When you create a new release series (e.g. 4.5.x) and start diverging from this series (e.g. 4.6.x), you need a maintenance branch for commits to the current series so new commits can continue going to master
- Proposal approved by attendees on the call