- version v19.04 (latest stable/1904)
- fixed GSO (Generic Segmentation Offload) issue on VPP
- fixed an SRv6 k8s service implementation issue
- GSO does not work correctly with SRv6
- (IPv6 only): network Policies are implemented using ip6tables rules in individual pods. Because of this, the policy programming is a bit slower (compared to policy programming on VPP for IPv4)
- version v19.04 (latest stable/1904)
- GSO (Generic Segmentation Offload) is enabled on VPP by default, which brings better performance for pod-to-pod communication via TAP interfaces
- SRv6 as a new node-to-node transport option
- new helm option
contiv.nodeToNodeTransportwith possible values:vxlan(default),srv6, ornooverlay - experimental (PoC) support for service chaining between pods
- (IPv6 only): service load-balancing in IPv6 setup is not equal, node-local backend pods are always preferred and a request is never load-balanced to a remote node's pod if there is a local backend
- (IPv6 only): network Policies are implemented using ip6tables rules in individual pods. Because of this, the policy programming is a bit slower (compared to policy programming on VPP for IPv4)
- version v19.01 (latest stable/1901)
- update to vpp-agent
v2.1.0 - multi-master and external ETCD support
- experimental SRv6 implementation of k8s services
- experimental support for multiple pod interfaces, including memif interfaces
- load-balancing between backends of a service now supports unlimited backend pod count
- (IPv6 only): service load-balancing in IPv6 setup is not equal, node-local backend pods are always preferred and a request is never load-balanced to a remote node's pod if there is a local backend
- (IPv6 only): network Policies are implemented using ip6tables rules in individual pods. Because of this, the policy programming is a bit slower (compared to policy programming on VPP for IPv4)
- version v18.10 (latest stable/1810)
- update to vpp-agent
v2.0.1(includes a GoVPP fix for unix-domain socket communication) - fixed IPv6 policy rendering for host-network pods
- load-balancing between backends of a service is limited to the first 256 PODs (the others will not receive any traffic until some of the first 256 PODs disappear)
- (IPv6 only): service load-balancing in IPv6 setup is not equal, node-local backend pods are always preferred and a request is never load-balanced to a remote node's pod if there is a local backend
- (IPv6 only): network Policies are implemented using ip6tables rules in individual pods. Because of this, the policy programming is a bit slower (compared to policy programming on VPP for IPv4)
- version v18.10 (latest stable/1810)
- IPv6 Support
- internal process manager is now used to manage Contiv-Agent & VPP processes inside of the vswitch container (instead of the supervisor tool)
- ability to use unix domain socket for connection to VPP for better performance
(
contiv.vswitch.useSocketVPPConnectionhelm option) contiv.useL2Interconnecthelm option renamed tocontiv.useNoOverlay- Vagrant optimized to boot faster using pre-build images
- fixed Prometheus stats exporter
- load-balancing between backends of a service is limited to the first 256 PODs (the others will not receive any traffic until some of the first 256 PODs disappear)
- (IPv6 only): service load-balancing in IPv6 setup is not equal, node-local backend pods are always preferred and a request is never load-balanced to a remote node's pod if there is a local backend
- (IPv6 only): network Policies are implemented using ip6tables rules in individual pods. Because of this, the policy programming is a bit slower (compared to policy programming on VPP for IPv4)
- version v18.10 (latest stable/1810)
- added
etcd.service.useNodeIPHELM option
- vswitch base image updated to the newest ubuntu:18.04 build, which fixes some security vulnerabilities
- load-balancing between backends of a service is limited to the first 256 PODs (the others will not receive any traffic until some of the first 256 PODs disappear)
- version v18.10 (latest stable/1810)
- use helm chart version as the default tag for contiv images in the helm template
- fixed some Contiv UI issues
- load-balancing between backends of a service is limited to the first 256 PODs (the others will not receive any traffic until some of the first 256 PODs disappear)
- version v18.10 (latest stable/1810)
- global helm option for default gateway
- documentation & dev guide
- fixed issue where default route in pod was not configured after IP address change
- load-balancing between backends of a service is limited to the first 256 PODs (the others will not receive any traffic until some of the first 256 PODs disappear)
- version v18.10 (latest stable/1810)
- default ETCD version update to 3.3.11
- enhanced
vppctlscript
- load-balancing between backends of a service is limited to the first 256 PODs (the others will not receive any traffic until some of the first 256 PODs disappear)
- version v18.10 (latest stable/1810)
- both Contiv/VPP and the underlying ligato/VPP-Agent have underwent a major refactoring
- VPP-Agent is now based on a newly implemented framework, called kvscheduler, providing transaction-based configuration processing with a generic mechanism for dependency resolution between configuration items, which in effect simplifies and unifies the configurators
- for Contiv, the VPP-Agent refactor yields more stability and also visibility to what changes are being made in the configuration in the form of additional REST interfaces and with much more readable and compact logs
- Contiv/VPP itself has been refactored into a synchronous event-driven control-flow model, further simplifying debugging and limiting potential race conditions
- the previously bloated
contivplugin has been split into multiple smaller plugins to improve readability and modularity - it will be substantially easier in the future to add new features as separate plugins without the need to change anything or too much of the existing code base, potentially opening Contiv/VPP for external contributors - the support for re-synchronization has been fine-tuned to ensure that the system state gets properly recovered after a restart or any kind of failure/outage
- network configuration is no longer persisted in
etcd, mitigating the load onto the datastore, instead the agent is now able to re-calculate the full configuration state on demand (only Kubernetes state remains reflected intoetcdby KSR) - Kubernetes state data are also mirrored from
etcdinto localboltDB on every node, allowing the agent to restart and recover the state without immediate connectivity toetcd - run-time change of the DHPC-assigned main IP address is now supported
- run-time change of
NodeConfigCRD is now also properly applied without the need for a hard restart - external applications are able to extend/customize the Contiv dataplane
by submitting additional configuration via
etcdorgRPCAPI, which then gets merged with Contiv's own internal configuration before it gets applied to VPP-Agent - more information are available here - L2 (no overlay) networking mode is again supported
podVPPSubnetCIDRhas been removed, POD-facing interfaces on VPP are now unnumbered
- load-balancing between backends of a service is limited to the first 256 PODs (the others will not receive any traffic until some of the first 256 PODs disappear)
- version v18.10 (latest stable/1810)
- fix contiv-netctl when http uses self-signed certs
- added
cipherSuitesoption for etcd into helm - native PCI driver support for VMware vmxnet3
- load-balancing between backends of a service is limited to the first 256 PODs (the others will not receive any traffic until some of the first 256 PODs disappear)
- version v18.10 (latest stable/1810)
- fixed various NAT (high load) & ACL (large scale) issues on VPP
- do not route ServiceCIDR towards VPP from Linux by default
- minor vpp-agent infra fixes
- mark ETCD disconnect as non-fatal for liveness & readiness probe in KSR & CRD
- parametrized liveness & readiness probe intervals for all components
- parametrized CPU requests for all Contiv components
- load-balancing between backends of a service is limited to the first 256 PODs (the others will not receive any traffic until some of the first 256 PODs disappear)
- version v18.10 (latest stable/1810)
- use parametrized ETCD image repo&tag in the vswitch init container
- lowered log verbosity
- minor CRD improvements
- version v18.10 (formal release)
- properly handle change events in service processor before the first resync
- close opened fd on unsuccessful switch to network ns
- netctl supports clusters where etcd & rest is secured
- version v18.10 (formal release)
- fix nodePort issues after change of management IP address between restarts
- STN: Fix proxy ARP for /31 subnets
- bump ETCD version to 3.3.10
- add new liveness probe for Contiv ETCD
- Policy Configurator: test for rule duplicity in O(log(n)) time
- remove persistent storage for IPAM (use ETCD instead)
- topology validator enhancements
- version v18.10-rc1~15-g347c523
- race condition in DHCP notification handling
- added missing pull policy for init containers
- fixed issues in STN setup
- fixed some
contiv-netctland CRD issues - minor IPAM fixes
- better handling of identity NAT on VPP
- k8s 1.12 compatibility
- the topology validator gives false positives for L3-FIB entries
- version 18.10-rc0-505
- support for more than one IP on the management interface
- concurrent map access fix in ligato/vpp-agent
- option for simplified IPAM config (
ContivCIDR) - ability to define IPAM via CRD
- cluster state validator
contiv-netctlcommand line tool
- fragmentation issues in STN setup (STN is still experimental)
- due to new VPP version and some recent changes in the VPP NAT plugin, the following is the new default recommended VPP startup config for the NAT:
nat {
endpoint-dependent
translation hash buckets 1048576
translation hash memory 268435456
user hash buckets 1024
max translations per user 10000
}
(for more details see VPP NAT Documentation).
- version 18.04 stable + Contiv bugfixes
- lots of bugfixes in both VPP and Contiv.
- MTU-related issues in STN setup (STN is still experimental).
- version 18.04 stable + Contiv bugfixes
- dedicated POD VRF and interconnect VRF,
- local Bolt DB used for IPAM storage,
- port names fully supported in k8s policies.
- most of NAT related issues fixed in STN setup.
- lots of other bugfixes in both VPP and Contiv.
- MTU-related issues in STN setup (STN is still experimental).
- version 18.01 stable + Contiv bugfixes
- vSwitch POD waits for ETCD becoming ready before starting,
- secured internal GRPC connections (CNI, STN),
- options for securing connections to Contiv ETCD & Contiv Agent HTTP services,
- configurable IP neighbor scanning.
- fixed issues in some restart / upgrade scenarios,
- Gratuitous ARP fix on VPP.
- port names not fully supported in k8s policies,
- NAT-related issues in STN setup (STN is still experimental).
- version 18.01 stable + Contiv bugfixes
- full k8s services support,
- full k8s policies support,
- support for vSwitch / node restarts,
- experimental STN (Steal The NIC) functionality.
- connection to Contiv ETCD, HTTP services and internal GRPC connections (CNI, STN) are not secured,
- port names not fully supported in k8s policies,
- NAT-related issues in STN setup (STN is still experimental).