refactor(backend): base owns the gateway-attach flow, fail hard (0081 review)
lint / lint (push) Successful in 1m1s
test / unit (pull_request) Successful in 1m5s
tracker-policy-pr / check-pr (pull_request) Successful in 10s
test / integration-docker (pull_request) Failing after 1m30s
test / coverage (pull_request) Has been skipped

Addresses review on #519 (@didericis 6143 + the fail-hard direction over the
codex skip).

Base owns the flow (6143 / ADR 0006). `attach_bottled_agents_to_gateway()` is
now a concrete method on `BottleBackend` that delegates to
`gateway_attach.reconcile_running_bottles`; backends override only three
primitives — `_gateway_attach_resources()`, `_running_bottles()`,
`_attach_bottle_to_gateway()`. The shared control flow + error policy live in
one place so a backend can't drift onto a bespoke loop or a silent skip. Keeps
the reconcile flow + `GatewayAttachResources` out of `backend/base.py` (the
size guardrail) in a dedicated `backend/gateway_attach.py`. New ADR 0006 records
the "shared behaviour in the base backend, subclasses override primitives"
theme.

Fail hard, never skip (the maintainer's direction over the codex review's
skip). Any attach failure now aborts bring-up instead of being logged and
skipped: a bottle that silently can't reach the fresh gateway (its egress just
starts failing TLS) is worse than a loud failure. Every bottle is attempted and
the failures are raised together (aggregate `InfraLaunchError`) so one bring-up
surfaces the full blast radius. Resource gathering (the CA fetch) also fails
hard. PRD 0081 goal + design updated to match (reversed from the earlier
"tolerate per-bottle failures" draft).

Bumps the base.py guardrail cap 580->600 for the new contract surface (the
delegator + 3 abstract primitives); the flow itself lives in gateway_attach.py.

Refs #516, #519.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-07-27 01:50:06 +00:00
parent 4f2ccaed29
commit 3c0d2fb66f
13 changed files with 420 additions and 251 deletions
@@ -0,0 +1,65 @@
# ADR 0006: Shared behaviour lives in the base backend
- **Status:** Accepted
- **Date:** 2026-07-27
- **Deciders:** didericis
## Context
`BottleBackend` has three concrete implementations (Firecracker, docker,
macOS-container). Cross-backend operations can be expressed two ways:
1. **Abstract method per backend** — the base declares `@abstractmethod foo()`
and each backend writes its own `foo()` end to end.
2. **Template method** — the base owns the *control flow* of `foo()` as a
concrete method and each backend overrides small **primitives** it calls.
Form 1 gives each backend total freedom, which is exactly the problem: the three
implementations drift. The bring-up reconcile (PRD 0081) first shipped as form 1
and immediately diverged — each backend re-implemented "gather CA → list running
bottles → push to each", and one backend's copy silently skipped failures while
another's aborted, an inconsistency caught only in review (#519). The behaviour
that must be identical (the loop, the error policy, the ordering) was duplicated
in the place least likely to stay in sync.
## Decision
Prefer the template-method form for cross-backend behaviour: **encode shared
control flow and policy in the base backend; backends override primitives, not
control flow.** Lean toward *less* freedom in the concrete backends and *more*
consistency in the base.
Concretely, a cross-backend operation on `BottleBackend` should be a concrete
method that:
- owns the sequencing, iteration, and error policy (e.g. fail-hard vs.
best-effort, fail-fast vs. attempt-all-then-aggregate); and
- delegates only the irreducibly backend-specific steps to `@abstractmethod`
primitives with narrow, well-typed signatures.
The first application is `attach_bottled_agents_to_gateway()` (PRD 0081): the
base gathers resources once, attempts every running bottle, and raises an
aggregate `InfraLaunchError` if any failed; backends supply only
`_gateway_attach_resources()`, `_running_bottles()`, and
`_attach_bottle_to_gateway()`.
A backend-specific `@abstractmethod` with no shared control flow (e.g.
`_launch_impl`, `enumerate_active`) stays as form 1 — this ADR is about
operations whose *shape* should be uniform, not about banning abstract methods.
## Consequences
- The error policy, ordering, and loop for a cross-backend operation have one
source of truth; a new backend cannot forget them or diverge.
- New backends implement smaller, more focused primitives instead of re-deriving
a whole flow.
- The base carries more logic and needs generic-enough primitive signatures
(e.g. a backend-specific attach-target type parameter on `BottleBackend`).
- Genuinely backend-specific one-offs still use a plain abstract method; the
guidance is a default, not an absolute.
## Links
- PRD 0081 (`docs/prds/0081-reprovision-gateway-state-on-bringup.md`).
- Review that prompted this: PR #519.
- `bot_bottle/backend/base.py` (`BottleBackend.attach_bottled_agents_to_gateway`).