Type Catalog¶

Every component in a model has a type — a type defined by a provider. This page lists all types from the official providers, the facts you can observe, the states the engine can derive, and the failure modes it uses for reasoning.

Use this to write correct inject blocks in scenarios and healthy overrides in models without needing to install mgtt first.

Kubernetes provider¶

Install: mgtt provider install kubernetes

Auth: KUBECONFIG, ~/.kube/config, or in-cluster service account. Read-only access via kubectl.

Variable: namespace (default: default)

`ingress`¶

A Kubernetes ingress or reverse proxy (e.g., nginx ingress controller).

Facts¶

Fact	Type	Cost	Description
`upstream_count`	`mgtt.int`	low	Number of upstream endpoints backing this ingress

Health conditions¶

upstream_count > 0

States¶

State	Condition	Description
`live`	`upstream_count > 0`	Serving traffic normally
`draining`	`upstream_count == 0`	No upstream endpoints

Default active state: live

Failure modes¶

State	Can cause
`draining`	`upstream_failure`, `5xx_errors`

`deployment`¶

A Kubernetes Deployment — the most common workload type.

Facts¶

Fact	Type	Cost	Description
`ready_replicas`	`mgtt.int`	low	Number of pods in Ready state
`desired_replicas`	`mgtt.int`	low	Configured replica count (`.spec.replicas`)
`restart_count`	`mgtt.int`	low	Container restart count (highest across pods)
`endpoints`	`mgtt.int`	low	Number of endpoint IPs in the Service

Health conditions¶

ready_replicas == desired_replicas
endpoints > 0
restart_count < 5

States¶

States are evaluated top-to-bottom — first match wins.

State	Condition	Description
`degraded`	`ready_replicas < desired_replicas & restart_count > 5`	Crash-looping — pods restarting repeatedly
`draining`	`desired_replicas == 0`	Scaled to zero intentionally
`starting`	`ready_replicas < desired_replicas`	Pods initializing (not yet crash-looping)
`live`	`ready_replicas == desired_replicas`	All replicas ready

Default active state: live

State ordering matters

degraded must be checked before starting because both match when ready_replicas < desired_replicas. The difference is restart_count > 5 — without checking restarts, a crash-looping deployment looks like it's still starting up.

Failure modes¶

State	Can cause
`degraded`	`upstream_failure`, `timeout`, `connection_refused`, `5xx_errors`
`draining`	`upstream_failure`, `connection_refused`
`starting`	`upstream_failure`, `timeout`

AWS provider¶

Install: mgtt provider install aws

Auth: AWS_PROFILE, AWS_ACCESS_KEY_ID+AWS_SECRET_ACCESS_KEY, ~/.aws/credentials, or instance profile. Read-only AWS API access.

`rds_instance`¶

An AWS RDS database instance.

Facts¶

Fact	Type	Cost	Description
`available`	`mgtt.bool`	low	Whether the instance is accepting connections
`connection_count`	`mgtt.int`	low	Current database connections (from CloudWatch)

Health conditions¶

available == true
connection_count < 500

States¶

State	Condition	Description
`live`	`available == true`	Accepting connections
`stopped`	`available == false`	Not accepting connections

Default active state: live

Failure modes¶

State	Can cause
`stopped`	`upstream_failure`, `connection_refused`, `query_timeout`

Docker provider (community)¶

Install: mgtt provider install https://github.com/mgt-tool/mgtt-provider-docker

`container`¶

A Docker container.

See the provider repository for the full type definition.

Stdlib primitive types¶

Every provider fact declares a type from mgtt's built-in stdlib. These are the base types:

Type	Base	Unit/Range	Example
`mgtt.int`	integer	—	`42`
`mgtt.float`	float	—	`0.95`
`mgtt.bool`	boolean	—	`true`
`mgtt.string`	string	—	`"running"`
`mgtt.duration`	float	ms, s, m, h, d	`500` (ms)
`mgtt.bytes`	integer	b, kb, mb, gb, tb	`1024`
`mgtt.ratio`	float	0..1	`0.95`
`mgtt.percentage`	float	0..100	`95.0`
`mgtt.count`	integer	0..	`12`
`mgtt.timestamp`	string	ISO 8601	`"2024-02-05T07:50:00Z"`

Inspect at runtime:

mgtt stdlib ls              # list all primitive types
mgtt stdlib inspect count   # details for a specific type

Standard failure mode vocabulary¶

Providers declare failure modes using a standard vocabulary. These are the recognized terms:

Failure mode	Meaning
`upstream_failure`	Downstream components cannot reach this one
`connection_refused`	TCP connections actively rejected
`timeout`	Responses too slow or no response
`5xx_errors`	HTTP 5xx errors returned to callers
`query_timeout`	Database queries timing out
`dns_failure`	DNS resolution failing
`auth_failure`	Authentication/authorization rejected
`resource_exhaustion`	CPU, memory, disk, or connection limits hit

When the type you need doesn't exist¶

The catalog above is small — three types across two providers. Real systems have many more component types (ElastiCache clusters, message brokers, S3 buckets, CDNs, secrets stores, etc.). When you need a type that isn't listed, you have three options:

Option 1: Write a vocabulary-only provider (fastest)¶

Define a manifest.yaml with your types, facts, states, and inline probe commands. No compiled binary needed — mgtt executes the shell commands directly.

# my-aws-extras/manifest.yaml
meta:
  name: my-aws-extras
  version: 0.1.0
  description: Additional AWS types for my project
  requires:
    mgtt: ">=0.2.0"

install:
  source:
    build: hooks/install.sh        # no-op script for vocabulary-only providers

# read_only defaults to true; vocabulary-only providers are inherently
# read-only. Document the AWS credential chain this provider uses in a
# README rather than in manifest.yaml — that's narrative that doesn't fit
# a structured field.

types:
  elasticache_cluster:
    description: AWS ElastiCache Redis/Memcached cluster
    facts:
      available:
        type: mgtt.bool
        ttl: 60s
        probe:
          cmd: "aws elasticache describe-cache-clusters --cache-cluster-id {name} --query 'CacheClusters[0].CacheClusterStatus' --output text"
          parse: bool
          cost: low
          access: AWS API read-only
    healthy:
      - available == true
    states:
      live:
        when: "available == true"
        description: accepting connections
      stopped:
        when: "available == false"
        description: not available
    default_active_state: live
    failure_modes:
      stopped:
        can_cause: [upstream_failure, connection_refused]

Install it:

mgtt provider install ./my-aws-extras

This is the right starting point. You can always add a compiled binary later for better performance or more complex probe logic.

Option 2: Extend an existing provider¶

Fork the official aws or kubernetes provider repository and add types to its manifest.yaml. This keeps all your AWS types in one provider.

Option 3: Contribute upstream¶

If your type is broadly useful (e.g., elasticache_cluster for the AWS provider), open a PR against the official provider. See Writing Providers for the full provider development guide.

The key point: mgtt's type system is open. The catalog above is what ships today, not what's possible. Any component you can observe via a shell command can be modeled.

Type Catalog¶

On this page¶

Kubernetes provider¶

ingress¶

Facts¶

Health conditions¶

States¶

Failure modes¶

deployment¶

Facts¶

Health conditions¶

States¶

Failure modes¶

AWS provider¶

rds_instance¶

Facts¶

Health conditions¶

States¶

Failure modes¶

Docker provider (community)¶

container¶

Stdlib primitive types¶

Standard failure mode vocabulary¶

When the type you need doesn't exist¶

Option 1: Write a vocabulary-only provider (fastest)¶

Option 2: Extend an existing provider¶

Option 3: Contribute upstream¶

`ingress`¶

`deployment`¶

`rds_instance`¶

`container`¶