|
| 1 | +# Verda Cloud |
| 2 | +[Verda Cloud](https://verda.com/) is a Helsinki-based GPU/compute provider. The Verda provider in Claudie requires OAuth2 client credentials (`clientid` + `clientsecret`) in the Kubernetes Secret. An optional `baseurl` can override the default API endpoint. |
| 3 | + |
| 4 | +## Compute example |
| 5 | +```yaml |
| 6 | +apiVersion: v1 |
| 7 | +kind: Secret |
| 8 | +metadata: |
| 9 | + name: verda-secret |
| 10 | +data: |
| 11 | + clientid: <base64-encoded-client-id> |
| 12 | + clientsecret: <base64-encoded-client-secret> |
| 13 | + baseurl: <base64-encoded-base-url> # optional, defaults to https://api.verda.com/v1 |
| 14 | +type: Opaque |
| 15 | +``` |
| 16 | +
|
| 17 | +!!! note "No DNS support" |
| 18 | + Verda Cloud does not provide DNS resources. If you need load balancer DNS records for Verda clusters, use a separate DNS provider (e.g., Cloudflare, AWS Route53, GCP Cloud DNS). |
| 19 | +
|
| 20 | +!!! warning "Choose a non-Docker Ubuntu image" |
| 21 | + Verda offers Ubuntu images both with and without Docker pre-installed (e.g., `ubuntu-24.04-cuda-12.8-open-docker` vs `ubuntu-24.04-cuda-12.8-open`). Claudie uses KubeOne for cluster bootstrap, which installs and pins its own version of `containerd.io`. Pre-installed Docker on the image conflicts with this and causes apt-get to refuse the downgrade. **Always pick a Verda image without the `-docker` suffix.** Use `ubuntu-24.04` for non-GPU nodes and `ubuntu-24.04-cuda-12.8-open` (or similar) for GPU nodes. |
| 22 | + |
| 23 | +## Create Verda API credentials |
| 24 | + |
| 25 | +Generate OAuth2 client credentials from the [Verda console](https://console.verda.com/) under **Keys > Cloud API Credentials**. You receive a `client_id` and a `client_secret` (the secret is shown only once at creation time). |
| 26 | + |
| 27 | +The provider uses the OAuth2 client-credentials grant flow with scope `cloud-api-v1`. Claudie's Verda template embeds the credentials at apply time (the OpenTofu `verda` provider handles token exchange internally). |
| 28 | + |
| 29 | +## Available locations and instance types |
| 30 | + |
| 31 | +Locations and instance types can change. Query the live catalog with the credentials you just created: |
| 32 | + |
| 33 | +```bash |
| 34 | +# Get an OAuth bearer token |
| 35 | +TOKEN=$(curl -sS -X POST https://api.verda.com/v1/oauth2/token \ |
| 36 | + -d grant_type=client_credentials \ |
| 37 | + -d "client_id=<your-client-id>" \ |
| 38 | + -d "client_secret=<your-client-secret>" \ |
| 39 | + -d scope=cloud-api-v1 | jq -r .access_token) |
| 40 | +
|
| 41 | +# List instance types (CPU, RAM, GPU per type) |
| 42 | +curl -sS https://api.verda.com/v1/instance-types \ |
| 43 | + -H "Authorization: Bearer $TOKEN" -H "user-agent: " | jq |
| 44 | +
|
| 45 | +# List locations |
| 46 | +curl -sS https://api.verda.com/v1/locations \ |
| 47 | + -H "Authorization: Bearer $TOKEN" -H "user-agent: " | jq |
| 48 | +
|
| 49 | +# List images |
| 50 | +curl -sS https://api.verda.com/v1/images \ |
| 51 | + -H "Authorization: Bearer $TOKEN" -H "user-agent: " | jq |
| 52 | +
|
| 53 | +# Live availability per instance type and location |
| 54 | +curl -sS https://api.verda.com/v1/instance-availability \ |
| 55 | + -H "Authorization: Bearer $TOKEN" -H "user-agent: " | jq |
| 56 | +``` |
| 57 | + |
| 58 | +Common values at the time of writing: |
| 59 | + |
| 60 | +| Field | Examples | |
| 61 | +|---|---| |
| 62 | +| `region` | `FIN-01`, `FIN-02`, `FIN-03` (all in Helsinki) | |
| 63 | +| `serverType` | `CPU.4V.16G` (CPU only), `1B200.30V` (1× B200 GPU) | |
| 64 | +| `image` | `ubuntu-24.04`, `ubuntu-24.04-cuda-12.8-open`, `ubuntu-22.04-cuda-13.0-open` | |
| 65 | + |
| 66 | +!!! warning "Storage quota and trash" |
| 67 | + Verda's deleted volumes go to a per-account trash bin and continue to count toward your storage quota until permanently purged. If a Claudie apply fails after creating volumes, those volumes accumulate in the trash. Purge them with `DELETE /v1/volumes/<id>` and JSON body `{"is_permanent": true}` (HTTP 202 returned). See `samuelstolicny/claudie-config` Verda integration notes for a copy-paste purge one-liner. |
| 68 | + |
| 69 | +## GPU instances |
| 70 | +Verda specializes in GPU compute (A100, H100, B200, etc.). GPU instances are selected by setting the `serverType` to a GPU instance type (e.g., `1B200.30V` for 1× B200 with 30 vCPU). No additional `machineSpec` configuration is needed — the GPU is included in the instance type. Use the `*-cuda-*-open` (no Docker) image variants for GPU nodes; CUDA drivers are baked in. |
| 71 | + |
| 72 | +## Custom SSH port |
| 73 | + |
| 74 | +Like all Claudie-managed nodes, Verda VMs listen for SSH on **port 22522**. The cloud-init script reconfigures `sshd_config` and the `ssh.socket` listener accordingly during node provisioning. |
| 75 | + |
| 76 | +## Input manifest examples |
| 77 | + |
| 78 | +### Create a secret for Verda provider |
| 79 | +The secret for a Verda provider must include the following mandatory fields: `clientid`, `clientsecret`. The `baseurl` field is optional. |
| 80 | + |
| 81 | +```bash |
| 82 | +kubectl create secret generic verda-secret-1 \ |
| 83 | + --namespace=<your-namespace> \ |
| 84 | + --from-literal=clientid='<your-client-id>' \ |
| 85 | + --from-literal=clientsecret='<your-client-secret>' |
| 86 | +``` |
| 87 | + |
| 88 | +### Single-provider, single-location cluster example |
| 89 | + |
| 90 | +```yaml |
| 91 | +apiVersion: claudie.io/v1beta1 |
| 92 | +kind: InputManifest |
| 93 | +metadata: |
| 94 | + name: verda-example-manifest |
| 95 | + labels: |
| 96 | + app.kubernetes.io/part-of: claudie |
| 97 | +spec: |
| 98 | + providers: |
| 99 | + - name: verda-1 |
| 100 | + providerType: verda |
| 101 | + templates: |
| 102 | + repository: "https://github.com/berops/claudie-config" |
| 103 | + tag: v0.11.0 |
| 104 | + path: "templates/terraformer/verda" |
| 105 | + secretRef: |
| 106 | + name: verda-secret-1 |
| 107 | + namespace: <your-namespace> |
| 108 | +
|
| 109 | + nodePools: |
| 110 | + dynamic: |
| 111 | + - name: control-verda |
| 112 | + providerSpec: |
| 113 | + # Name of the provider instance. |
| 114 | + name: verda-1 |
| 115 | + # Location of the nodepool. |
| 116 | + region: FIN-01 |
| 117 | + count: 1 |
| 118 | + # Instance type name. |
| 119 | + serverType: CPU.4V.16G |
| 120 | + # OS image name (use a non-Docker variant). |
| 121 | + image: "ubuntu-24.04" |
| 122 | +
|
| 123 | + - name: compute-verda |
| 124 | + providerSpec: |
| 125 | + name: verda-1 |
| 126 | + region: FIN-01 |
| 127 | + count: 2 |
| 128 | + serverType: CPU.4V.16G |
| 129 | + image: "ubuntu-24.04" |
| 130 | + storageDiskSize: 50 |
| 131 | +
|
| 132 | + kubernetes: |
| 133 | + clusters: |
| 134 | + - name: verda-cluster |
| 135 | + version: "1.34.0" |
| 136 | + network: 192.168.2.0/24 |
| 137 | + pools: |
| 138 | + control: |
| 139 | + - control-verda |
| 140 | + compute: |
| 141 | + - compute-verda |
| 142 | +``` |
| 143 | + |
| 144 | +### GPU compute pool example |
| 145 | + |
| 146 | +```yaml |
| 147 | +apiVersion: claudie.io/v1beta1 |
| 148 | +kind: InputManifest |
| 149 | +metadata: |
| 150 | + name: verda-gpu-manifest |
| 151 | + labels: |
| 152 | + app.kubernetes.io/part-of: claudie |
| 153 | +spec: |
| 154 | + providers: |
| 155 | + - name: verda-1 |
| 156 | + providerType: verda |
| 157 | + templates: |
| 158 | + repository: "https://github.com/berops/claudie-config" |
| 159 | + tag: v0.11.0 |
| 160 | + path: "templates/terraformer/verda" |
| 161 | + secretRef: |
| 162 | + name: verda-secret-1 |
| 163 | + namespace: <your-namespace> |
| 164 | +
|
| 165 | + nodePools: |
| 166 | + dynamic: |
| 167 | + - name: control-verda |
| 168 | + providerSpec: |
| 169 | + name: verda-1 |
| 170 | + region: FIN-01 |
| 171 | + count: 1 |
| 172 | + serverType: CPU.4V.16G |
| 173 | + image: "ubuntu-24.04" |
| 174 | +
|
| 175 | + - name: gpu-verda |
| 176 | + providerSpec: |
| 177 | + name: verda-1 |
| 178 | + region: FIN-01 |
| 179 | + count: 1 |
| 180 | + # GPU instance type, e.g. 1× B200. |
| 181 | + serverType: 1B200.30V |
| 182 | + # CUDA image without Docker (KubeOne installs its own containerd). |
| 183 | + image: "ubuntu-24.04-cuda-12.8-open" |
| 184 | + storageDiskSize: 100 |
| 185 | +
|
| 186 | + kubernetes: |
| 187 | + clusters: |
| 188 | + - name: verda-gpu-cluster |
| 189 | + version: "1.34.0" |
| 190 | + network: 192.168.2.0/24 |
| 191 | + pools: |
| 192 | + control: |
| 193 | + - control-verda |
| 194 | + compute: |
| 195 | + - gpu-verda |
| 196 | +``` |
| 197 | + |
| 198 | +### Autoscaling cluster example |
| 199 | + |
| 200 | +```yaml |
| 201 | +apiVersion: claudie.io/v1beta1 |
| 202 | +kind: InputManifest |
| 203 | +metadata: |
| 204 | + name: verda-autoscale-manifest |
| 205 | + labels: |
| 206 | + app.kubernetes.io/part-of: claudie |
| 207 | +spec: |
| 208 | + providers: |
| 209 | + - name: verda-1 |
| 210 | + providerType: verda |
| 211 | + templates: |
| 212 | + repository: "https://github.com/berops/claudie-config" |
| 213 | + tag: v0.11.0 |
| 214 | + path: "templates/terraformer/verda" |
| 215 | + secretRef: |
| 216 | + name: verda-secret-1 |
| 217 | + namespace: <your-namespace> |
| 218 | +
|
| 219 | + nodePools: |
| 220 | + dynamic: |
| 221 | + - name: control-verda |
| 222 | + providerSpec: |
| 223 | + name: verda-1 |
| 224 | + region: FIN-01 |
| 225 | + count: 1 |
| 226 | + serverType: CPU.4V.16G |
| 227 | + image: "ubuntu-24.04" |
| 228 | +
|
| 229 | + - name: gpu-verda |
| 230 | + providerSpec: |
| 231 | + name: verda-1 |
| 232 | + region: FIN-01 |
| 233 | + # Autoscaler configuration (mutually exclusive with count). |
| 234 | + autoscaler: |
| 235 | + min: 0 |
| 236 | + max: 3 |
| 237 | + serverType: 1B200.30V |
| 238 | + image: "ubuntu-24.04-cuda-12.8-open" |
| 239 | + storageDiskSize: 100 |
| 240 | +
|
| 241 | + kubernetes: |
| 242 | + clusters: |
| 243 | + - name: verda-cluster |
| 244 | + version: "1.34.0" |
| 245 | + network: 192.168.2.0/24 |
| 246 | + pools: |
| 247 | + control: |
| 248 | + - control-verda |
| 249 | + compute: |
| 250 | + - gpu-verda |
| 251 | +``` |
0 commit comments