feat(launch): switch start to docker compose project per bottle
test / unit (pull_request) Successful in 18s
test / integration (pull_request) Successful in 1m5s

PRD 0018 chunk 3. Each instance is now one `docker compose` project:

  - launch.py renders the compose spec via chunk-1's
    bottle_plan_to_compose, writes it to state/<slug>/docker-compose.yml,
    `docker compose up -d`s, and (on teardown) dumps
    `docker compose logs --no-color --timestamps` to
    state/<slug>/compose.log before `docker compose down`.
  - Networks are pre-created (`docker network create --internal` +
    user-defined bridge) so pipelock yaml can know the internal CIDR
    before compose-up. Compose references them with `external: true`;
    the launch step's ExitStack still owns network removal.
  - Agent still runs `sleep infinity`; claude reaches it via
    `docker exec -it` exactly like before (per the PRD's resolved
    TTY question).
  - metadata.json grows a `compose_project` field so dashboard /
    cleanup tooling can derive compose invocations without
    re-deriving the slug.

Security follow-ups from chunk-2 review:

  (b) CA private keys: pipelock + egress ca-key.pem land at 0o600
      explicitly. The mitmproxy cert+key concat stays 0o644 because
      the egress container's uid-1000 user reads it through the
      bind mount; parent dir at 0o700 still restricts host-side
      reach.
  (c) Apply atomicity: egress_apply + pipelock_apply switch from
      `docker cp` to host-side write-temp-then-rename on the
      bind-mount source. POSIX rename is atomic on the same
      filesystem, so a sidecar SIGHUP racing the apply can't see
      a half-written routes.yaml / pipelock.yaml.

Per-sidecar Docker{Sidecar}.start/stop methods stay in place — the
integration test suite drives them directly to validate each image
in isolation, which is still useful. launch.py no longer calls
them; a follow-up chunk can prune if the integration tests move to
the compose lifecycle.

git-gate entrypoint's chmod 600 on the keyfile + known_hosts now
tolerates EROFS (`|| true`) — the host SSH key is already 0600
(SSH refuses to load otherwise), so the inside-container chmod
was already a no-op in the docker-cp path and now just needs to
not error on the read-only bind mount.

422 unit tests pass; supervise integration test passes; end-to-end
`./cli.py start implementer` brings up the project, attaches,
captures full merged logs on teardown, and reaps all containers +
networks.
This commit is contained in:
2026-05-25 23:16:40 -04:00
parent b9f6889d09
commit cefdc8c6e9
11 changed files with 362 additions and 302 deletions
+135 -7
View File
@@ -39,6 +39,9 @@ aren't rebuilt on every up.
from __future__ import annotations
import json
import subprocess
import sys
from pathlib import Path
from typing import Any
@@ -46,6 +49,7 @@ from ...egress import (
EGRESS_HOSTNAME,
EGRESS_ROUTES_IN_CONTAINER,
)
from ...log import die, warn
from ...git_gate import git_gate_aggregate_extra_hosts
from ...supervise import (
CURRENT_CONFIG_DIR_IN_AGENT,
@@ -126,18 +130,21 @@ def bottle_plan_to_compose(plan: DockerBottlePlan) -> dict[str, Any]:
def _networks(plan: DockerBottlePlan) -> dict[str, Any]:
"""Two compose-managed networks with explicit `name:` matching
the existing slug-suffixed convention. The internal one is
`--internal` (no default gateway); the egress one is a normal
user-defined bridge so the upstream-bound sidecars can resolve
+ reach the outside world."""
"""Both networks are `external: true` — chunk 3 pre-creates them
via `docker network create` so pipelock's yaml can embed the
internal-network CIDR in its SSRF allowlist before compose-up.
Compose just references the pre-existing networks by name.
Network lifecycle (create / remove) is owned by the compose-
lifecycle helpers, not compose itself; `docker compose down`
leaves external networks alone."""
return {
"internal": {
"name": plan.proxy_plan.internal_network,
"internal": True,
"external": True,
},
"egress": {
"name": plan.proxy_plan.egress_network,
"external": True,
},
}
@@ -382,4 +389,125 @@ def _agent_no_proxy(plan: DockerBottlePlan) -> str:
return ",".join(hosts)
__all__ = ["bottle_plan_to_compose"]
# --- Lifecycle helpers (PRD 0018 chunk 3) ----------------------------------
#
# The renderer above is pure. The helpers below own the I/O side:
# serialize the spec to disk, drive `docker compose up`, dump the
# merged log file on teardown, and `docker compose down` to clean up
# (networks are pre-created externally so `down` leaves them alone;
# the launch step removes them in its own teardown step).
COMPOSE_FILE_NAME = "docker-compose.yml"
COMPOSE_LOG_NAME = "compose.log"
def compose_project_name(slug: str) -> str:
"""Stable mapping from slug → compose project. Matches the
`name:` field the renderer emits, so `docker compose ls`
enumeration and direct CLI invocations agree on the project
identifier."""
return f"claude-bottle-{slug}"
def compose_file_path(state_dir: Path) -> Path:
return state_dir / COMPOSE_FILE_NAME
def compose_log_path(state_dir: Path) -> Path:
return state_dir / COMPOSE_LOG_NAME
def write_compose_file(spec: dict[str, Any], path: Path) -> Path:
"""Serialize the compose dict to disk. JSON content with a
`.yml` filename — JSON is a strict subset of YAML 1.2 for the
constructs the renderer uses (mappings, lists, strings, bools,
nulls), and `docker compose -f file.yml` parses it as YAML.
Avoids a yaml dependency while keeping the file `cat`-readable.
"""
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text(json.dumps(spec, indent=2, sort_keys=False) + "\n")
path.chmod(0o644)
return path
def _compose_argv(project: str, compose_file: Path, *cmd: str) -> list[str]:
return [
"docker", "compose",
"-p", project,
"-f", str(compose_file),
*cmd,
]
def compose_up(
project: str,
compose_file: Path,
*,
env: dict[str, str] | None = None,
) -> None:
"""`docker compose up -d` for the project. Env-inheritance is
via `env=` on the subprocess — every `environment: [NAME]` (bare
name) entry in the compose file resolves to whatever value
`NAME` has in `env` at exec time. Secrets never land on argv or
in the compose file."""
argv = _compose_argv(project, compose_file, "up", "-d")
result = subprocess.run(
argv, capture_output=True, text=True, env=env, check=False,
)
if result.returncode != 0:
sys.stderr.write(result.stderr)
die(f"docker compose up failed for project {project}")
def compose_dump_logs(project: str, compose_file: Path, output: Path) -> None:
"""Write the merged stdout/stderr of every service to `output`
using `docker compose logs --no-color --timestamps`. Best-effort
— failures here shouldn't block teardown. The interleaved single
file is what the user reads post-mortem; per-service tail still
works through `docker compose logs -f <service>` while the
project is up."""
output.parent.mkdir(parents=True, exist_ok=True)
argv = _compose_argv(project, compose_file, "logs", "--no-color", "--timestamps")
try:
with open(output, "wb") as f:
subprocess.run(
argv,
stdout=f,
stderr=subprocess.STDOUT,
check=False,
)
output.chmod(0o644)
except OSError as e:
warn(f"failed to write compose log to {output}: {e}")
def compose_down(project: str, compose_file: Path) -> None:
"""`docker compose down` for the project. External networks are
intentionally NOT removed by compose (`external: true` on the
networks block); the launch step's own teardown removes them
via `network_remove` so the per-bottle ephemeral subnet doesn't
accumulate."""
argv = _compose_argv(project, compose_file, "down")
result = subprocess.run(
argv, capture_output=True, text=True, check=False,
)
if result.returncode != 0:
warn(
f"docker compose down failed for project {project}: "
f"{result.stderr.strip()}"
)
__all__ = [
"COMPOSE_FILE_NAME",
"COMPOSE_LOG_NAME",
"bottle_plan_to_compose",
"compose_down",
"compose_dump_logs",
"compose_file_path",
"compose_log_path",
"compose_project_name",
"compose_up",
"write_compose_file",
]