Skip to content

TODO.md

Latest commit

 

History

History
86 lines (82 loc) · 6.61 KB

TODO.md

File metadata and controls

86 lines (82 loc) · 6.61 KB

Bugs

  • Fix repeated/failed creation of connectors on the router
  • Figure out the issue with incorrect return values from API functions
  • Deleting sites and backbones causes exceptions to be repeatedly caught in the management controller
  • There is a name-collision thing happening when a backbone is deleted and then re-created
  • Multiple Access Points can be created on the same site with the same name
  • Claim against an expired invitation is honored

General Stuff

  • Generalize and parameterize the "hash-sets" for site-to-management reconciliation allowing for greater flexibility in development
  • Use PG notifications instead of polling to detect database changes
  • Explore ways to virtualize the Kubernetes content on backbones/member-sites for non-kube environments
  • Consolidate string definitions
  • When reconciling changes that are immutable, delete the synced object and re-reconcile for the create
  • Do something sensible when AMQP destinations are no longer reachable (i.e. credit runs out)
  • Consider issuing temporary credentials so that the first site of a new backbone can be bootstrapped on another backbone
  • Set up router policies to restrict access to management-plane APIs
  • Router-feature: Use a PKI signature for the router configuration so the configuration cannot be altered
    • HMAC digest for the initial configuration that is signed by the router's private key
    • Digest is carried in the connection properties and is enforced by the connected router
    • If policy (or other security) configuration is altered after the initial startup, the router is shut down
    • Consider also sending the cryptographic signatures of the pod images in the connection properties
    • Inter-router/Edge connections are:
      • Authenticated via MTLS
      • Accepted only if the connecting site has an authentic initial configuration
  • Consider having the sites generate their own certificates and CSRs to be sync'd to the MC for signing
    • This would require the installation of openssl into the site-controller container image
  • Build APIs and front-end for managing the tree of certificates. This includes:
    • API for fetching all certificates signed by root or a CA (to support a tree-table)
    • configuring and commanding certificate rotation
    • viewing the important fields of a certificate

Backbone Feature Tasks

  • In initial setup, specify which ingresses are needed on the site based on configured access points (skupperx-incoming configmap?)
  • In site YAML, include outgoing peer-links to known access points
  • Figure out how to establish connectivity from the management controller to the backbone
  • Define the backbone-router-to-management-controller protocol and APIs
  • Implement the link APIs inband to the management controller
  • Demonstrate automatic backbone site bootstrap
  • On initial creation for backbone sites, only include manage and peer incoming links
  • Review the inband APIs to allow for full link updates at run time (add/remove links, rotate certificates)
  • Review the API for coherence in the URL paths
  • Review the module structure
  • Ensure that Kube watches are working as expected
  • Disable API for backbone-site config after the site has "checked in" with the management controller
  • Define the APIs for the front-end
  • Change the field validator to a validator/normalizer which returns a copy of the fields with their expected types
  • Trigger hash updates and heartbeats to sites when relevant database changes occur
  • Rename the site-controller image
  • Consider adding a "metadata" field to the interiorsites table to store opaque (JSON) data from the front-end
  • Provide guidance in the deployment of backbone sites - bootstrapping the first site; indicating which sites can be deployed in sequence - deployment-state
  • Simplify the JSON fields in the bootstrap process to remove unneeded fields and depth
  • Add a shell command in one of the site containers that extracts the host info for bootstrap-deployment
  • Delete orphaned TlsCertificate records when their owners are deleted
  • Use router names that are more descriptive and derived from the backbone site name
  • Clean up the async structure of the "prune" module
  • Add an entry point in the "prune" module to clean things up after objects are deleted
  • Add ingress capability for non-OpenShift kubernetes
  • Consider replacing the boolean ingress indication with "no-ingress", "any", or ingress-style suggestions
  • Add a PUT to allow backbones to be renamed
  • Add a "platform" column for sites - probably enumerated for formality
  • Return an error on the ingress-post (during bootstrap) if the input is not validated
  • Consider invitation-templates for backbones
    • Every new application network automatically comes with a set of pre-definied invitations which can be deleted or added-to

VAN Feature Tasks

  • Create non-backbone mode for the site-controller - disable ingress and other backbone-specific functions
  • Add a claim-handshake module to the site-controller
  • Design a mechanism by which invited sites become member sites
  • Add interactive invitations so member-console action is needed before the claim is asserted
  • Remove the skx-inject annotation for claim secrets as they are not needed in the router
  • Add the User API for access to application networks, invitations, and member sites
  • Remove backbone-mode and replace with backbone-enabled, member-enabled, claim-enabled. This allows a site to be a member and a backbone at the same time
  • Design the member API
    • Site status and site activation
    • Active and inactive application state
  • Allow users to rename member sites (affects site-scope addressing) - Alternatively, consider a separate site-scope-prefix field.
  • Bug: Invitation can be created with a primary access that is not a 'member' access-point

Application Definition Tasks

  • Define the data structures to support full application definition
  • Design the kube-state representation of the above data
  • Implement APIs to access the full application definition
  • Consider pre-loading a directory tree of sync-state in the management controller and separating the sync processing from the database (to reduce load on the DB)
  • Add transitive dependency - A "mixed" block might be dragged in by a South interface but have a North interface that recursively drags in another dependent block.
  • Consider un-constraining interface roles. Just make them arbitrary strings, defined by connector writers. Alternatively, allow users to add to the role list. Namespaced roles?? too much??