Prototype variant of the auto-deploy architecture where approved PRs are accumulated into a batch window (e.g. N minutes) and deployed as a single unit, instead of triggering an immediate deploy per approval.
Why
The current architecture (see #24) starts with immediate deploy on every approval. We want a second implementation to compare:
immediate: approve → deploy within ~minute. Predictable latency. May produce multiple deploy cycles if several PRs are approved in quick succession. No batching, no human curation of release sets.
window: PRs accumulate for N minutes (or until M approvals), then one deploy ships the whole batch. Enables deliberate release sets. Higher latency per PR. Smoother load on the swarm.
A prototype of both lets us compare operationally and pick one (or run them in parallel for different stacks).
Scope
Implement the window variant as a separate reusable action (brencher3 or just a config flag in brencher2). Use a single watched repo (e.g. admin/hermes) as the test target.
Out of scope
Migrating brencher to brencher2 — separate effort.
Multi-repo orchestration — start with one repo, generalize later.
Acceptance
Approve 3 PRs in a row within 5 minutes → exactly one deploy cycle ships all three
A PR approved alone → either deploys in the next window or waits until the window closes
Window size is configurable (env var or workflow input)
Visible in PR / commit comments which deploy cycle included this PR
## What
Prototype variant of the auto-deploy architecture where approved PRs are accumulated into a batch window (e.g. N minutes) and deployed as a single unit, instead of triggering an immediate deploy per approval.
## Why
The current architecture (see #24) starts with `immediate` deploy on every approval. We want a second implementation to compare:
- **immediate:** approve → deploy within ~minute. Predictable latency. May produce multiple deploy cycles if several PRs are approved in quick succession. No batching, no human curation of release sets.
- **window:** PRs accumulate for N minutes (or until M approvals), then one deploy ships the whole batch. Enables deliberate release sets. Higher latency per PR. Smoother load on the swarm.
A prototype of both lets us compare operationally and pick one (or run them in parallel for different stacks).
## Scope
Implement the `window` variant as a separate reusable action (`brencher3` or just a config flag in brencher2). Use a single watched repo (e.g. `admin/hermes`) as the test target.
## Out of scope
- Migrating brencher to brencher2 — separate effort.
- Multi-repo orchestration — start with one repo, generalize later.
## Acceptance
- Approve 3 PRs in a row within 5 minutes → exactly one deploy cycle ships all three
- A PR approved alone → either deploys in the next window or waits until the window closes
- Window size is configurable (env var or workflow input)
- Visible in PR / commit comments which deploy cycle included this PR
## Related
- #24 -- main design ticket, immediate variant
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
What
Prototype variant of the auto-deploy architecture where approved PRs are accumulated into a batch window (e.g. N minutes) and deployed as a single unit, instead of triggering an immediate deploy per approval.
Why
The current architecture (see #24) starts with
immediatedeploy on every approval. We want a second implementation to compare:A prototype of both lets us compare operationally and pick one (or run them in parallel for different stacks).
Scope
Implement the
windowvariant as a separate reusable action (brencher3or just a config flag in brencher2). Use a single watched repo (e.g.admin/hermes) as the test target.Out of scope
Acceptance
Related