v1.3.0

Changes by Kind

Feature

• Add azure specific rules and routes to be configured for secondary nic. (#156, @tamilmani1989)
• Dranet no longer injects child VFs attached to an excluded uplink as pod network interfaces. (#176, @anson627)

Bug or Regression

• Address several critical edge cases in default interface detection (#153, @aojea)
• Fixed DRANET failing to publish any ResourceSlice on nodes where an interface had enough global-scope IP addresses for the joined dra.net/ipv4 or dra.net/ipv6 attribute to exceed DRA's 64-byte string attribute limit (commonly triggered by kube-ipvs0 on clusters running kube-proxy in IPVS mode). The oversized IP attribute is now omitted on the offending device; all other attributes and devices in the slice are unaffected. (#198, @dkennetzoracle)
• Fixed issue where RDMA device is not restored to the host network namespace in rdma netns=exclusive mode (#180, @kanlkan)
• PCI network devices whose kernel driver has been unbound or replaced with a userspace driver (vfio-pci, uio_pci_generic, igb_uio, pci-stub) are no longer published in the ResourceSlice. Previously such devices were published but every NodePrepareResources call for them failed, trapping pods in FailedPrepareDynamicResources. (#193, @wevans-ant)

All changes

• fix: Remove 'dranet' from registry name to avoid duplication by @gauravkghildiyal in #154
• fix: Install npm during netlify build by @gauravkghildiyal in #155
• accurately detect default gateways by @aojea in #153
• helm: default image tag to Chart.AppVersion by @fmuyassarov in #160
• Add OKE GB200 examples: DRA NIC allocation, MNNVL, placement-group by @dkennetzoracle in #157
• Remove obsolete github pages action by @gauravkghildiyal in #171
• docs: fix dead link in with .md extension by @ngcxy in #173
• prevent uplink child virtual function injected into pods by @anson627 in #176
• Azure: generate per-device network config from IMDS metadata by @tamilmani1989 in #156
• Bump golang version to 1.26 by @ngcxy in #184
• Fix postsubmit failure caused by Bats tests for metric server by @ngcxy in #189
• add GPU EFA example for AWS EKS by @anson627 in #182
• Add AKS AMD GPU example and consolidate Nvidia GPU example by @anson627 in #188
• Update release process doc and improve release image tagging by @gauravkghildiyal in #185
• Add tamilmani1989 as reviewer by @tamilmani1989 in #192
• Bug Fix: RDMA device is not restored to the host network namespace in rdma netns=exclusive mode by @kanlkan in #180
• Improve look for the docs and add dark mode by @gauravkghildiyal in #195
• fix: Skip PCI network devices whose kernel driver is unbound by @wevans-ant in #193
• Fix typos: priviledge, directoy, lenght in docs and Dockerfile by @SAY-5 in #197
• Resourceslice attr length overflow by @dkennetzoracle in #198
• Consolidate pod-level state into podConfigStore by @purvavj in #191
• chore: drop unnecessary helm installation during workflow by @fmuyassarov in #201
• ci: declare contents: read on bats, helm-lint, periodics, test by @arpitjain099 in #200
• Add PyTorch training example by @anson627 in #202
• Add NIXL kv cache transfer example by @anson627 in #203

New Contributors

• @ngcxy made their first contribution in #173
• @wevans-ant made their first contribution in #193
• @SAY-5 made their first contribution in #197
• @purvavj made their first contribution in #191
• @arpitjain099 made their first contribution in #200

Full Changelog: v1.2.0...v1.3.0