Stable Channel (default)

v1.0.5

Improvements:

  • allow own namespace and single namespace

Bug fixes

  • unavailability of unrelated subnet Issue #117
  • try getting daemon pod if cache not found when calling interface update Issue #118
  • continue after net-attach-def create/udpate error in PR #114

v1.0.4

CRD changes:

  • config:
    • add following dynamic config values
      // type ConfigSpec struct
UrgentReconcileSeconds int        `json:"urgentReconcileSeconds,omitempty"`
NormalReconcileMinutes int        `json:"normalReconcileMinutes,omitempty"`
LongReconcileMinutes   int        `json:"longReconcileMinutes,omitempty"`
ContextTimeoutMinutes  int        `json:"contextTimeoutMinutes,omitempty"`
LogLevel               int        `json:"logLevel,omitempty"`

Improvements:

  • apply linter code analysis removing unused function, handling errors
  • allow changing reconciler time and log level on the fly from config.multinic
  • change file strcuture
    • separate unit-test folder
    • move constatnt and shared varaible to vars package
  • allow changing synchronization ticker period (TickerIntervalKey) and maximum size of daemon pod watching queue (MAX_QSIZE) from environment variable
  • change logger time encoder to ISO 8601
  • support OwnNamespace/SingleNamespace namespace deployment

Bug fixes:

  • computeResult is reset at one point after CIDR update with no change
  • potentially hang at scale due to API server never return for large amount of listing
  • continue creating NetworkAttachmentDefinition even if it is failed to create/update NetworkAttachmentDefinition in some namespaces

v1.0.3

CRD changes:

  • cidrs:
    • add corresponding IPPool to CIDR spec (spec.cidr.hosts[*].ippool)
  • multinicnetwork:
    • add host discovery and processing progress (status.discovery)
    • add network config status (status.configStatus)
  • config:
    • add toleration spec for e2e integration test and future taint usecases (spec.daemon.tolerations)

Improvements:

  • separate log level on controller manager

    Verbose Level Information
    1 - critical error (cannot create/update resource by k8s API)
    - "Set Config" key
    - set up log
    - config error
    2 - significant events/failures of multinicnetwork
    3 - significant events/failures of cidr
    4 (default) - significant events/failures of hostinterface
    5 - significant events/failures of ippools
    6 - significant events/failures of route configurations
    7 - requeue
    - get deleted resource
    - debug pointers (e.g., start point of function call)

  • log CNI message on host

    • main plugin: /var/log/multi-nic-cni.log
    • IPAM plugin: /var/log/multi-nic-ipam.log
  • add CI tests
    • multi-nicd test (daemon and CNI components)
    • end-to-end 200-node scale test on kind cluster using kwok

Bug fixes:

  • sequential CIDR update blocking in scale (scale issue)
  • unexpected HostInterface deletion when API server is stressed (scale issue)
  • unexpected interface updates on HostInterface when API server is stressed (scale issue)
  • missing HostInterface at initialization (fault-tolerance issue)

v1.0.2

  • First release (as open source)

Core Features:

  • Host-interface auto-discovery
  • Single definition for multiple secondary network attachments
  • Multi-NIC IPAM (CIDR computation, IP allocation/deallocation) for multiple secondary subnets
  • L3 configurations on host neighbour route table corresponding to ipvlan CNI plugin with l3 mode

Supplementary Features:

  • NIC selection based on specific requested number or specific interface name list
  • Dynamic CIDR updates when
    • detecting added/removed hosts at creation/deletion of multi-nic daemon
    • (periodically) discovering added/removed secondary interfaces
  • Fault tolerance in scale (tested upto 100 nodes x 2 secondary interfaces) with
    • initial synchronization of CIDR, IPPool after controller restarted
    • periodic synchronization of L3 routes for hosts which were restarted and lost the configuration