acts-package-http

The acts http package plugin for acts.

Installation

cargo add acts-package-http

Start

use acts::Engine;
use acts_package_http::HttpPackage;

#[tokio::main]
async fn main() {
    let engine = Engine::builder()
        .add_pacakge::<HttpPackage>()
        .start();
}

Example

name: http example
id: http-example
inputs:
  key1: 1
  key2: 2
steps:
  - name: http step
    uses: acts.core.http
    params:
      url: http://127.0.0.1:1234/hello
      method: GET
      # params from workflow.inputs
      params: 
        - key: key1
          value: '${{ key1 }}'
        - key: key2
          value: '${{ key2 }}'
  - name: http step 2
    uses: acts.core.http
    params:
      url: http://127.0.0.1:1234/world
      method: POST
      content-type: json
      # body data from prev http response data
      body:
        data: '${{ $inputs().data }}'

Configuration

The package reads an optional [http] section from the engine config (acts.toml), or HttpPackage::from_config when building it directly:

[http]
# Egress allowlist. When non-empty only these hosts can be requested;
# "*.example.com" matches subdomains but not example.com itself.
allowed-hosts = ["api.example.com", "*.example.org"]
# Opt-in for internal/on-prem endpoints. Cloud metadata addresses stay
# blocked even when this is true.
allow-private-addresses = false
# Hard cap on a response body; larger bodies fail the act.
max-response-bytes = 67108864
# Connect timeout; 1..=300000.
connect-timeout-ms = 10000
# Whole-request timeout, including reading the body; 1..=3600000.
timeout-ms = 30000

Neither timeout can be disabled: a request without a deadline waits on the remote server for as long as it keeps the connection, and that wait is a scheduler lane the engine cannot hand to anyone else. timeout-ms and connect-timeout-ms out of range fail startup rather than being clamped.

An act may set its own timeout-ms param, but only to tighten the configured value — a request that must not outlive a workflow's own horizon fails fast instead of waiting out the deployment's bound. A value above the configured one (or zero) fails the act before anything is sent.

Cancellation reaches the request too: when the engine shuts down, or an action (abort, cancel, skip, remove, next, error) overrides the task while the act runs, the request is dropped and the act reports no failure of its own — the action owns the task's state, and a shutdown leaves the task for the next start to resume.

Egress policy

Every target is validated before connecting and again at DNS resolution / connection time:

only http and https are accepted;
with allowed-hosts set, the host must match an entry;
loopback, RFC1918, link-local, carrier-grade NAT and IPv6 unique-local addresses are rejected unless allow-private-addresses = true;
cloud metadata endpoints (169.254.169.254, 100.100.100.200, fd00:ec2::254) are always rejected;
redirects are re-validated, so a permitted URL cannot bounce the client to an internal target.

The request uses one package-level async client, so connections and TLS sessions are reused. timeout-ms, connect-timeout-ms and max-response-bytes bound how long an act can hang and how much a single response can allocate. The request deadline covers the whole request — connecting, redirects, and streaming the body — so a server that answers the headers and then stalls cannot hold the act open.

Local example servers (e.g. http://127.0.0.1:1234) need allow-private-addresses = true.