Skip to main content
Version: v1.4

Release Notes

1.4.4​

This release introduces high availability for Luna. The Luna Manager can now run multiple replicas, with one elected as the leader to handle all incoming requests. Enable this feature by setting the Helm value manager.leaderElection.enabled to true.

Fixed an issue where a pod’s destination ID hash would change if it had an attached volume. This affected bin-selected nodes created with Luna versions 1.3.2 and earlier.

Avoid including the deprecated includeArmInstance field in the ConfigMap when it is empty.

Google GKE​

Fixed a configuration matching issue related to node pool reuse for bin-select pods with placement-related annotations.

1.4.3​

This release resolves several bugs.

Google GKE​

Resolved a crash that occurred when an existing nodepool LinuxNodeConfig is nil.

Added hugepages configuration matching with LinuxNodeConfig.

Microsoft AKS​

Fixed an issue with tracking requests for bin-select node pool reuse.

1.4.2​

Fixed slow eviction pods with an associated pod disruption budget.

Previously, Luna could enter a continuous loop of cordon, drain, timeout, uncordon, and pod rescheduling for pods with a PDB and graceful termination duration longer than Kubernetes’s default.

1.4.1​

Switched to fully-qualified image names.

Amazon EKS​

Added the configuration option aws.disablePqcCurves to disable Post-Quantum Cryptography (PQC) algorithms when using Luna. These new curves may not be compatible with certain firewalls, such as AWS Network Firewall or Suricata.

Updated the AWS Go SDK to v1.39.2.

Google GKE​

Added support for the gcp.linuxNodeConfig configuration value, which allows changes to sysctls, cgroup mode, and hugepages configuration for Linux nodes.

Updated the Google SDK to v1.44.1.

1.4.0​

This release introduces enhanced support for multiple architectures. Nodes with amd64 and arm64 architectures can now be provisioned based on the kubernetes.io/arch node selector.

Resolved an issue with updating the pricing category for existing bin-packing node pools.

Added an event to pods when no suitable node can be provisioned for them.

Amazon EKS​

Uses EC2 instances rather than EKS managed node groups to determine the node instance profile.

Exclude AWS instance type a1, that are not supported by Amazon Linux 2023.

Google GKE​

Fixed an issue with node type selection on GKE. Certain node types were overlooked if they were not available in the default zone.

Prevent the ARM64 taint added by GKE from being interpreted as a node pool change.

Microsoft AKS​

Fixed disk configuration matching in AKS to prevent unnecessary node pool churn.

Oracle OKE​

Update OKE node pool naming to avoid overflow issues with new bin-packing node pool versions.

Ensure that flex node types with A2 processors respect the required memory ratio.

1.3.4​

This release resolves an issue with node instance profile detection in Amazon EKS.

Amazon EKS​

In version 1.3.3, Luna introduced automatic detection of the node instance profile based on the EKS node groups’ configuration. However, this method proved unreliable in certain scenarios. To address this, Luna now queries the EC2 instances attached to the cluster—similar to the old deployment script’s approach. This update ensures compatibility with EKS clusters that lack node groups.

1.3.3​

This release includes two new features, which are determining the node instance profile on EKS automatically and generating node cost estimate events for pods with the nodecostestimate schedulingGate. The release updates the available node types and their pricing for Amazon EKS. It also contains several bugfixes, including addressing a problem with the binSelectNodeTypeRegexp option, improving scale-up time to add multiple gpu-enabled nodes, avoiding logging spurious error messages when pod garbage collection fails with pod not found, and filtering node types whose prices are unexpectedly set to 0, seen for a dynamically-fetched price for a new node type on AKS.

Amazon EKS​

Luna was enhanced to automatically determine the node instance profile if aws.nodeInstanceProfile is empty. This feature requires an update to the IAM policy for Luna Manager to determine the node instance profile. Refer to the upgrade notes for this release.

The set of available node types and their pricing was updated.

Azure AKS​

Luna's AKS code was modified to filter node types whose dynamically-fetched prices are reported as 0, which was observed for a node type that was recently added.

1.3.2​

This release includes metrics enhancements, improved pod garbage collection behavior, AWS Spot Advisor support, and cloud provider updates.

We made some changes to Luna manager’s metrics:

Added:

  • scale_errors_total — tracks scaling operation failures

Renamed:

  • cluster_gpu_limit_exceeds_total → gpu_requests_exceeding_cluster_limit
  • insufficient_free_addresses_in_subnet_thrown_total → insufficient_free_addresses_in_subnet_errors_total

When Luna scales down a node, daemonset pods that were running on that node occasionally persist in the Kubernetes inventory as orphaned zombie pods; Luna pod garbage collection cleans up those pods. Luna now uses a two-stage process to improve pod garbage collection reliability: First it attempts to remove the pods with foreground deletion. After loopPeriod, if a pod isn’t removed, it switches to background deletion.

Amazon EKS​

Luna now supports AWS Spot Instance Advisor, which provides spot savings data and interruption frequency metrics.

To enable it, set aws.useSpotAdvisor to true. You can configure constraints on using spot instances based on the spot advisor data: the spot savings rate with aws.maxSpotPriceRatio, and the interruption frequency filter with aws.maxSpotInterruptBucket.

Google GKE​

The node listing and pricing was updated.

1.3.1​

This release includes changes to Luna’s internal database format. Before upgrading to version 1.3.1, ensure all node creation requests with spot pricing are complete and no spot node provisioning operations are in progress. In the rare case a spot node creation request fails during the upgrade, Luna may incorrectly mark the node type as unavailable for on-demand instead of spot. The duration of unavailability depends on the failure type and the option configured in the backoffDurations configuration section, by default 1 hour or less.

General​

This release introduces several configuration options:

  1. maxPodsPerNode sets the maximum pods parameter for both bin-select and bin-packed nodes
  2. nodeTypeRegexp and binSelectNodeTypeRegexp filter node types using regular expressions
  3. nodeTypePricing and binSelectNodePricing configure pricing models (on-demand or spot) for all nodes or bin-selected nodes specifically

This release resolves the following issues:

  • Resolved an issue where pods with long graceful termination periods would block the Luna loop, causing 10+ minute delays in processing pending pods and evictions before specified graceful termination period elapsed
  • Fixed a crash that occurred when no bin-selected node types were available
  • Fixed a regression where bin-packing placement groups could be formed from pods with mismatched zone affinity, zone spread, and host spread [rare; introduced in release 1.2.13 (February 2025)]

Oracle OKE​

Enhanced unavailable shape detection during node creation, enabling Luna to retry with alternative shapes more quickly.

1.3.0​

This release introduces support for Amazon Linux 2023 with EKS. Note that EKS 1.33 requires Amazon Linux 2023 and drops support for Amazon Linux 2.

This release optimizes the provisioning of bin-selected nodes by improving the calculation of required nodes. Luna now avoids creating unnecessary bin-selected nodes when multiple bin-selected pods can be scheduled on a single node.

Anti-affinity configuration has been added to webhook pods to ensure they run on different nodes whenever possible.

A bug that caused Luna to crash during pod eviction from nodes has been fixed.

Amazon EKS​

With the addition of Amazon Linux 2023 support, Luna now supports ARM-based instances with GPUs, including the g5g and p6e families. ARM-based GPU instance types use dedicated image configuration values: aws.amiIdGpuArm and imageSsmQueryGpuArm.

Security group tags can now be configured using aws.securityGroupSelector.

1.2.21​

This release introduces minor features and improvements:

Added a new binSelectInstanceFamilyExclusions option, allowing users to exclude specific bin-select node types from consideration. This serves as an alternative to the node.elotl.co/instance-family-exclusions pod annotation.

Introduced the caBundle configuration option, enabling the use of a custom certificate authority bundle with the admission webhook.

Luna pods now include a node selector to ensure they run on the appropriate architecture and operating system.

Improved logging output during node draining operations.

1.2.20​

We’ve resolved an issue introduced in Luna 1.2.19 where the node.elotl.co/force-bin-select annotation was accidentally removed from the configuration annotations list. This caused pods using this annotation to generate different hashes on restart, changing their node selector. As a result of this regression, existing bin-selected nodes weren’t being reused, forcing the creation of new nodes for restarted pods.

1.2.19​

You can now add extra environment variables to the Luna manager and webhook pods via the manager.extraEnv and webhook.extraEnv Helm values.

The deployment script now cleans up temporary files from the installation directory after completion.

Luna now deletes Kubernetes node objects before terminating the underlying node instances. This change allows finalizers on RKE2 nodes to be removed, preventing orphaned node objects from lingering in the etcd database.

SUSE RKE2​

Luna now supports running on SUSE RKE2 clusters provisioned with Amazon EC2 nodes.

Installation documentation for RKE2 on EC2 will be available soon. In the meantime, please contact us for deployment assistance.

Amazon EKS​

Updated pricing data and added support for the new instance types: c8gd, m8gd, and r8gd.

Introduced configuration options: aws.subnets and aws.securityGroups. If these are not specified, Luna automatically queries subnets and security groups using the EKS cluster information. These options allow you to restrict the subnets in which Luna nodes run and to assign additional security groups to Luna nodes as needed.

Google GKE​

Luna no longer attempts to garbage-collect node pools that are in provisioning, reconciling, or stopping states. This prevents logs from being cluttered with non-actionable messages (e.g., Nodepool ... garbage collect failed (...); nodepool is rechecked in future pass).

1.2.18​

Amazon EKS​

This change ensures that the Luna manager pod will have the environment variables AWS_REGION, AWS_DEFAULT_REGION, and AWS_STS_REGIONAL_ENDPOINTS defined via the deployment manifest. A new helm chart value, aws.clusterLocation, is introduced, which when set to the cluster's region ensures that these environment variables are injected in to the Luna manager pod.

This release introduces a Helm chart value, aws.userDataType, to configure how user data is passed to EC2 instances. Setting aws.userDataType to Template allows the operator to create custom scripts to configure their EC2 nodes. This can be used to support non-standard nodes.

1.2.17​

This release addresses several issues which could cause Luna to overshoot its clusterGPULimit option. It also modifies Luna to reduce a scale request node replica count when the full replica count would exceed clusterGPULimit.

The release fixes issues that interfered with the expected coalescing of bin-select pod placement requests when reuseBinSelectNodes is set.

The release adds the option webhookConfigPrefix to allow changing the prefix of Luna's webhook, so that the order in which it is run with respect to other webhooks can be modified.

The release updates the Luna manager to log all environment variables at startup, to facilitate troubleshooting configuration issues.

Amazon EKS​

The Luna manager service account was updated to include the annotation eks.amazonaws.com/sts-regional-endpoints: "true", to reflect Luna's use of AssumeRoleWithWebIdentity.

Google GKE​

Luna's spot instance offering was changed from using preemptible virtual machine instances to using spot virtual machine instances.

1.2.16​

This release adds two new bin-select pod annotations, which can be used to constrain attributes of the instance type Luna may allocate for the pod. The annotation node.elotl.co/instance-max-cost limits the estimated USD cost per hour of the instance and node.elotl.co/instance-max-gpus limits the GPU count of the instance, i.e., Luna will not choose an instance type for the bin-select pod that violates these limits, if specified.

The release introduces the option spotPriceRatioEstimate (default 0.5) to make Luna's spot to on-demand price ratio estimate user-visible and configurable.

The release adds the option pendingPodReasonRegexp, which can be set to a regular expression for which a pod's pending reason message must match for Luna to consider it.

The release updates Luna's backoff handling for an instance type to capture the CPU, memory, and GPU configuration used for the instance type being avoided, since some instance types on some clouds allow multiple configurations.

The release removes some system configuration variables from the configuration map; they are now passed to luna executables via the command line.

This release includes several improvements specific to GKE as described in the next section.

Google GKE​

This release modifies Luna to require that the instance type be explicitly constrained if gcp.minCpuPlatform is used.

The release fixes an issue where Luna could generate unsupported instance type candidates for GPU SKUs added to N1 instances. It also fixes an issue where scale down of a node pool, which Luna performs after a failed scale-up operation, would fail on GKE regional clusters.

1.2.15​

This release addresses all high severity vulnerabilities reported by Dockerhub scanning. At this point, Dockerhub scanning shows no critical or high severity vulnerabilities.

The release improves the handling of cloud scale-up operations that take longer than scaleUpTimeout. When this occurs, Luna's subsequent scaling attempt backs off from using the node type and pricing category associated with the long latency for a period specified by backoffDurations.ScaleUpTimeoutBackoff.

The release improves contention detection and handling on GKE, as described in the GKE-specific section below.

The release addresses a price handling issue on AKS, as described in the AKS-specific section below.

Google GKE​

Luna was updated to detect and handle two additional kinds of contention on GKE.

GCP scaling operations can fail due to an incompatible operation running, e.g., when the GKE control plane is being upgraded. Luna now detects when a nodepool scale-up operation encounters this contention, and backs off from re-attempting the operation for a period configurable via backoffDurations.nodepoolIncompatOpBackoff.

GCP limits the number of concurrent operations that can run in a GKE cluster. Luna now detects when cluster operations fail due to this limit, and backs off from running scale-up operations for period configurable via backoffDurations.tooManyOpsBackoff.

Also, an issue impacting the Luna code that enhances scaling efficiency by opportunistically combining scale operations for multiple pods was addressed.

Azure AKS​

Luna's AKS price fetching code was updated to filter prices which have an effective end date that has passed.

1.2.14​

This release adds support for specifying the minimum CPU platform for GKE nodes.

The release enhances Luna's logging when no bin-select node meets the configured requirements, so that the user can more easily determine how to update the requirements to allow a node to be allocated.

The release reverts the change made in 1.2.12 to avoid a rare certificate setup issue at webhook startup, given its unintended impact on a key customer installation use case. Installation retry can instead be used if the rare certificate setup issue occurs.

The release adds options to configure the backoff times for allocating classes of unavailable resources, so that the user can set those values to match their operating environment, if desired.

The release improves Luna's spot price estimates to better handle spot vs on-demand selection. It also fixes a role policy issue that impacted Luna's spot termination event handling feature on EKS.

Google GKE​

Adds the option gcp.minCpuPlatform to specify the minimum CPU platform for GKE nodes, and the annotation node.elotl.co/gcp.minCpuPlatform to override the setting for bin-select pods. Please see minCpuPlatform for details.

Amazon EKS​

Fixes an issue that impacted Luna's access to spot termination warning messages when the option aws.spotSqsQueueName is set.

1.2.13​

This release updates the operation of binPackingNodePricing and node.elotl.co/instance-offerings when set to "spot,on-demand" to try on-demand allocation if the spot capacity for a node type is exhausted.

The release also improves particular scale-up situations on EKS and OKE.

Amazon EKS​

For EKS, the release enhances scale-up handling of insufficient capacity in a zone by retrying the allocation in other zones recommended in the API response. The release also adds recognition and response to spot quota issues during scale-up.

Oracle OKE​

For OKE, on clusters using OCI VCN IP Native networking, the release fixes an issue in the default value handling of the Luna options binSelectMaxPodsPerNode and binPackingMaxPodsPerNode that could lead scale-up to fail due to unexpected IP address exhaustion.

1.2.12​

This release improves the management of instance type choices on clouds for which Luna uses node pools [AKS, GKE, and OKE].

For GKE, it updates the machine types and prices, handles the newly-added goog-gke-accelerator-type label on GPU node pools, and adds support for detecting, logging, and directly responding to node pool scale operations that fail at runtime.

Common​

Add the option binPackingKeepNodeType [default true] to avoid bin-packing node pool churn solely to change the instance type for lower node cost [applies to AKS, GKE, and OKE].

Reduce latency in bin-selection placement when original instance type choice is unavailable on clouds for which Luna uses node pools [applies to AKS, GKE, and OKE].

Fix rare certificate setup issue at webhook startup.

Amazon EKS​

Filter unsupported instance types. This includes any metal instance types and instance types with accelerators other than Nvidia, i.e., AMD GPU, AWS Neuron GPU, Intel GPU, Qualcomm GPU, and FPGAs.

Google GKE​

Update GCP machine types and prices.

Ignore the newly-added GPU instance label goog-gke-accelerator-type during node pool configuration matching.

Add support for detecting, logging, and directly responding to node pool scale operations that launch successfully but fail at runtime.

1.2.11​

This release adds node preservation features from scale-down operations, fixes two rare scaling issues in GKE and AKS, and handles the newly-added label goog-gke-node-pool-provisioning-model in GKE.

Common​

You can now prevent specific nodes from being scaled down by adding node.elotl.co/do-not-remove=true as a label or an annotation to the node. Protected nodes will not be scaled down, drained, or affected by nodeTTL.

Two rare scaling issues are fixed:

  1. When scaling up node pools in GKE and AKS, failure wasn’t handled properly and led to unnecessary scale-down retries on the node pool.
  2. Bin-selected node pools may have mismatched instance types leading to failure when scaling up bin-selected node pools with no node.

Amazon EKS​

Added support for i8* instance types.

Google GKE​

The label goog-gke-node-pool-provisioning-model is ignored during node pool configuration matching.

1.2.10​

This release adds two features: a pod annotation to force Luna to use bin-select node allocation and an option to support Artifact Streaming on Luna-allocated Azure AKS nodes.

Common​

Add the pod annotation node.elotl.co/force-bin-select: "true", to have Luna use bin-select for a pod when it would have otherwise used bin-packing.

Azure AKS​

Add the Luna option azure.enableArtifactStreaming to enable Artifact Streaming, an Azure preview feature that can reduce pull time for container images stored in ACR premium tier, on Luna-allocated AKS nodes.

1.2.9​

This release fixes minor issues, adds a label with the packing mode to node objects, and adds support for Artifact Registry Image Streaming with GKE.

Common​

Adds a new label to node objects: node.elotl.co/mode with values bin-packing or bin-selection. This label allows operators to determine the packing mode for the node.

Fix pod retry delay when node pool scale-up operation fails. If a scale-up operation fails, the pod was ignored until the timeout podRetryPeriod expired. Now if the node creation failed, the pod will be retried at the next iteration.

Fix eviction path and add extra logging to avoid a rare case where a node with low utilization would get cordoned and uncordoned repeatedly.

Google GKE​

Add support for Artifact Registry Image Streaming image streaming, which can significantly reduce image loading time. Luna detects if Image Streaming is enabled on the cluster and takes the extra memory overhead associated with it into account when calculating the available memory on the nodes it allocates.

1.2.8​

This release fixes minor issues and introduces Bottlerocket support for AWS.

Common​

Fixes node over-allocation for zone-spread StatefulSets with bound PVC placement. Luna over-allocated bin-selected nodes during the initial zone-spread placement of the StatefulSet pods.

Fixes out-of-capacity error handling for bin-selection.

Allows critical pods like CoreDNS to run on Luna-managed nodes even when these pods aren’t handled by Luna. This prevents resource exhaustion of non-Luna nodes when CoreDNS is scaled up.

Optimizes API calls to the control plane by using PATCH operations instead of UPDATE when modifying pods and nodes.

Reduces Luna manager’s memory consumption by 5% to 10%.

Updates Kubernetes’ client-go dependency to 0.29.10.

Amazon EKS​

Adds the option aws.isBottlerocketImage. When enabled, Luna handles node initialization and userData with the TOML format expected by the Bottlerocket images.

1.2.7​

This release adds disk options for AKS nodes, allowing operators to better control the OS disk and use Ultra SSD volumes. It also improves GKE support with better IAM role configuration and node overhead calculation.

Common​

The deployment script now has a --skip-cert-manager option to skip the detection and auto-installation of cert-manager. Note that if cert-manager isn’t properly installed in the cluster when using this option, the Luna installation will fail.

For cloud providers on which zone affinity and zone spread are supported (GKE and EKS), handle zone-spread pods for which zone affinity is inherited from a bound PVC.

Update recommended NVIDIA driver version and update the cuda vector add examples to use the latest image from NVIDIA.

Google GKE​

Luna now requires fewer permissions. Luna manager no longer uses roles/container.developer, roles/compute.instanceAdmin.v1, or roles/container.admin; instead, it uses roles/iam.serviceAccountUser and a custom role created via the deployment script.

Microsoft AKS​

Luna supports new disk options for AKS nodes:

1.2.5​

This release enhances Luna with improved support for various cloud providers and addresses several minor issues.

Google GKE​

Added support for additional raw block devices. These options allow you to build node-level caches for pods running on the nodes.

You can specify a number of NVMe or Ephemeral raw block devices to attach to each node.

For bin-selected nodes, you can also specify node-specific options with pod annotations.

Microsoft Azure​

Recognize new node types with A10 Nvidia GPUs: Standard_NC, Standard_ND, Standard_NV36, Standard_NV72, and instance types with the v5 suffix.

Oracle OKE​

The resource requirements for running the Kubernetes system on nodes have increased. This release updates nodes’ estimated available resources based on those new requirements.

Fixed a crash with Luna manager during node deletion failures.

1.2.4​

This release officially transitions OKE support from Beta to GA.

Amazon EKS​

The release updates the AWS instance types and prices, adding the G6 series of GPU systems and the R8G series of Graviton4-based ARM systems.

Oracle OCI​

The release updates the OCI instance types and prices, adding the VM.GPU.A10.1 and VM.GPU.A10.2 GPU systems and the VM.Standard.A2.Flex Ampere ARM systems.

The release adds support for Spot pricing, using OCI preemptible instances.

1.2.3​

This release significantly enhances Luna's security and stability. We audited Luna and fixed the critical and high vulnerabilities reported by various scanning tools. We added automated security scanning and reporting around Luna to ensure we ship a secure product.

General​

Our installation script deploy.sh now requires cmctl to be installed. We have also updated the version of cert-manager installed by the script to 1.15.2.

The telemetry option is now false by default.

The default CPU and memory requests and limits have been set. See the default Helm values for more information.

Added support for requests and limits ephemeral-storage to manager and webhook:

manager: requests: ephemeral-storage: 500Mi limits: ephemeral-storage: 1Gi

Added NetworkPolicy for both manager and webhook pods. You can configure the ingress and egress rules:

manager:
  networkPolicy:
    ingress:
      - ports:
        - protocol: TCP
          port: 9090
    egress:
      - ports:
        - protocol: TCP
          port: 8443

Added pod disruption budget for webhook pods to ensure there's always 1 pod available when upgrading or restarting.

We fixed various minor bugs with Luna.

There were also various improvements to the documentation like the Hot Node Mitigation Tutorial.

Microsoft AKS​

Password authentication support was removed, and the azure.Username & azure.Password helm values have also been removed. We also added documentation about Spot usage with AKS to make it easier to get started.

Google GKE​

GKE now supports Spot pricing.

1.2.2​

This release includes significant improvements and bug fixes for Microsoft AKS and Oracle OCI platforms, as well as general enhancements to logging and performance.

Microsoft AKS​

The nodeTags option is now supported on AKS. You can now add custom tags to Azure VMs managed by Luna.

Oracle OCI​

Bin packing nodepool configuration matching is now correctly handled; a configuration mismatch will result in the creation of a new nodepool with the new configuration.

Fixed a bug where Luna would "overshoot" the target number of nodes when scaling down a nodepool because of a race condition with the API. Luna now scales down nodepools with a single API call, eliminating this race condition.

General​

Better logging for instance selection, previously covered pods, and node utilization.

1.2.1​

Google GKE​

Added gcp.diskSizeGb to specify the disk size for GKE nodes.

Implemented support for instance out-of-stock error handling.

Webhook​

Luna uses pod CPU and memory resources requests for key elements of its operation. Therefore warnings are logged when bin-selected pods don’t have full resource requirements specified.

General​

Add max pods per node support for all cloud providers. See the documentation for details.

Luna now features management of over-utilized bin-packed nodes. Users can configure Luna to distribute CPU and/or memory usage across multiple nodes by evicting the overburdening pod(s). For details, see the BinPacking section in the documentation.

1.2.0​

Amazon EKS​

aws.maxPodsPerNode has been deprecated and replaced with binPackingMaxPodsPerNode and binSelectMaxPodsPerNode.

Google GKE​

gcp.maxPodsPerNode has been deprecated and replaced with binPackingMaxPodsPerNode and binSelectMaxPodsPerNode.

Oracle OCI​

oci.maxPodsPerNode has been deprecated and replaced with binPackingMaxPodsPerNode and binSelectMaxPodsPerNode.

General​

We have unified the management of maximum pods per node parameter across cloud providers by replacing the provider-specific options (*.maxPodsPerNode) with the new binPackingMaxPodsPerNode and binSelectMaxPodsPerNode parameteRs/index. Additionally, this update introduces support for configuring Max Pods Per Node on Azure, bringing consistency to all cloud environments we support.

1.1.0​

Google GKE​

  • Luna can now configure auto-upgrade, auto-repair, secure boot, and integrity monitoring for its node pool.

General​

We have updated the Kubernetes client libraries to version 0.29.1. This update inadvertently broke the computation of the reuse hash value. We have since fixed this issue in Luna to prevent such regressions in the future. However, upgrading from versions prior to 1.0.1 may result in bin-selected node pools not being reused. See the upgrade page for more details.

1.0.1​

Fixed an issue with the trial version of Luna where images were being pulled from the incorrect Docker repository, causing installation failures. The regular versions were not affected.

1.0.0​

Amazon EKS​

  • Fix EKS zone spread node overshoot. Too many nodes were created once the node(s) required for zone spread expired. Now the correct number of nodes is maintained during node expiration.

Google GKE​

  • Luna now avoids triggering node garbage collection on locked node pools, and logs a warning instead of an error when it attempts to operate a locked node pool.
  • Node pool garbage collection now includes orphaned instance templates. No need to clean these manually anymore.
  • Support network tags, and GCE instance metadata. The Helm values gcp.networkTags and gcp.gceInstanceMetadata can be used to customize these configuration options.
  • Fix unnecessary creation of new node pool when there are minor differences in the nodes metadata.

General bug fixes​

  • Fix annotation when nodes are successfully drained.
  • Set minimum value for non-zero nodeTTL to prevent unnecessary node churn. The minimum value for scaleDown.nodeTTL is now scaleUpTimeout.