SDL Syntax Reference | Akash Network - Your Guide to Decentralized Cloud

Version

Required. SDL version identifier.

version: "2.0"

Value	Description
`"2.0"`	Current stable version

Services

Define containers and networking.

image

Required. Docker image to deploy.

image: nginx:1.25.3

Supports any public registry (Docker Hub, GHCR, GCR, ECR, etc.)

Best Practice: Always specify a version tag (e.g., nginx:1.25.3) instead of using :latest or omitting the tag. This ensures reproducible deployments and prevents unexpected updates.

command & args

Optional. Override container entrypoint.

command:
  - /bin/sh
  - -c
args:
  - |
    npm start

env

Optional. Environment variables as KEY=value strings.

env:
  - NODE_ENV=production
  - DATABASE_URL=postgresql://db:5432/myapp

expose

Required for individual services. And at least one service in the deployment must have global: true in its expose configuration. Services with only internal endpoints (e.g., database backends) are accessible only to other services within the deployment.

expose:
  - port: 80              # Container port (required)
    as: 80                # External port (default: same as port)
    proto: tcp            # Protocol: tcp or udp (default: tcp)
    accept:               # Accepted hostnames
      - example.com
    to:
      - global: true      # Expose to internet (at least one required)
    http_options:         # HTTP-specific options
      max_body_size: 1048576
      read_timeout: 60000
      send_timeout: 60000

HTTP Options:

Option	Default	Max	Description
`max_body_size`	1048576	104857600	Max request body (bytes)
`read_timeout`	60000	60000	Read timeout (ms)
`send_timeout`	60000	60000	Send timeout (ms)
`next_tries`	3	-	Retry attempts
`next_cases`	`[error, timeout]`	-	When to retry

credentials

Optional. Private registry authentication.

credentials:
  host: https://registry.example.com
  username: user
  password: pass

params

Optional. Service-level runtime parameters. Groups storage mounts and inter-service RBAC permissions.

params:
  permissions:
    read:
      - logs
      - events

params.storage covers persistent-volume mounts and is documented under advanced features.

permissions

Optional. Grant a sibling service in the same deployment RBAC access to read this service’s runtime data. Used by log-collector and observability sidecars so they can tail logs and events from the target service’s pods.

params:
  permissions:
    read:
      - logs        # pods, pods/log, deployments — read-only
      - events      # kubernetes events — read-only

Scope	RBAC granted
`logs`	`get/list/watch` on `pods`, `get` on `pods/log`, `get/list/watch` on `apps/deployments`
`events`	`get/list/watch` on `events`

The collector pod authenticates against the Kubernetes API using a ServiceAccount the provider mounts into every container. When the target service declares permissions.read, the provider attaches a matching Role and RoleBinding so collectors in the same deployment can call the API server. See Log Forwarding for the full sidecar pattern.

Profiles

Define resources and provider requirements.

compute

cpu

Required. CPU units (1000 = 1 core).

cpu:
  units: 1.0              # Cores (0.1, 0.5, 1.0, etc.)

Limits: Min: 0.01 cores, Max: 384 cores

memory

Required. Memory allocation.

memory:
  size: 1Gi               # RAM (Mi or Gi)

Limits: Min: 1Mi, Max: 2Ti

storage

Required. At least one ephemeral storage volume required. Persistent storage is optional.

storage:
  - size: 1Gi             # Ephemeral root volume (required)
  - name: data            # Persistent volume (optional)
    size: 10Gi
    attributes:
      persistent: true    # Survives restarts
      class: beta3        # Performance tier

Limits: Min: 5Mi, Max: 32Ti per volume

Classes: default, beta1, beta2, beta3, ram

gpu

Optional. GPU resources.

gpu:
  units: 1                # Number of GPUs
  attributes:
    vendor:
      nvidia:
        - model: rtx4090
          ram: 24Gi       # Optional: filter by VRAM
          interface: pcie # Optional: pcie or sxm

Limits: Max: 24 GPUs per service

Common models: a100, a40, rtx4090, rtx3090, t4, v100, h100

placement

Provider selection and pricing.

placement:
  akash:
    attributes:           # Optional filters
      region: us-west
    pricing:              # Required: max bid per service
      web:
        denom: uact
        amount: 100       # uakt per block (~6 sec)

Pricing calculation:

Daily cost = (amount × 14,400 blocks) / 1,000,000 AKT
Example: 100 uakt/block = 1.44 AKT/day = ~43 AKT/month

Deployment

Map services to profiles.

deployment:
  web:                    # Service name
    akash:                # Placement name
      profile: web        # Compute profile name
      count: 1            # Number of replicas

Note: count cannot be changed after deployment. Max: 50 replicas per service.

Endpoints

Optional. Named IP endpoints for static IPs.

endpoints:
  myendpoint:
    kind: ip

Use in services:

expose:
  - port: 1194
    to:
      - global: true
        ip: myendpoint

Quick Reference

Minimum viable SDL:

version: "2.0"
services:
  web:
    image: nginx:1.25.3
    expose:
      - port: 80
        as: 80
        to:
          - global: true
profiles:
  compute:
    web:
      resources:
        cpu:
          units: 0.5
        memory:
          size: 512Mi
        storage:
          size: 512Mi
  placement:
    akash:
      pricing:
        web:
          denom: uact
          amount: 100
deployment:
  web:
    akash:
      profile: web
      count: 1

Common patterns:

GPU deployments
Persistent storage
Private registries
Static IPs

Resource Limits (per service):

Resource	Minimum	Maximum
CPU	0.01 cores	384 cores
Memory	1Mi	2Ti
Storage	5Mi	32Ti (per volume)
GPU	0 (optional)	24 units
Replicas (count)	1	50

See also:

Complete SDL Example

1---
2version: "2.0"
3
4services:
5  web:
6    image: nginx:1.25.3
7    command:
8      - /bin/sh
9      - -c
10    args:
11      - nginx -g 'daemon off;'
12    env:
13      - NODE_ENV=production
14      - PORT=3000
15    expose:
16      - port: 80
17        as: 80
18        proto: tcp
19        accept:
20          - example.com
21        to:
22          - global: true
23        http_options:
24          max_body_size: 1048576
25          read_timeout: 60000
26          send_timeout: 60000
27    credentials:
28      host: https://registry.example.com
29      username: user
30      password: pass
31    params:
32      permissions:
33        read:
34          - logs
35          - events
36
37profiles:
38  compute:
39    web:
40      resources:
41        cpu:
42          units: 1.0
43        memory:
44          size: 1Gi
45        storage:
46          - size: 1Gi
47          - name: data
48            size: 10Gi
49            attributes:
50              persistent: true
51              class: beta3
52        gpu:
53          units: 1
54          attributes:
55            vendor:
56              nvidia:
57                - model: rtx4090
58  
59  placement:
60    akash:
61      attributes:
62        region: us-west
63      pricing:
64        web:
65          denom: uact
66          amount: 100
67
68deployment:
69  web:
70    akash:
71      profile: web
72      count: 1
73
74endpoints:
75  myendpoint:
76    kind: ip