Attendees

General

Goals

Minutes

Walk through by goals

  • Clustering
    • Scott got "new relic" up and running (monitors nodes, report back to common dashboard), doesn't play well with existing setup
      • Fedora Primary is reporting (public node), but ones on private network not reporting
      • Some of the nodes can't see the New Relic coordinator, which is causing errors
      • Also causing initialization issues with infinispan which go away when the New Relic jars removed.
      • Creating a ticket for this issue, will speak with the infinispan group about it
    • Greg is going to have a meeting with local sys admins later today to work out puppet issues and handshake issues across nodes. Afterwards he will try to get Benchtool working on the UNC cluster
    • Scott discussed Tomcat shutdown problem when running fcrepo on a cluster configuration
      • Frank had been using ctrl-c to shut down tomcat, so hadn't run into the problem of the shutdown script not completing
      • https://www.pivotaltracker.com/s/projects/684825/stories/65988656
      • Frank suspect this may be a modeshape config issue, will investigate further
    • Some discussion about multicast UDP
  • Wiring and Config
    • Adam mentioned that the work here is taking place on the indexer code right now rather than in the fcrepo main code base
  • Unit test coverage > 75%
    • More unit tests the better, we should try to get close to 100% unit test coverage when submitting a new pull request
  • Prepare for AuthZ feature sign-off
    • We may be close to finished, but we need to take stock of what features have been completed and which use cases have been satisfied
    • There are tickets for each of the use cases
    • Talk with stakeholders about how to verify that their use case has been satisfied and go about doing so.
  • Performance testing with transactions
    • Using transactions in high throughput situations should be helpful, but we still need to verify that this is the case
    • To do this, we should use the existing tools and tests (namely Benchtool) and adapt it to work with transactions for comparison
    • Need to determine how to group the actions into reasonable transactions
    • Are there use cases for transactions already that we can use as the basis for this determination?

Please create additional tickets by stand up on Tuesday, so that we can prioritize them in that meeting.


Going through Pivotal Tracker

  • Establish Performance Goals for UNC - Greg
    • Been speaking to locals about what are likely to be the largest ingests they will receive in the next year, coming up with calculations of how long these actions should ideally take.
    • Adam suggested comparing against SCAPE requirements, and Frank spoke about how the SCAPE requirements were determined by the various stakeholders in the organization.
  • Figure out how to verify consistent data across clusters - Greg
    • Consistency does not seem to be an issue
    • This needs to be broken down into smaller individual tickets
    • Scott: this ties into transaction performance testing across clustered nodes
    • Adam: Up to this point, transactions have mostly
  • Remove dead SessionCodeFactory - KS

    • Too tight coupling between transaction apartaus and http, led to problems where you couldn't change clients between steps of transactions.  Now only the security principal needs to be shared across parts of transactions

    • Is SessionFactory still needed now that we have an authentication framework?  Investigate better ways to handle the changes.
  • Test different config settings with infi/jgroups in order to raise performance of FIZ cluster - Frank
    • Need to be able to use 20 threads in a 6 node cluster.  Right now can only use one thread.
      • Can't fire multiple threads against cluster, one request locks all nodes
    • Check into locking, including the granularity.  Frank has been looking at modeshape locking config changes, but they have not worked so far. 
    • He wants to analyze the issue some more before taking it to the modeshape group.  Is the problem with how fcrepo handle binaries or how modeshape/infinispan does?

All content on the LYRASIS Wiki is licensed under the CC BY (Attribution) license, unless otherwise noted.