---

name: netgraph-dsl description: > NetGraph scenario DSL for defining network topologies, traffic demands, failure policies, and analysis workflows in YAML. Use when: creating or editing .yaml/.yml network scenarios, defining nodes/links/groups, writing link rules with patterns, configuring selectors or blueprints, setting up traffic demands or failure policies, running scenarios and interpreting results, debugging DSL syntax or validation errors, or asking about NetGraph scenario structure.

NetGraph DSL

Define network simulation scenarios in YAML format.

Quick Start: See Minimal Example below. Complete Reference: See references/REFERENCE.md for full documentation.

Instructions

When working with NetGraph scenarios:

1. Creating new scenarios: Start with the Minimal Example, then add sections as needed
2. Editing existing scenarios: Identify the relevant section (network, demands, failures, etc.)
3. Understanding selection: Review Selection Models to understand path-based vs condition-based selection
4. Debugging issues: Check Common Pitfalls and Validation Checklist
5. Complex topologies: Use Blueprints for reusable patterns
6. Failure simulation: Define Risk Groups before creating failure policies

Refer to specific sections below for detailed syntax and examples.

Quick Reference

Minimal Example

Copynetwork:
  nodes:
    A: {}
    B: {}
  links:
    - source: A
      target: B
      capacity: 100
      cost: 1

Core Patterns

Selection Models

The DSL implements two distinct selection patterns:

1. Path-based Node Selection (link rules, traffic demands, workflow steps)

• Uses regex patterns on hierarchical node names
• Supports capture group-based grouping
• Supports attribute-based grouping (group_by)
• Supports attribute filtering (match conditions)
• Supports active_only filtering

2. Condition-based Entity Selection (failure rules, membership rules, risk group generation)

• Works on nodes/links; failure + membership can also target risk_groups
• Optional regex path filter on entity names/IDs (no capture grouping)
• Attribute filtering via match.conditions (match.logic defaults vary by context)

These patterns share common primitives (condition evaluation, match specification) but serve different purposes and should not be confused.

For comprehensive details on entity creation flows, processing steps, and comparison tables, see the Entity Creation Architecture section in the full reference.

Nodes and Links

Copynetwork:
  nodes:
    Seattle:
      attrs:           # Custom attributes go here
        role: core
      risk_groups: ["RG1"]
      disabled: false
    Portland:
      attrs:
        role: edge

  links:
    - source: Seattle
      target: Portland
      capacity: 100    # Direct properties (no wrapper)
      cost: 10
      attrs:
        distance_km: 280
      count: 2         # Parallel links

Node Groups

Copynetwork:
  nodes:
    leaf:
      count: 4
      template: "leaf-{n}"
      attrs:
        role: leaf

Creates: leaf/leaf-1, leaf/leaf-2, leaf/leaf-3, leaf/leaf-4

Template Syntaxes

These are NOT interchangeable. See REFERENCE.md for details.

Bracket Expansion

Copynetwork:
  nodes:
    dc[1-3]/rack[a,b]:    # Cartesian product
      count: 4
      template: "srv-{n}"

Creates: dc1/racka, dc1/rackb, dc2/racka, dc2/rackb, dc3/racka, dc3/rackb

Scope: Bracket expansion works in group names, risk group definitions (including children), and risk group membership arrays. Component names and other fields treat brackets as literal characters.

Link Patterns

Copynetwork:
  links:
    - source: /leaf
      target: /spine
      pattern: mesh        # Every source to every target
      capacity: 100

    - source: /group_a     # 4 nodes
      target: /group_b     # 2 nodes
      pattern: one_to_one  # Pairwise with modulo wrap (sizes must have multiple factor)

Selectors with Conditions

Copynetwork:
  links:
    - source:
        path: "/datacenter"
        match:
          logic: and         # "and" or "or"; defaults vary by context (see below)
          conditions:
            - attr: role
              op: "=="
              value: leaf
      target: /spine
      pattern: mesh

Operators: ==, !=, <, <=, >, >=, contains, not_contains, in, not_in, exists, not_exists

Logic defaults by context (for match.logic):

Capturing Groups for Grouping

Copy# Single capture group creates groups by captured value
source: "^(dc[1-3])/.*"     # Groups: dc1, dc2, dc3

# Multiple capture groups join with |
source: "^(dc\\d+)/(spine|leaf)/.*"  # Groups: dc1|spine, dc1|leaf, etc.

Variable Expansion

Copylinks:
  - source: "plane${p}/rack"
    target: "spine${s}"
    expand:
      vars:
        p: [1, 2]
        s: [1, 2, 3]
      mode: cartesian  # or "zip" (equal-length lists required)
    pattern: mesh

Expansion limit: Maximum 10,000 expansions per template. Exceeding this raises an error.

Blueprints

Copyblueprints:
  clos_pod:
    nodes:
      leaf:
        count: 4
        template: "leaf-{n}"
      spine:
        count: 2
        template: "spine-{n}"
    links:
      - source: /leaf
        target: /spine
        pattern: mesh
        capacity: 100

network:
  nodes:
    pod[1-2]:
      blueprint: clos_pod
      params:
        leaf.count: 6  # Override defaults

Alternative: Inline nested nodes (no blueprint needed):

Copynetwork:
  nodes:
    datacenter:
      nodes:              # Inline hierarchy
        rack1:
          count: 2
          template: "srv-{n}"

Node and Link Rules

Modify entities after creation with optional attribute filtering:

Copynetwork:
  node_rules:
    - path: "^pod1/.*"
      match:                      # Optional: filter by attributes
        conditions:
          - {attr: role, op: "==", value: compute}
      disabled: true

  link_rules:
    - source: "^pod1/.*"
      target: "^pod2/.*"
      link_match:                 # Optional: filter by link attributes
        conditions:
          - {attr: capacity, op: ">=", value: 400}
      cost: 99

Rules also support expand for variable-based application.

Traffic Demands

Copydemands:
  production:
    - source: "^dc1/.*"
      target: "^dc2/.*"
      volume: 1000
      mode: pairwise       # or "combine"
      flow_policy: SHORTEST_PATHS_ECMP

Flow policies: SHORTEST_PATHS_ECMP, SHORTEST_PATHS_WCMP, TE_WCMP_UNLIM, TE_ECMP_16_LSP, TE_ECMP_UP_TO_256_LSP

Failure Policies

Copyfailures:
  single_link:
    expand_groups: false         # Expand to shared-risk entities
    modes:                       # Weighted modes (one selected per iteration)
      - weight: 1.0
        rules:
          - scope: link          # Required: node, link, or risk_group
            mode: choice         # all, choice, or random
            count: 1
            # Optional: weight_by: capacity  # Weighted sampling by attribute

Rule modes: all (select all matches), choice (sample count), random (each with probability)

Risk Groups

Risk groups model failure correlation (shared infrastructure, geographic regions, vendor dependencies, or any custom domain). Three methods:

Direct definition:

Copyrisk_groups:
  - name: "RG1"              # Full form
  - "RG2"                    # String shorthand (equivalent to {name: "RG2"})

Membership rules (assign entities by attribute matching; optional path filter):

Copyrisk_groups:
  - name: HighCapacityLinks
    membership:
      scope: link            # Required: node, link, or risk_group
      match:
        logic: and           # "and" or "or" (default: "and" for membership)
        conditions:
          - attr: capacity
            op: ">="
            value: 1000

Generate blocks (create groups from unique attribute values; optional path filter):

Copyrisk_groups:
  - generate:
      scope: node            # Required: node or link only
      path: "^prod_.*"       # Optional: narrow entities before grouping
      group_by: region       # Any attribute to group by
      name: "Region_${value}"

Validation: Risk group references are validated at load time (undefined references and circular hierarchies detected).

See REFERENCE.md for complete details.

Workflow

Copyworkflow:
  - type: NetworkStats
    name: stats
  - type: MaximumSupportedDemand
    name: msd
    demand_set: production
    alpha_start: 1.0
    resolution: 0.05
  - type: TrafficMatrixPlacement
    name: placement
    demand_set: production
    failure_policy: single_link
    iterations: 1000
    alpha_from_step: msd          # Reference MSD result
    alpha_from_field: data.alpha_star
  - type: MaxFlow
    source: "^(dc[1-3])$"
    target: "^(dc[1-3])$"
    mode: pairwise
    failure_policy: single_link
    iterations: 1000
    seed: 42                      # Optional: for reproducibility

Step types: BuildGraph, NetworkStats, MaxFlow, TrafficMatrixPlacement, MaximumSupportedDemand, CostPower

Common Pitfalls

1. Custom fields must go in attrs

Copy# WRONG
nodes:
  A:
    my_field: value    # Error!

# CORRECT
nodes:
  A:
    attrs:
      my_field: value

2. Link properties are flattened (no wrapper)

Copy# WRONG
links:
  - source: A
    target: B
    link_params:       # Error! No wrapper
      capacity: 100

# CORRECT
links:
  - source: A
    target: B
    capacity: 100      # Direct property

3. one_to_one requires compatible sizes

Sizes must have a multiple factor (4-to-2 OK, 3-to-2 ERROR).

4. Path patterns are anchored at start

Copypath: "leaf"       # Only matches names STARTING with "leaf"
path: ".*leaf.*"   # Matches "leaf" anywhere

Note: Leading / is stripped and has no effect. /leaf and leaf are equivalent. All paths are relative to the current scope (blueprint instantiation path or network root).

5. Variable syntax uses $ prefix

Copy# WRONG (conflicts with regex {m,n})
source: "{dc}/leaf"

# CORRECT
source: "${dc}/leaf"

6. zip requires equal-length lists

Copy# WRONG
expand:
  vars:
    a: [1, 2]
    b: [x, y, z]     # Length mismatch!
  mode: zip

7. Processing order matters

1. Groups and direct nodes created
2. Node rules applied
3. Blueprint links expanded
4. Top-level links expanded (including direct links)
5. Link rules applied

Rules only affect entities that exist at their processing stage.

Validation Checklist

• Custom fields inside attrs
• Link properties directly on link (no wrapper)
• Referenced blueprints exist
• Node names in direct links exist
• one_to_one sizes have multiple factor
• zip lists have equal length
• Selectors have at least one of: path, group_by, match

More Information

• Full DSL Reference - Complete field documentation, all operators, workflow steps
• Working Examples - 22 complete scenarios from simple to advanced

Running Scenarios

Always validate before running:

Copy./venv/bin/ngraph inspect scenario.yaml && ./venv/bin/ngraph run scenario.yaml

CLI Commands

Success Indicators

inspect success:

Copy✓ YAML file loaded successfully
✓ Scenario structure is valid

run success:

Copy✅ Scenario execution completed
✅ Results written to: scenario.results.json

Exit code 0, results JSON created.

Failure Indicators

Schema validation error:

Copy❌ ERROR: Failed to inspect scenario
  ValidationError: Additional properties are not allowed ('bad_field' was unexpected)
On instance['network']['nodes']['A']:
    {'bad_field': 'x'}

Fix: Move custom fields inside attrs: {}.

Missing node reference:

CopyValueError: Source node 'X' not found in network

Fix: Check node name spelling or pattern matching.

Empty selector match:

CopyWARNING: No nodes matched selector

Fix: Verify regex pattern matches actual node names.

Interpreting Results

Results JSON structure:

Copy{
  "steps": {
    "<step_name>": {
      "metadata": { "duration_sec": 0.5 },
      "data": { ... }
    }
  }
}

Key metrics by step type:

Quick validation after run:

Copy# Check alpha_star (should be >= 1.0)
grep -o '"alpha_star": [0-9.]*' scenario.results.json

# Check for dropped traffic (should be 0)
grep -o '"total_dropped": [0-9.]*' scenario.results.json | head -5

Common Errors and Fixes

Profiling Scenarios

When performance matters:

Copy./venv/bin/ngraph run scenario.yaml --profile --profile-memory

Output shows:

• Step timing: Time per workflow step
• Bottlenecks: Steps taking >10% of total time
• Memory: Peak memory per step (with --profile-memory)
• Recommendations: Optimization suggestions

Iteration Pattern

1. Write scenario YAML
2. ./venv/bin/ngraph inspect scenario.yaml - fix any errors
3. ./venv/bin/ngraph run scenario.yaml
4. Check results: alpha_star, total_dropped
5. If results bad -> adjust topology/demands -> repeat