Tags

v0.36.2

Small fix for v0.36.x, cleaning up an unintentional kernel logging
config change.

v0.36.1

Small fix for v0.36.0, removing unused kube-rbac-proxy roles and role
bindings.

v0.36.0

A lot of commits this time, grouped into 4 categories.

1. Bug fixes, which change the production behavior:
- 5e5465be neonvm-controller: recreate the pod for VM even if the older pod still exists (#1055)
- bfa69c2e neonvm-controller: check if deletion timestamp is in the past (#1055)
- dee47a11 neonvm-controller: add explicit "node not ready" tolerations with 30s grace period (#1055)
- c893f77e neonvm-controller: Release leader election lease on exit (#1091)
- 72267bab vxlan-controller: support dualstack (#1057)

2. Tech debt removal, which is not supposed to influence the
    production behavior, but might do so.
- a2f53e32 Flatten & unify component structure (#961)
- 498573bb neonvm-controller: Use virtio-mem when status.memoryProvider is nil (#1110)
- 11c8c933 neonvm-controller: Use virtio-mem by default (#1098)
- 0f3917ab runner: fix comments to reflect the removal of conatiner-mgr
- 9f002830 Revert "neonvm: Use container statuses, not pod phase, to trigger restart (#749)"
- b96a1ebd Revert "neonvm: Use crictl to change container CPU, ditch cgroup (#738)"
- ccf77daa neonvm: disable runner cgroup by default
- 954ec878 Revert "neonvm-controller: Enable container-mgr by default (#803)"

3. prep work for ARM support, not yet used:
- 89f08827 neonvm-daemon: Move main.go into cmd subdir (#1112)
- 4156e390 neonvm: Add new neonvm-daemon binary skeleton (#1090)
- 7194bd5d vm-builder: makefile targets to introduce cross-compilation of the linux kernel (#1073)

4. devprod changes, impact should be localized to the repo itself:
- 41924067 ci: call go-coverage-report only if base is the main branch (#1118)
- 42446905 e2e-tests: fix ps path in vm-neonvmd test (#1113)
- 351b7b4d Fix ECR access (#1088)
- 4ed6744f util: Remove BuildGitInfo (#1109)
- 1a7acbcf ci/vm-kernel: change CACHE_TAG calculation (#1106)
- d836bf3e ci: gh-workflow-stats-action: bump version and move action repo (#1105)
- 03612c61 ci/vm-kernel: Fix outdated comments (#1101)
- edbe23cc ci: report workflow stats  (#1099)
- ecdb9ede CI: Add PR title checker (#1087)
- 0caa474e vm-builder: Use fresh base for 'builder' stage (#1086)

Upgrade path from v0.35.0:
- No ordering requirements.

v0.36.0-test-ci

Test release to test CI.

v0.35.0

Somewhat bigger release, with a handful of new features, fixes, and
smaller improvements.

Features:

- neonvm: Add controller flag to disable runner cgroup (#1034)
- agent/core: Use cached LFC for memory scaling signal (#1031)
- neonvm: Support disk quotas (#1062)
- neonvm/multus: Add AKS support (#1059)

Breaking changes:

- neonvm: Remove `GuestSettings.SwapInfo` (#1064)

No protocol changes.

Fixes:

- vm-builder: Change sshd config to use internal-sftp subsystem (#1051)
  - Note: This allows scp to work normally.
- plugin/trans: Fix over-allowing from forceApprovalMinimum (#1025)
- Specify minor go version in .mod file (#1068)
- vm-builder: Remove qemu-img from built vm images (#1077)

Other changes:

- plugin: Add warning log message for the pods with assigned scheduler (#1037)
- Bump alpine base images: 3.18 -> 3.19 (#1058)
  - NOTE: This also bumps QEMU in neonvm-runner: 8.0.5 -> 8.1.5
- vm-builder: Use docker auth (#977)
- agent: Remove unused "error" exit status (#1061)
  - NOTE: this PR removes the autoscaling_agent_runner_fatal_errors_total metric.

Upgrade path from v0.34.x:

- No ordering requirements.

v0.34.0

Relatively small release, with some bugfixes.
NOTE: This release upgrades to k8s 1.28 !

Features:

- agent: add metric and log warning for denied scaling (#1029)
  - new metrics are:
    - autoscaling_agent_scaling_full_denials_total
    - autoscaling_agent_scaling_partial_approvals_total

Breaking changes:

- Upgrade k8s deps to 1.28 (#1005)

Protocol changes:
Fixes:

- neonvm-controller: Fix immutable `.spec.guest.memoryProvider` (#1026)
- agent: Don't include VM name in neonvm request metric labels (#1028)
- plugin: Fix missing buffer info in handleRequested verdict (#1024)

Other changes:

- neonvm-runner: Set hostname in the VM to "vm-<compute name>" (#998)
- Bump alpine base images: 3.17 → 3.18 (#1022)
  - NOTE: This also bumps QEMU from 7.1.0 -> 8.0.5

Upgrade path from v0.33.x:

- No ordering requirements.

v0.33.0 with dependecies upgraded to 1.28
(https://github.com/neondatabase/autoscaling/pull/1005)

v0.33.0

Big release with some big features.
NOTE: This release upgrades to k8s 1.27 !

Features:

- agent/billing: support Azure Blob Storage to store consumption events (#994)
- neonvm-controller: Add '--skip-update-validation-for' (#990)
- Implement scaling latency metrics through revisions (#983)
- agent/core: Implement LFC-aware scaling (#1003)

No breaking changes.
No protocol changes.

Fixes:

- vm-builder: don't use ANSI terminal formatting in vector logs (#984)
- neonvm-runner: Silence "Unknown parameter 'mode'" messages at VM startup (#997)
- go.mod: fix dependency version (#1004)
- agent/core: Fix "but but" in warning message (#1017)

Other changes:

- go.mod: Remove unused cilium replace directives (#992)
- vm-builder: Use script to copy dynamically linked deps to /neonvm/lib (#1010)
- Bump alpine base images: 3.16 -> 3.17 (#1010)
  - NOTE: This also bumps QEMU from 7.0.0 -> 7.1.0
- agent: Reduce logging (#1013)
- neonvm-controller: Requeue after 15s if Pending or Running (#1016)
- plugin: Log only once on successful agent request (#1015)

Upgrade path from v0.32.x:

- NeonVM MUST be upgraded before the autoscaler-agents.

v0.32.2-latency-metrics

v0.32.2 but with latency metrics (#983)

v0.32.2 with dependecies upgraded to 1.27 (https://github.com/neondatabase/autoscaling/pull/929)
And also with https://github.com/neondatabase/autoscaling/pull/984 and https://github.com/neondatabase/autoscaling/pull/996

v0.32.2

Fix for v0.32.1, just containing a fix to the generated neonvm.yaml, so
that we don't have "neonvm-neonvm-runner-image-loader". See #933.

v0.32.1

Small release, three notable changes:

1. plugin: fix Score reserving when over-budget (#987)
   (fixes a critical bug with scheduler >= v0.31.0)
2. Remove go.mod dependency on controller-tools (#985)
3. neonvm: Move runner image env var into YAML (#988)

v0.32.0

Overall, relatively small release. One big item about adding support for
virtio-mem.
NOTE the breaking changes below.

Features:

- neonvm: Add support for virtio-mem (#970)

Breaking changes:

- neonvm: Make spec min/max/use resources required (#971)

No protocol changes.

Fixes:

- neonvm-controller: Set podName only while Pending (#973)
- neonvm: Use memory_hotplug.online_policy=auto-movable for virito-mem (#981)
  - This notably fixes OOM issues that can occur by running out of
    ZONE_NORMAL when there is a lot of hotplugged memory.

Other changes:

- neonvm-controller: Emit CpuInfo/MemoryInfo every reconcile (#974)

Upgrade path from v0.31.0:

- All VMs in the cluster must have spec min/max/use resources set.
- All clients that may create VMs must set spec min/max/use resources.
- NOTE: neonvm-controller still uses DIMM hotplug (instead of
  virtio-mem) by default. It IS NOT SAFE to roll back if any VMs are
  using virtio-mem; they will permanently break. Refer to #970 for more.

Test release to verify image upload to ECR. Do not use.

Test release to vefiry image upload to ECR

Same as v0.31.0 but with 1.27 k8s API

v0.31.0

Small-ish release by number of commits, but some larger refactorings
thrown in. NOTE: This release upgrades to k8s 1.26 !

No new features.
No breaking changes.
No protocol changes.

Fixes:

- neonvm-controller: Fix error in panic catching middleware (#965)
- neonvm-runner: Exit with error if QEMU fails (#964)
- neonvm-controller: Fix log sampling (#968)
  - (again)

Other changes:

- agent,api: Add support for fetching LFC metrics (#895)
  - Currently, this doesn't actually do anything! It's just setting up
    the piping so we can test it now and use it later.
- plugin: Add transactions for speculative operations (#936)
- plugin: Unify handleReserve+handleRequested (#936)
- Upgrade k8s dependencies to 1.26.15 (#928)
- util: Add taskgroup (#922)
- neonvm-runner: Remove parted,sfdisk from final image (#966)
- agent: Set log level to info (#969)

Upgrade path from v0.30.x:

- No ordering requirements.

v0.30.1-k8s-1.26

v0.30.1 but with 1.26 k8s API

v0.30.1

A hotfix after v0.30.0, only one change:
neonvm-controller: bring back RecordSuccess for conflicting (#957

v0.30.0-k8s-1.26

v0.30.0 but with 1.26 k8s API