iRODS Development Update: January 2016

During January we held a Technology Working Group, and worked with a Consortium member to migrate from 3.3.x to the 4.1 series.  We also continued work on iRODS 4.2: reimplementation of the build system in CMake, adding serialization to the rule engine framework, and further work on the python configuration system.

Technology Working Group

The TWG met on January 11 with the following agenda (slides):

  • New Plugin Framework
  • Resource IDs
  • Proposed Deprecations
    • XMessage
    • iphybun and the bundleResc
New Plugin Framework

We discussed the new plugin framework that has been enabled from our use of clang.  In short, the old framework relied on variadic parameters for type erasure which provided no type safety across the shared object boundary.  With the use of clang we can now use variadic templates along with std::function to hold plugin operations.  We store these operations in a map of boost::any for type erasure, but given the variadic templates we are guaranteed type safety at the point of call. This is a breaking change from the previous framework which will require an update to any plugins not shipped with the iRODS core. A follow up technology update on how to do this will be posted shortly.

Resource IDs

Currently the resource hierarchy is stored in the database as a string, and this has several draw backs. Primarily it makes the update of a hierarchy potentially slow given a large number of data objects in a given leaf. In order to address this issue we have proposed to deprecate the use of hierarchy strings in the catalog and move to the use of the resource id of the leaf on which the data object resides.  This abstraction allows for the immediate renaming of the resource, as well as swift editing of resource hierarchies. This change will require a schema update as an automatic part of the upgrade process to 4.2.

Proposed Deprecations

We proposed that XMessage, iphybun, and the bundleResc be deprecated for 4.2, and potentially removed or replaced at a future date. We invite comment from the user community regarding this issue.  If these are mission critical features to any of our users we would like to hear as soon as possible.

iRODS 4.1.8

The next release in the 4.1 series of iRODS is largely finished and awaiting sign-off by a Consortium member who is in the process of migrating from the legacy 3.3.x series.  Efforts this month have mostly been related to documentation and testing. The migration of our supported plugins has been completed, and a full suite of resource and authentication plugins can be expected alongside this release.

iRODS 4.2

Development of iRODS 4.2 continues in the experimental-build-system branch of the github repository in order to insulate the community from broken builds while doing the heavy lifting.  As mentioned above during the TWG, the resource ID work began and is mostly complete with updates to the replication rebalance code left to finish.

Development has also begun on the new build system using CMake, with the added benefit of the move to dynamically linked executables and plugins.  CMake will generate a build environment in a target directory which will be used as a basis for the packaging system, or as the target location for a run-in-place deployment.

We have also began work on the serialization system for the new rule engine framework. When a plugin operation is called, iRODS will automatically attempt to trigger a dynamic policy enforcement point. Currently, data is serialized into key-value pairs for the native iRODS rule engine.  Now that we have a range of choices in rule language, a common representation of data is necessary for cross language communication. Having chosen Avro, we have began creating schemas for the native iRODS data types which will allow for the maximal communication between the rule engine and iRODS. This implies that when a new data type is introduced to iRODS which is passed to a plugin operation an associated serialization function will also be required.