This is a list of development priorities the Ironic team is prioritizing for Pike development, in order of priority. The primary contact(s) listed is/are responsible for tracking the status of that work and herding cats to help get that work done. They are not the only contributor(s) to this work! The number of primary contacts is limited to 2 maximum to simplify communication. We expect at least one of them to have core privileges to simplify getting changes in.
|Standalone CI tests||vsaienk0|
|Generic boot-from-volume||TheJulia, dtantsur|
|Rolling upgrades||rloo, jlvillal|
|Reference architecture guide||jroll|
|Python 3.5 compatibility||JayF, hurricanerix|
|Deploying with Apache and WSGI in CI||vsaienk0|
|Driver composition||jroll, dtantsur|
|Feature parity between two CLIs||rloo, dtantsur|
|OSC default API version change||mariojv, dtantsur|
|Rescue mode||mariojv, JayF|
|Finish node tags||zhenguo, dtantsur|
|Specific fault support||JayF, mariojv|
|Post-deploy VIF attach/detach||sambetts, vsaienk0|
|Physical network awareness||sambetts, vsaienk0|
|Routed networks support||sambetts, vsaienk0|
|Neutron event processing||vdrok, vsaienk0|
|Documentation reorganization||JayF, TheJulia|
|IPA REST API versioning||sambetts|
|Split away the tempest plugin||soliosg, jlvillal|
|Deploy steps||yolanda, rloo|
|Redfish driver||lucasagomes, jroll|
|Supported power states API||dtantsur|
|Available clean steps API||rloo|
|E-Tags in API||galyna, vdrok|
|IPA authentication||jroll, joanna|
We are working on a set of tests that can be run without nova present. We expect this work to improve CI time, coverage and help test certain features (e.g. adoption) that are not quite compatible with nova.
This work allows generic hardware to boot from remote storage, allowing diskless nodes to be managed by ironic. This also lays down the framework for hardware-specific implementations to be built.
Many OpenStack projects are beginning to support rolling upgrades - we should too. Let’s do our part to make downtimes a thing of the past. This involves code changes, new multi-node grenade CI jobs, and reviewer/developer documentation. The target now is Ocata -> Pike rolling upgrades.
To help new deployers make the right choices, we need a document describing a reference architecture for an ironic deployment, especially around multi-tenant networking and co-existing with VMs.
We’ve done most of the coding. Let’s concentrate on stabilizing the feature, writing new hardware types, polishing documentation and helping vendors with onboarding.
Let us make sure that all features implemented in the old ironic CLI are available in the new OSC-based openstack baremetal CLI.
Currently the default API version OSC sends to ironic is an old Kilo-era one. We need to figure out the path towards making the latest version the default.
This is necessary for users that lose regular access to their machine (e.g. lost passwords). The spec was merged in Newton, the code is partially done, let’s put some effort into making progress here in Pike.
Currently we communicate certain node errors via setting maintenance mode. This is not obvious, and is a constant source of complaints from operators. This work allows us to better communicate such failures, by providing a new API for working with them.
We already support attaching and detaching VIFs to nodes as part of the deployment process. Now we need to support the same for active nodes.
We need to make sure instances are not scheduled on nodes that cannot physically reach the networks they’re connected to. This may require data model changes around ports. This is required for Routed networks support.
Ironic should become aware of L2 segments available to connected networks as well as which L2 networks are actually available to nodes to correctly pick subnet (IP address) when doing provisioning/cleaning.
Currently ironic has no way to determine when certain asynchronous events actually finish in neutron, and with what result. Nova, on the contrary, uses a special neutron driver, which filters out notifications and posts some of them to a special nova API endpoint. We should do the same.
For some time, our documentation has grown without strict definition of structure, as well as difficulties getting documentation into the primary OpenStack documentation. In the Pike cycle we plan on reaching a consensus of direction for our documentation structure so we can better self-service as OpenStack evolves to better support projects such as ironic.
IPA API is currently not versioned, which causes problems when ironic starts relying on new features. Versioning similar to ironic API is expected to fix it and simplify upgrades.
Currently we rely on certain hacks to make our CI use master version of the tempest plugin on all branches. The QA team suggests moving our tempest plugin to a separate branch instead, let’s do it. We should also merge ironic and ironic-inspector plugins for simpler maintenance and consumption.
This is an effort to split parts of our monolithic deployment process into steps, similar to cleaning. That will give driver authors a bit more freedom in customizing the deploy process, and simplify potential additions to it (like RAID, for example).
The soft power/NMI spec proposes exposing available power states in the API. We didn’t implement this part in Ocata, let us finish it now.
We need to expose available clean steps in the API, so that users know which actions they can run during manual cleaning. This is a part of the manual cleaning spec which was never implemented, despite the spec being marked as done.
Neither IPA API, nor the lookup/heartbeat endpoints it uses are currently authenticated. While this issue is hard to solve completely in generic case, we can at least consider reducing potential attack surface.