# Deployments

Deploy your models to production with auto-scaling and load balancing.

![Deployments View](/files/A2f3MmKxrjL07odwiYh5)

## Creating a Deployment

Navigate to **Deployments** → Click **Create**

![Create Deployment Form](/files/p3aND1HlZGKzeo5JpfUo)

### Basic Information

**Deployment Name**\* (Required)

* Enter a descriptive name for the deployment
* Example: `resnet-prod`, `bert-api-v1`

**Description** (Optional)

* Deployment purpose and details

**Model ID**\* (Required)

* ID of the model to deploy
* Helper text: "ID of the model to deploy"

**Model Version**\* (Required)

* Version of the model to deploy
* Helper text: "Version of the model to deploy"

**Environment**\* (Required)

* Select deployment environment:
  * Development
  * Staging
  * Production
* Default: `development`

### Resource Configuration

**CPU Cores**\* (Required)

* Number of CPU cores per instance
* Example: `4`, `8`, `16`

**Memory (GB)**\* (Required)

* Memory allocation per instance in GB
* Example: `8`, `16`, `32`

**GPU Count** (Optional)

* Number of GPUs (0 or GPU Count)
* Default: `0`

**GPU Type** (Optional)

* Select GPU type if GPU Count > 0:
  * NVIDIA T4
  * NVIDIA V100
  * NVIDIA A100

**Min Replicas**\* (Required)

* Minimum number of instances
* Example: `1`, `2`

**Max Replicas**\* (Required)

* Maximum number of instances
* Example: `10`, `20`

**Target CPU Utilization (%)**\* (Required)

* CPU threshold for scaling
* Example: `70`, `80`

**Target Memory Utilization (%)**\* (Required)

* Memory threshold for scaling
* Example: `80`, `90`

### Scaling Configuration

**Enable Auto-Scaling** (Checkbox)

* Enable automatic scaling based on metrics

When enabled:

* **Min Instances**\*: Minimum instances to maintain
* **Max Instances**\*: Maximum instances allowed
* **Target GPU Utilization (%)**: GPU threshold
* **Target Memory Utilization (%)**: Memory threshold

### Load Balancer

**Enable Load Balancer** (Checkbox)

* Enable load balancing across instances

When enabled:

* **Service Type**\*: Round Robin, Least Connections, IP Hash
* **Health Check URL**\*: Endpoint for health checks (e.g., `/health`)
* **Health Check Interval (seconds)**\*: Frequency of health checks
* **Sticky Sessions**: Enable session affinity

### Actions

* **Cancel**: Discard and close
* **Create Deployment**: Submit and create the deployment

## Example Configuration

```yaml
Deployment Name: resnet-production-v1
Model ID: model_abc123
Model Version: 1.0.0
Environment: Development

Resource Configuration:
  CPU Cores: 4
  Memory (GB): 16
  GPU Count: 0
  Min Replicas: 2
  Max Replicas: 10

Scaling:
  Enable Auto-Scaling: ✓
  Min Instances: 2
  Max Instances: 10
  Target CPU: 70%

Load Balancer:
  Enable: ✓
  Service Type: Round Robin
  Health Check URL: /health
  Health Check Interval: 30s
  Sticky Sessions: ✓
```

## Viewing Deployment Details

To view detailed information about a deployment:

1. Navigate to **Deployments**
2. Click on a deployment from the list
3. View comprehensive details in the modal dialog

![View Deployment Details](/files/0zrf97sTHRzJsvf72BQ1)

**Details Panel Sections**:

**Basic Information**:

* **Deployment Name**: e.g., "BERT Sentiment API"
* **Description**: Full description of the deployment
* **Model ID**: ID of the deployed model (e.g., "model-001")
* **Model Version**: Version being deployed (e.g., "v2.1.0")
* **Environment**: Production, Staging, or Development
* **Deployment Strategy**: Blue-Green, Rolling Update, Canary

**Resource Configuration**:

* **CPU Cores**: Number of CPU cores per instance (e.g., 4)
* **Memory (GB)**: Memory allocation (e.g., 16 GB)
* **GPU Count**: Number of GPUs (e.g., 2)
* **GPU Type**: GPU model (e.g., A100, V100)
* **Storage (GB)**: Storage allocation (e.g., 200 GB)
* **Network Bandwidth (Mbps)**: Network bandwidth (e.g., 1000 Mbps)

**Scaling Configuration**:

* **Current Instances**: Number of running instances (e.g., 2)
* **Enable Auto-Scaling**: Checkbox status
* **Min Instances**: Minimum replicas (e.g., 2)
* **Max Instances**: Maximum replicas (e.g., 10)
* **Target CPU Utilization (%)**: CPU scaling threshold (e.g., 70%)
* **Target GPU usage for trigger scaling**: GPU threshold
* **Target Memory Utilization (%)**: Memory threshold (e.g., 80%)
* **Target memory usage for trigger scaling**: Memory threshold

**Load Balancer**:

* **Enable Load Balancer**: Checkbox status
* **Load Balancing Algorithm**: Round Robin, Least Connections, IP Hash
* **Health Check Path**: Endpoint for health checks (e.g., "/health")
* **Health Check Interval (seconds)**: Check frequency (e.g., 30)
* **Sticky Sessions**: Checkbox for session affinity

## Editing a Deployment

To update deployment configuration:

1. Open deployment details page
2. Click **Edit** button (or three-dot menu → Edit)
3. Modify editable fields in the Edit Deployment modal

![Edit Deployment Form](/files/0yjUgJzuowLKvaedEEb6)

4. Click **Update Deployment** to save changes

> \[!NOTE] The Edit form is identical to the View form, but with editable fields and an "Update Deployment" button. Some changes may require a deployment restart.

> \[!NOTE] Some changes may require a deployment restart to take effect.

**Editable Fields**:

* ✅ Description
* ✅ Environment variables
* ✅ Min/Max replicas
* ✅ Auto-scaling thresholds
* ✅ Health check settings
* ✅ Load balancer configuration
* ⚠️ CPU/Memory (requires restart)
* ❌ Model ID (use Update Model instead)
* ❌ Deployment name (cannot edit)

## Updating Model Version

To deploy a new model version:

1. Open deployment details
2. Click **Update Model** button
3. Select new model version
4. Choose update strategy:
   * **Rolling Update**: Gradual replacement (zero downtime)
   * **Blue-Green**: Switch all at once
   * **Canary**: Test with small percentage first
5. Click **Update**

**Update Strategies**:

**Rolling Update** (Recommended):

* Gradually replaces old instances
* Zero downtime
* Automatic rollback on failure

**Blue-Green**:

* Deploys new version alongside old
* Switches traffic all at once
* Quick rollback possible

**Canary**:

* Routes small % of traffic to new version
* Monitor performance
* Gradually increase if successful

## Scaling a Deployment

**Manual Scaling**:

1. Open deployment details
2. Click **Scale** button
3. Adjust number of replicas
4. Click **Apply**

**Auto-scaling**:

1. Open deployment details
2. Click **Edit**
3. Enable auto-scaling
4. Set min/max replicas
5. Configure scaling triggers
6. Save changes

## Stopping a Deployment

To temporarily stop a deployment:

1. Open deployment details
2. Click **Stop** button
3. Confirm action
4. All instances will shut down
5. Endpoint will become unavailable

**Use Cases**:

* Maintenance window
* Cost optimization
* Testing in isolation

## Restarting a Deployment

To restart a stopped deployment:

1. Open deployment details
2. Click **Start** button
3. Deployment will resume with previous configuration

## Deleting a Deployment

To permanently remove a deployment:

1. Navigate to deployment details
2. Click **Delete** button
3. Confirm deletion

> \[!WARNING] Deleting a deployment will:
>
> * Shut down all instances
> * Remove the endpoint
> * Delete deployment configuration
> * This action cannot be undone!

**Before Deleting**:

* Stop sending traffic to the endpoint
* Update client applications
* Export logs if needed
* Verify you have the correct deployment

## Monitoring Deployments

**Real-time Metrics**:

* Request rate
* Latency (p50, p95, p99)
* Error rate
* Resource usage

**Actions**:

* Scale up/down
* Update model version
* View logs
* Rollback

## Best Practices

* Set appropriate min/max replicas
* Configure auto-scaling thresholds
* Enable health checks
* Use load balancing for high traffic
* Monitor performance continuously

## Next Steps

* Monitor in [Analytics](/kaisar-network/kaisar-ai-ops/deep-learning-platform/analytics.md)
* View logs and metrics
* Set up alerts


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.kaisar.io/kaisar-network/kaisar-ai-ops/deep-learning-platform/deployments.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.