Docs

Volcengine Kubernetes Engine

User guide

Pod annotation descriptions

When creating pods in VCI via virtual nodes in a Kubernetes cluster, you can add annotations to pods as needed. This allows you to fully utilize VCI's features without changing Kubernetes semantics. This topic describes the annotations supported by VCI and provides configuration examples.

Annotations supported for instance creation

Annotation key	Annotation sample value	Description	Required	Related documentation
vke.volcengine.com/burst-to-vci	enforce	Whether to forcibly deploy the pod on VCI. Valid values: enforce: Forcibly deploy the pod on VCI during its creation.	Required for VCI scenarios	Creating instances
vke.volcengine.com/preferred-subnet-ids	subnet-3tispp1nai****	Specify a subnet for the VCI. tip The default setting is the subnets specified during creation of the cluster. You can specify multiple subnets across multiple Availability Zones. Separate them with commas (,). If you specify multiple subnets, the system selects the subnet with the most available IP addresses to create the VCI.	no	Creating instances by specifying subnets
vke.volcengine.com/security-group-ids	sg-3thbinrkp94bwbh**, sg-2b49fbsd723bf4r**	Specify security groups for the VCI. To specify multiple security groups, separate them with commas (,). For security group quantity limits, see Security group limits. tip The default security group used is the pod security group, which is created by default during VKE cluster creation and allows all inbound and outbound traffic.	no	Creating instances by specifying security groups
vci.vke.volcengine.com/preferred-instance-types	vci.g1v.2xlarge	The instance specification of the VCI. You can specify multiple instance specifications (annotation values) with commas (,) as separators. Priority is determined by the order in which they are provided.	Required to specify specifications when creating an instance	Creating GPU-accelerated instances based on the specified VCI specification Creating general-purpose instances with specified VCI specifications
vci.vke.volcengine.com/kube-proxy-enabled	false	Proxy can be disabled to reduce overhead and accelerate startup. Valid values: false: Do not install the proxy. true: (Default) Install the proxy.	no	No
vci.vke.volcengine.com/preferred-instance-family	vci.u1	The instance family of the VCI. You can specify multiple instance families (annotation values) with commas (,) as separators. Priority is determined by the order in which they are provided. For available instance families, see VCI specifications.	General-purpose: This setting is not required. The default value is vci.u1. GPU-accelerated: This setting is required. No default value applies.	Creating general-purpose instances by specifying container vCPUs and memory Creating GPU-accelerated instances based on specified GPU resources
vci.vke.volcengine.com/desired-system-storage	"50"	Size of system volume storage. Value: a positive integer in string format. Unit: GiB. Maximum value: 500. Minimum value: 40. tip [Beta - apply for trial]: The feature indicated by the annotation is in the beta stage. Submit an application if you need to use it.	no	No
vci.vke.volcengine.com/stdlog-max-size	10Mi	The maximum capacity of container log files that can be stored in case of collecting standard output logs via a sidecar. Value range: 1MiB to 1GiB (that is, 1 MiB to 1024 MiB). The value format is number + binary unit. For example, 10Mi means 10 MiB and 1Gi means 1 GiB (1024 MiB). Default value: 10Mi.	no	Collecting VCI container logs by using a sidecar
vci.vke.volcengine.com/stdlog-max-files	5	The maximum number of container log files that can be stored in case of collecting standard output logs via a sidecar. Valid values: 2–100 (The value must be a positive integer). Default value: 5.	no	Collecting VCI container logs by using a sidecar
vci.vke.volcengine.com/stdlog-max-line-size	16384	Maximum length for a single line of container standard output when collecting logs via a sidecar. Value range: 4096 to 1048576 bytes. The value follows the number or number+binary unit format. For the number format, the unit is bytes. An example of the number+binary unit form is 100Ki, which means 100 KiB. Default value: 16384.	no	Collecting VCI container logs by using a sidecar
vci.volcengine.com/tls-enable	false	Whether to collect VCI logs to TLS. Valid values: false: (Default) Disable log collection. true: Enable log collection.	no	Collecting VCI container logs via TLS
vci.vke.volcengine.com/gpu-driver-version	tesla-470.129.06	Specify the driver version of the GPU-accelerated instance. Different GPU-accelerated specifications support different configurable driver versions. For details, see Driver versions for different specifications. tip If you do not specify this annotation, the system automatically specifies the default driver version that corresponds to the GPU-accelerated instance specification that you selected.	no	Creating GPU-accelerated instances based on specified container GPU resources Creating GPU-accelerated instances based on the specified VCI specification
vke.volcengine.com/eviction-grace-period-seconds	50	Grace period for pod eviction. The value is in seconds. It is an integer from 1 to 60.	no	No
vci.vke.volcengine.com/resource-tags	"cluster-id:ccq****,env:test"	Specify custom tags for the VCI to enable bill splitting of VCI resources by tag. The requirements for custom tags are as follows: The supported format is string. Each tag consists of a key-value pair in the key:value format. Multiple tags are separated by commas (,). For example, "cluster-id:**,env:test" means that two tags are set for the VCI: cluster-id:** and env:test. The requirements for a tag key are as follows: It must not start with volc: in any case form. It can only contain characters, digits, spaces, and special characters _.:/=+-@. The length is limited to 1 to 128 characters. The requirements for a tag value are as follows: This can be left blank. It can only contain characters, digits, spaces, and special characters _.:/=+-@. Its length does not exceed 256 characters. A single VCI supports up to 50 tags. Adding a tag with the same key as an existing tag key by using the annotation will overwrite the original tag value. Tag keys and values do not allow leading or trailing spaces.	no	Tag-based bill splitting
vci.vke.volcengine.com/project-name	default	The name of the project to which the VCI belongs.	no	Project-based bill splitting
vci.vke.volcengine.com/sysctls		Modify VCI kernel parameters. Valid values: The valid value type is Map[String]String. Only the vm.watermark_scale_factor parameter is supported: The kernel parameter watermark_scale_factor adjusts the differences between the min, low, and high values to handle situations where the service suddenly requests memory. For more information, see the kernel documentation. Default value: 10. A valid value is a positive integer from 1 to 1000.	no	Creating general-purpose instances by specifying container vCPUs and memory Creating GPU-accelerated instances based on specified container GPU resources
vci.volcengine.com/insecure-registries	"registry1,registry:5000,registry:80"	Specify container registries for which TLS authentication is skipped. If you specify multiple ones, separate them with commas (,). If the address of the image registry includes a port number, you also need to enter the port number.	no	Pulling images from self-built image registries during the creation of instances
vci.volcengine.com/plain-http-registries	"registry1-http,registry-http:5000,registry-http:80"	Specify HTTP-accessible image registries, separating multiple ones with commas (,). If the image registry address includes a port number, you also need to enter it.	no	Pulling images from self-built image registries during the creation of instances
vci.vke.volcengine.com/image-cache-id	imc-3uqroqkx0i3j********	Image cache ID specified for creating the VCI.	no	Creating VCIs by using manually created image caches
vci.vke.volcengine.com/ignore-image-cache-unavailable	false	During VCI creation, if the image cache identified by the image cache ID does not exist or is unavailable, the image cache configuration will be ignored. true: (Default) Ignore it. false: Do not ignore it. If the image cache with the corresponding ID does not exist or is unavailable, the VCI creation will fail.	no	Creating VCIs by using manually created image caches
vci.volcengine.com/gpu-compute-mode	exclusive	CUDA compute mode for the GPU. This setting is used for management of GPU device context creation and multi-thread or multi-process access. Supported options: exclusive: Only one CUDA context can be created on the GPU device across the entire system. However, within the process that creates this context, multiple threads can use the context. This ensures process-level exclusivity and prevents resource contention among multiple processes. tip This mode locks the process, making it suitable only for high-performance scientific computing scenarios. For general inference scenarios, it will cause significant performance degradation.	no	Creating GPU-accelerated instances
vci.vke.volcengine.com/kernel-transparent-hugepage	madvise	Whether the kernel uses Transparent Huge Pages. Valid values: never: Do not use transparent memory. always: Prioritize using transparent memory as much as possible. madvise: (Default) Avoid altering memory usage patterns.	no	No

Major annotations displayed after creating an instance

Annotation key	Annotation sample value	Description
vci.vke.volcengine.com/instance-id	vci-3todtpl3ng4dgfwz****	VCI ID.
vci.vke.volcengine.com/used-instance-types	vci.g1v.2xlarge	Instance specification of the VCI.
vke.volcengine.com/eni-id	eni-bp1fgnh68xyz9****	ID of the ENI for the VCI.
vke.volcengine.com/used-subnet-id	subnet-128fhnvjjgg****	Subnet of the VCI.
vci.vke.volcengine.com/vk-name	vci-node1-cn-beijing-a	The virtual node to which the VCI belongs.

Last updated: 2026.10.16 17:40:20