Deployment Config
DeploymentConfig
Defines the runtime configuration of an application for one environment.
It overrides settings of an Application manifest, for example resource limits and volume mounts.
| Property | Type | Description |
|---|---|---|
| apiVersion | string | API schema version. Must be lowops.manifests.v1. |
| kind | string | Resource kind. Must be DeploymentConfig. |
| metadata | map | The engine reads the reserved keys name and project to resolve this manifest and its parent Project. |
| spec | Spec | Desired configuration. See Spec. |
BucketMountsEntry
| Property | Type | Description |
|---|---|---|
| key | string | |
| value | VolumeMountBucketRef |
EventArcNotificationsEntry
| Property | Type | Description |
|---|---|---|
| key | string | |
| value | string |
PortsEntry
| Property | Type | Description |
|---|---|---|
| key | int64 | |
| value | string |
VolumeMountsEntry
| Property | Type | Description |
|---|---|---|
| key | string | |
| value | ContainerVolumeMount |
Spec
User-provided deployment overrides.
Environment-specific settings that the engine layers on the shared Application definition.
| Property | Type | Description |
|---|---|---|
| description | string | Optional free text. The engine copies it to the compute resource. AI assistants use it as context. |
| deleteBeforeCreate | bool | NOT YET IMPLEMENTED. The engine ignores this field and always does rolling updates. It will remove the old revision first. |
| stateful | bool | NOT YET IMPLEMENTED. The engine ignores this field. It will signal that the application needs durable storage. For VMs, use the stateful block of VirtualMachine. |
| container | ContainerSpec | Overrides for the main container (command, args, environment variables, resource limits and requests). They merge over the BuildDefinition or ExternalApplication definition. |
| ports | list of PortsEntry | Map of port number (1-65535) to protocol (http, grpc or tcp). Each entry creates a backend endpoint and a mesh route. |
| health | Health | Health probe settings. If unset, the platform default applies. See Health. |
| volumes | list of VolumesEntry | Map of container mount path to volume source, for example a bucket-backed volume. Validation rejects mount paths under /tmp. |
| scaling | ScalingConfig | Min and max instance count. See ScalingConfig. |
| version | string | Image tag to deploy. The container also gets it in the INFRASTREAM_APP_VERSION environment variable. Required. |
| requestTimeout | int64 | Request timeout in seconds, from 0 to 3600. Applies only if greater than 0, otherwise the platform default applies (Cloud Run: 300). Kubernetes and Compute set the backend service timeout. |
| terminationGracePeriodSeconds | int64 | Seconds, from 0 to 3600, between SIGTERM and the forced stop of the container. Kubernetes sets Pod.spec.terminationGracePeriodSeconds. Compute sets the TERMINATION_GRACE_PERIOD_SECONDS environment variable. If 0, the platform default applies. |
| gpu | DeploymentGpuConfig | Attaches one GPU to the application container. Only for target: CLOUD_RUN. Validation rejects this block on any other target, even an empty block. container.resources.limits must meet the CPU and memory minimum of the accelerator. The region of the parent Project must serve it. |
| invokers | list of string | Applications allowed to call this service directly. Cloud Run only. If empty (default), invoker IAM is off. Only the ingress setting limits access, so an internal service accepts calls from any workload in the VPC. |
| allowDirectIngress | bool | NOT RECOMMENDED for general use. If true, an HttpRoute with ingress_path DIRECT can send traffic to this service. The default is false. The service declares that it checks every caller itself. A DIRECT route skips the mesh gateway, so the mesh does not check the caller. |
Health
Container health probes.
Startup and liveness probes gate the deployment and detect hung containers.
| Property | Type | Description |
|---|---|---|
| startup | HostHealthConfig | NOT YET IMPLEMENTED. The engine ignores this field. The startup probe comes from liveness. |
| liveness | HostHealthConfig | Protocol, port, path and timing settings. The engine uses them for the startup probe and the liveness probe. |
PortsEntry
| Property | Type | Description |
|---|---|---|
| key | int64 | |
| value | string |
ScalingConfig
Instance autoscaling bounds.
Lower and upper limits on the number of instances.
| Property | Type | Description |
|---|---|---|
| min | int64 | Minimum number of warm instances. Applies only if greater than 0. A value of 0 allows scale to zero. |
| max | int64 | Maximum number of instances. Applies only if greater than 0. |
| hasBackgroundOperations | bool | If true, the service does background work outside requests, for example timers or queue consumers. CPU stays always allocated, so request-based (CPU-idle) billing does not apply. Default: false (CPU throttled between requests). |
| maxConcurrency | int64 | Maximum concurrent requests for one instance. Applies only if greater than 0, otherwise the platform default (80) applies. Ignored on Kubernetes and Compute targets. |
VolumesEntry
| Property | Type | Description |
|---|---|---|
| key | string | |
| value | ContainerVolumeMount |
DeploymentGpuConfig
GPU accelerator for the application container of a deployment.
Cloud Run attaches at most one GPU to one container, so the block has no count or container selector. Validation rejects the block on targets other than CLOUD_RUN, even when accelerator is empty.
| Property | Type | Description |
|---|---|---|
| accelerator | string | Accelerator model, nvidia-l4 or nvidia-rtx-pro-6000. Required when this block is present. nvidia-l4 has 24 GB VRAM and needs at least 4 CPU and 16Gi memory. nvidia-rtx-pro-6000 (RTX PRO 6000 Blackwell) has 96 GB VRAM and needs at least 20 CPU and 80Gi memory. The container resources.limits.cpu and resources.limits.memory must meet the minimum. Validation rejects a smaller container and does not raise the values. nvidia-l4 must also fit the Cloud Run matrix: 4, 6 or 8 CPU with at most 16Gi, 24Gi or 32Gi of memory. For nvidia-rtx-pro-6000, validation enforces only the minimum and a whole number of vCPU. |
| zonalRedundancyDisabled | bool | Opt out of GPU zonal redundancy. Zonal redundancy reserves GPU capacity in a second zone for zonal outages. It is the Cloud Run default and costs more. If true, you pay less but get a weaker SLA and no protection from a zonal outage. |
HostHealthConfig
Health check for host endpoints.
Sets the probes and backend health checks for GCP load balancers.
| Property | Type | Description |
|---|---|---|
| protocol | string | Health check protocol: http, http2, https, grpc, or tcp. It applies to container probes (Cloud Run, GKE) and the VM backend health check. |
| port | int64 | TCP port that the health check targets. |
| checkIntervalSec | int64 | Interval between checks, in seconds. The default is 30 for generated probes. Default 30 |
| timeoutSec | int64 | Seconds to wait for a probe response before it fails. The engine sets 10 for generated probes. Default 5 |
| healthyThreshold | int64 | Consecutive successful probes that mark the endpoint healthy. The default is 1. Default 1 |
| unhealthyThreshold | int64 | Consecutive failed probes that mark the endpoint unhealthy. The default is 2. Default 2 |
| enableLogs | bool | NOT YET IMPLEMENTED. The engine ignores this field. Intended to enable health-check logs. |
| path | string | HTTP request path for HTTP-family probes. The default is /. Default "/" |
ContainerSpec
Core configuration block for a runtime container.
It overrides the entrypoint, variables and compute constraints of the base image.
| Property | Type | Description |
|---|---|---|
| enabled | bool | NOT YET IMPLEMENTED. The engine ignores this field. Intended use: turn the container off but keep its definition. |
| command | list of string | Container entrypoint. Overrides the image entrypoint (the container command). |
| args | list of string | Arguments for the entrypoint (the container args). |
| env | list of EnvVariableDefinition | Static environment variables. The engine merges them with variables from secrets and access control. |
| uid | string | NOT YET IMPLEMENTED. The engine ignores this field. Intended use: the user ID of the container process. |
| gid | string | NOT YET IMPLEMENTED. The engine ignores this field. Intended use: the group ID of the container process. |
| resources | ContainerResources | CPU and memory requests and limits. See ContainerResources. |
ContainerVolumeMount
Generated files for a container volume.
The files typically become config maps or secrets on Kubernetes, or files on Cloud Run.
| Property | Type | Description |
|---|---|---|
| files | list of FilesEntry | Map of relative file path to file content. Each entry becomes a file in the volume. |
VolumeMountBucketRef
Bucket mounted as a volume.
Describes the Cloud Storage FUSE mount of a bucket.
| Property | Type | Description |
|---|---|---|
| name | string | Name of the Bucket manifest to mount with Cloud Storage FUSE. |
| path | string | Path where the container mounts the bucket. |
| canWrite | bool | If true, the mount is read-write and the service account of the container gets object-write permission on the bucket. If false, the mount is read-only. |
FilesEntry
| Property | Type | Description |
|---|---|---|
| key | string | |
| value | VolumeMountFile |
EnvVariableDefinition
Static environment variable.
The engine adds it to the environment of the container.
| Property | Type | Description |
|---|---|---|
| name | string | Variable name. |
| value | string | Literal value. |
ContainerResources
Compute requirements and limits of a container.
They map to the resource requests and limits of Kubernetes pods or Cloud Run services.
| Property | Type | Description |
|---|---|---|
| requests | ContainerResource | Minimum CPU and memory guaranteed to the container. |
| limits | ContainerResource | Maximum CPU and memory that the container can use. |
VolumeMountFile
Generated file in a volume mount.
Startup scripts write the file to disk.
| Property | Type | Description |
|---|---|---|
| mimeType | string | MIME type of the file. If empty, the engine sets a default. |
| content | string | File content: plain text, or base64 if base64_encoded is true. |
| base64Encoded | bool | If true, the engine decodes content from base64 before it writes the file. |
ContainerResource
Single resource boundary for a container.
It sets the requests or the limits for CPU and memory.
| Property | Type | Description |
|---|---|---|
| cpu | string | CPU quantity in Kubernetes/Cloud Run notation, for example "500m" or "2". |
| memory | string | Memory quantity in the same notation, for example "256Mi" or "1Gi". |