Files
Wenxi 119418c07c fix(craft): stabilize local Kubernetes startup and image builds (#16571)
## Description

Make local Craft startup recover from stale Telepresence agents and use
the deployed cluster configuration. Zed reuses compatible intercepts,
preserves intercepts it does not own, and cleans up newly created
intercepts after failed readiness checks.

Allow image builds behind a trusted TLS inspection proxy through an
optional BuildKit CA secret. Sandbox package downloads use the secret
without persisting it. Add LibreOffice Writer and Calc for document and
spreadsheet conversions.

Readiness waits for API health and leaves services running if its check
fails. The optional build CA secret also covers backend dependencies and
Chromium downloads.

## How Has This Been Tested?

- Follow-up: all 29 local-development tests and Dockerfile build checks
pass.

- 29 local development workflow tests passed, including intercept
compatibility, ownership, and failed rollout cleanup.
- Zed shell syntax and pre-commit checks passed.
- No running Zed services or sandbox pods were restarted during review
fixes.

## Additional Options

- [ ] [Optional] Please cherry-pick this PR to the latest release
version.


<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
<!-- coderabbit-ai-generated:v1 -->

## Summary by CodeRabbit

* **Documentation**
* Updated local Kubernetes setup guidance for personal configuration
overrides, startup retries, and API/frontend readiness checks.
* Clarified that intercepts are reused only when they forward port 8080
correctly, and that Zed preserves reused intercepts while cleaning up
ones it creates if rollout checks fail.
* Added guidance for launching the API and worker together and building
images behind TLS inspection proxies with an optional certificate
bundle.
* **New Features**
* Sandbox images built with skills enabled now include LibreOffice
Writer and Calc.
  * Image downloads can use an optional trusted certificate bundle.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
2026-10-22 22:23:48 +00:00

804 lines
27 KiB
Python

#!/usr/bin/env python3
"""Local kind development commands shared by Make and VS Code."""
import argparse
import hashlib
import json
import os
import re
import shutil
import subprocess
import sys
from pathlib import Path
from typing import NotRequired, TypedDict, cast
from runtime import Cluster
ROOT = Path(__file__).resolve().parents[3]
LABEL = "app.onyx.dev.image"
SOURCE_LABEL = "app.onyx.dev.source"
OWNER_LABEL = "app.onyx.dev.checkout"
IMAGES = {
"backend": (
os.environ.get("BACKEND_IMAGE", "onyxdotapp/onyx-backend:dev"),
"backend",
),
"sandbox": (
os.environ.get("SANDBOX_IMAGE", "onyxdotapp/sandbox:dev"),
"backend/onyx/server/features/build/sandbox/image",
),
}
class ImageConfig(TypedDict):
Labels: NotRequired[dict[str, str] | None]
class ImageInfo(TypedDict):
Config: ImageConfig
def run(
*args: str,
capture: bool = False,
cwd: Path = ROOT,
timeout: float | None = 120,
) -> str:
"""Bound short commands; callers can allow longer builds, loads, and cleanup."""
result = subprocess.run(
args,
cwd=cwd,
check=True,
text=True,
timeout=timeout,
stdout=subprocess.PIPE if capture else None,
)
return result.stdout.strip() if capture else ""
def source_hash(context: str) -> str:
"""Fingerprint tracked and unignored source contents, paths, and file modes."""
paths = subprocess.check_output(
[
"git",
"ls-files",
"-z",
"--cached",
"--others",
"--exclude-standard",
"--",
context,
],
cwd=ROOT,
timeout=120,
).split(b"\0")
digest = hashlib.sha256()
for name in sorted(set(paths) - {b""}):
path = ROOT / os.fsdecode(name)
if path.is_symlink():
contents = os.fsencode(os.readlink(path))
elif path.is_file():
contents = path.read_bytes()
else:
continue
digest.update(name + b"\0" + str(path.lstat().st_mode).encode() + b"\0")
digest.update(hashlib.sha256(contents).digest())
return digest.hexdigest()
def nodes(cluster: str) -> list[str]:
"""Find the selected kind cluster nodes or reject a missing cluster."""
result = run("kind", "get", "nodes", "--name", cluster, capture=True).splitlines()
if not result:
raise ValueError(f"kind cluster {cluster!r} does not exist")
return result
def kube(
args: argparse.Namespace,
*command: str,
capture: bool = False,
timeout: float = 120,
) -> str:
"""Run kubectl against the selected local cluster and namespace."""
return run(
"kubectl",
"--context",
f"kind-{args.cluster_name}",
"--request-timeout=30s",
"-n",
args.namespace,
*command,
capture=capture,
timeout=timeout,
)
def inspect_image(image: str) -> ImageInfo | None:
"""Inspect a Docker image without treating daemon failures as missing images."""
run("docker", "info", "--format", "{{.ServerVersion}}", capture=True)
result = subprocess.run(
["docker", "image", "inspect", image],
capture_output=True,
text=True,
timeout=120,
)
if result.returncode:
if "No such image" in result.stderr or "No such object" in result.stderr:
return None
raise ValueError(result.stderr.strip())
return cast(ImageInfo, json.loads(result.stdout)[0])
def labels(info: ImageInfo) -> dict[str, str]:
"""Read image labels, including images with no labels."""
return info["Config"].get("Labels") or {}
def images(args: argparse.Namespace) -> int:
"""Check, build, load, and apply selected images while preserving live sandboxes."""
nodes(args.cluster_name)
selected = [args.only] if args.only else list(IMAGES)
stale = False
for name in selected:
image, context = IMAGES[name]
fingerprint = source_hash(context)
info = inspect_image(image)
matches = info is not None and labels(info).get(SOURCE_LABEL) == fingerprint
if args.check:
print(
f"{image}: {'matches checkout' if matches else 'missing or source changed'}"
)
stale |= not matches
continue
if args.reuse and info is None:
raise ValueError(f"{image} is missing; omit --reuse to build it")
if args.force or (info is None or (not args.reuse and not matches)):
build_options: list[str] = []
if ca_bundle := os.environ.get("CRAFT_BUILD_CA_BUNDLE"):
bundle_path = Path(ca_bundle).expanduser().resolve(strict=True)
build_options = ["--secret", f"id=build_ca_bundle,src={bundle_path}"]
run(
"docker",
"build",
*build_options,
"--label",
f"{LABEL}=true",
"--label",
f"{SOURCE_LABEL}={fingerprint}",
"--label",
f"{OWNER_LABEL}={ROOT}",
"-t",
image,
str(ROOT / context),
timeout=None,
)
else:
print(f"Reusing {image}; no build needed")
run(
"kind",
"load",
"docker-image",
image,
"--name",
args.cluster_name,
timeout=None,
)
if args.load_only:
continue
if name == "backend":
# Update the actual workload image, not just its rollout timestamp.
for deployment in ("onyx-sandbox-proxy", "onyx-api-server"):
workload = json.loads(
kube(
args,
"get",
"deployment",
deployment,
"-o",
"json",
capture=True,
)
)
containers = workload["spec"]["template"]["spec"]["containers"]
patches = []
for index, container in enumerate(containers):
# Telepresence's traffic-agent must retain its own image.
if container["name"] not in {"api-server", "sandbox-proxy"}:
continue
patches.extend(
[
{
"op": "add",
"path": f"/spec/template/spec/containers/{index}/image",
"value": image,
},
{
"op": "add",
"path": f"/spec/template/spec/containers/{index}/imagePullPolicy",
"value": "IfNotPresent",
},
]
)
kube(
args,
"patch",
"deployment",
deployment,
"--type=json",
"-p",
json.dumps(patches),
)
kube(args, "rollout", "restart", f"deployment/{deployment}")
kube(
args,
"rollout",
"status",
f"deployment/{deployment}",
"--timeout=180s",
"--request-timeout=200s",
timeout=210,
)
else:
template = json.loads(
run(
"kubectl",
"--context",
f"kind-{args.cluster_name}",
"--request-timeout=30s",
"-n",
args.sandbox_namespace,
"get",
"podtemplate",
"sandbox-pod",
"-o",
"json",
capture=True,
)
)
spec = template["template"]["spec"]
patches = []
for group in ("containers", "initContainers"):
for index, container in enumerate(spec.get(group, [])):
if container["name"] not in {"sandbox", "sandbox-init", "sidecar"}:
continue
for field, value in (
("image", image),
("imagePullPolicy", "IfNotPresent"),
):
patches.append(
{
"op": "add",
"path": f"/template/spec/{group}/{index}/{field}",
"value": value,
}
)
if not patches:
raise ValueError("Sandbox PodTemplate has no recognized containers")
run(
"kubectl",
"--context",
f"kind-{args.cluster_name}",
"--request-timeout=30s",
"-n",
args.sandbox_namespace,
"patch",
"podtemplate",
"sandbox-pod",
"--type=json",
"-p",
json.dumps(patches),
)
print("Existing sandboxes retain their image. No live sandbox was deleted.")
kube(
args,
"get",
"pods",
"-o",
"custom-columns=NAME:.metadata.name,IMAGE:.spec.containers[*].image,IMAGE_ID:.status.containerStatuses[*].imageID",
)
return int(stale)
def disk(args: argparse.Namespace) -> None:
"""Report host Docker storage and each selected kind node image store."""
run("docker", "system", "df")
for node in nodes(args.cluster_name):
print(f"\nkind node: {node}", flush=True)
run("docker", "exec", node, "df", "-h", "/var/lib/containerd")
run("docker", "exec", node, "crictl", "images")
def doctor(args: argparse.Namespace) -> int:
"""Inspect prerequisites without changing the cluster or exposing secrets."""
failures = 0
def report(level: str, message: str) -> None:
"""Print one diagnostic result and count required failures."""
nonlocal failures
failures += level == "FAIL"
print(f"{level}: {message}", flush=True)
def probe(*command: str) -> subprocess.CompletedProcess[str] | None:
"""Capture diagnostic commands without exposing their output on failure."""
try:
return subprocess.run(
command,
cwd=ROOT,
capture_output=True,
text=True,
timeout=20,
check=False,
)
except (OSError, subprocess.TimeoutExpired):
return None
available = {
name: shutil.which(name)
for name in ("docker", "kind", "kubectl", "helm", "telepresence", "uv")
}
for name, path in available.items():
report("PASS" if path else "FAIL", f"{name}: {path or 'missing from PATH'}")
free_gib = shutil.disk_usage(ROOT).free / 1024**3
report(
"PASS" if free_gib >= 10 else "WARN",
f"Checkout filesystem: {free_gib:.1f} GiB free; Docker storage can use another filesystem",
)
if available["docker"]:
result = probe("docker", "info", "--format", "{{.ServerVersion}}")
report(
"PASS" if result is not None and result.returncode == 0 else "FAIL",
"Docker daemon access"
+ (
f": {result.stdout.strip()}"
if result is not None and result.returncode == 0
else "; check the daemon and socket permissions"
),
)
context = f"kind-{args.cluster_name}"
if available["kubectl"]:
result = probe(
"kubectl",
"--context",
context,
"--request-timeout=10s",
"version",
"-o",
"json",
)
if result is not None and result.returncode == 0:
versions = json.loads(result.stdout)
client = versions["clientVersion"]["gitVersion"]
server = versions["serverVersion"]["gitVersion"]
client_match = re.match(r"v(\d+)\.(\d+)", client)
server_match = re.match(r"v(\d+)\.(\d+)", server)
supported = (
client_match is not None
and server_match is not None
and client_match[1] == server_match[1]
and abs(int(client_match[2]) - int(server_match[2])) <= 1
)
report(
"PASS" if supported else "FAIL",
f"kubectl {client}; server {server}; supported minor-version skew is at most one",
)
native_sidecars = server_match is not None and (
int(server_match[1]),
int(server_match[2]),
) >= (1, 33)
report(
"PASS" if native_sidecars else "FAIL",
"Craft requires Kubernetes 1.33 or later",
)
result = probe(
"kubectl",
"--context",
context,
"--request-timeout=10s",
"get",
"nodes",
"-o",
"json",
)
items = (
json.loads(result.stdout)["items"]
if result is not None and result.returncode == 0
else []
)
ready = bool(items) and all(
any(
c["type"] == "Ready" and c["status"] == "True"
for c in node["status"].get("conditions", [])
)
for node in items
)
report("PASS" if ready else "FAIL", f"{context}: node readiness")
else:
report(
"FAIL",
f"{context}: API unavailable; start or resume the selected cluster",
)
result = probe(
"kubectl",
"--context",
context,
"--request-timeout=10s",
"-n",
args.sandbox_namespace,
"get",
"podtemplate",
"sandbox-pod",
"-o",
"json",
)
if result is not None and result.returncode == 0:
spec = json.loads(result.stdout)["template"]["spec"]
containers = [
c
for group in ("containers", "initContainers")
for c in spec.get(group, [])
if c["name"] in {"sandbox", "sandbox-init", "sidecar"}
]
matches = {c["name"] for c in containers} == {
"sandbox",
"sandbox-init",
"sidecar",
} and all(
c["image"] == IMAGES["sandbox"][0]
and c.get("imagePullPolicy") in {"IfNotPresent", "Never"}
for c in containers
)
report(
"PASS" if matches else "FAIL",
"Sandbox PodTemplate uses the selected local image without registry pulls; repair with make craft-sandbox-image",
)
else:
report("FAIL", "Sandbox PodTemplate unavailable; run make craft-up")
if available["telepresence"]:
result = probe("telepresence", "status", "--format", "json")
status = (
json.loads(result.stdout)
if result is not None and result.returncode == 0
else {}
)
user = status.get("user_daemon", {})
connected = (
user.get("kubernetes_context") == context
and user.get("namespace") == args.namespace
and user.get("status") == "Connected"
and status.get("root_daemon", {}).get("running", False)
)
report(
"PASS" if connected else "FAIL",
f"Telepresence connection to {context}/{args.namespace}"
+ (
""
if connected
else "; run telepresence connect in an interactive terminal for sudo authorization"
),
)
version = user.get("version", "")
report(
"PASS" if version.removeprefix("v").startswith("2.") else "FAIL",
f"Telepresence v2 daemon: {version or 'not running'}",
)
binary = ROOT / ".venv/bin/python"
report(
"PASS" if binary.is_file() else "FAIL", "Checkout Python environment (.venv)"
)
if available["uv"]:
result = probe(
"uv",
"sync",
"--check",
"--offline",
"--frozen",
"--no-default-groups",
"--group",
"backend",
"--group",
"dev",
"--group",
"ee",
)
report(
"PASS" if result is not None and result.returncode == 0 else "FAIL",
"Checkout Python dependencies (backend, dev, ee)"
+ (
""
if result is not None and result.returncode == 0
else "; run uv sync --frozen --no-default-groups --group backend --group dev --group ee"
),
)
if binary.is_file():
script = """import json, sys
from pathlib import Path
sys.path.insert(0, sys.argv[1])
from runtime import Cluster
cluster = Cluster(Path(sys.argv[2]), sys.argv[3], sys.argv[4], sys.argv[5])
env = cluster.environment()
required = ('POSTGRES_HOST', 'SANDBOX_PROXY_HOST', 'SANDBOX_CONTAINER_IMAGE', 'ONYX_SERVER_URL', 'ONYX_SANDBOX_PUSH_PRIVATE_KEY')
print(json.dumps({
'missing': [key for key in required if not env.get(key) or env[key].startswith('<')],
'model_key': bool(env.get('GEN_AI_API_KEY')),
'image': env.get('SANDBOX_CONTAINER_IMAGE') == sys.argv[6],
}))
"""
result = probe(
str(binary),
"-c",
script,
str(Path(__file__).parent),
str(ROOT),
args.cluster_name,
args.namespace,
args.sandbox_namespace,
IMAGES["sandbox"][0],
)
if result is None or result.returncode:
report(
"FAIL",
"Cannot resolve deployed API configuration; check cluster access, secrets, and Python dependencies",
)
else:
env_status = json.loads(result.stdout)
missing = env_status["missing"]
report(
"FAIL" if missing else "PASS",
"Effective Craft configuration"
+ (f": fill {', '.join(missing)}" if missing else " is available"),
)
report(
"PASS" if env_status["image"] else "FAIL",
"Effective sandbox image matches the selected local image",
)
report(
"PASS" if env_status["model_key"] else "WARN",
"GEN_AI_API_KEY is set"
if env_status["model_key"]
else "GEN_AI_API_KEY is unset; agent turns need a configured model provider",
)
print(f"Doctor finished: {failures} failed check(s). No resources changed.")
return int(failures > 0)
def clean(args: argparse.Namespace) -> None:
"""Preview or remove unused checkout images, with separate optional builder cache cleanup."""
candidates = run(
"docker",
"image",
"ls",
"-q",
"--filter",
"dangling=true",
"--filter",
f"label={LABEL}=true",
capture=True,
).splitlines()
for image in sorted(set(candidates)):
info = inspect_image(image)
if info is None or labels(info).get(OWNER_LABEL) != str(ROOT):
continue
if run("docker", "ps", "-aq", "--filter", f"ancestor={image}", capture=True):
continue
print(
f"{'Removing' if args.apply else 'Would remove'} unused checkout image {image}",
flush=True,
)
if args.apply:
run("docker", "image", "rm", image, timeout=None)
if args.build_cache:
print(
"Build cache cleanup affects the selected Docker builder, across all projects.",
flush=True,
)
print(
"Keeps a 10GB cache target and only considers entries unused for seven days.",
flush=True,
)
if args.apply:
run(
"docker",
"builder",
"prune",
"--force",
"--filter",
"until=168h",
"--keep-storage",
"10GB",
timeout=None,
)
print("Volumes, tagged images and kind runtime images are preserved.")
def cluster_runtime(args: argparse.Namespace) -> Cluster:
"""Bind shared runtime commands to the selected development environment."""
return Cluster(ROOT, args.cluster_name, args.namespace, args.sandbox_namespace)
def readiness(args: argparse.Namespace) -> None:
"""Check the selected infrastructure or application readiness stage."""
if args.stage == "infrastructure":
cluster_runtime(args).infrastructure_ready(args.timeout)
else:
cluster_runtime(args).application_ready(args.frontend_url, args.timeout)
def pause_resume(args: argparse.Namespace) -> None:
"""Stop or restart cluster nodes while preserving their data."""
cluster_nodes = nodes(args.cluster_name)
if args.command == "pause":
# Never quit a Telepresence connection to another cluster.
status = json.loads(
run("telepresence", "status", "--format", "json", capture=True)
)
connection = status.get("user_daemon", {})
if connection.get("kubernetes_context") == f"kind-{args.cluster_name}":
run("telepresence", "quit")
run("docker", "stop", *cluster_nodes)
print(
"Cluster data preserved. Stop your local VS Code launch compound separately."
)
else:
run("docker", "start", *cluster_nodes)
kube(
args,
"wait",
"--for=condition=Ready",
"nodes",
"--all",
"--timeout=180s",
"--request-timeout=200s",
timeout=210,
)
cluster_runtime(args).infrastructure_ready()
cluster_runtime(args).connect()
print(
"Cluster resumed. Relaunch your local services to restore the API intercept."
)
def migrate(args: argparse.Namespace) -> None:
"""Inspect or upgrade the selected database after validating local migration prerequisites."""
import psutil
env = cluster_runtime(args).database_environment(args.env_file)
host = env.get("POSTGRES_HOST", "localhost")
port = env.get("POSTGRES_PORT", "5432")
if args.operation == "upgrade":
for proc in psutil.process_iter(["pid", "cmdline", "cwd", "uids"]):
command = proc.info["cmdline"] or []
directory = proc.info["cwd"]
owner = proc.info["uids"]
if (
not directory
or not Path(directory).is_relative_to(ROOT)
or owner is None
or owner.real != os.getuid()
):
continue
launcher = str(ROOT / "deployment/helm/dev/launch.py")
shared_launch = any(
str((base / arg).resolve()) == launcher
for arg in command
if Path(arg).name == "launch.py"
for base in (Path(directory), ROOT)
) and any(role in command for role in ("api", "celery"))
api = "onyx.main:app" in command
worker = any("celery" in Path(arg).name for arg in command) and any(
arg.startswith("onyx.background.") for arg in command
)
if shared_launch or api or worker:
raise ValueError(
f"Stop local Onyx API/workers before migration (PID {proc.pid})"
)
run(
"telepresence",
"connect",
"--context",
f"kind-{args.cluster_name}",
"-n",
args.namespace,
)
print(f"Database: {host}:{port}; checkout: {ROOT}", flush=True)
binary = ROOT / ".venv/bin/python"
subprocess.run(
[str(binary), "-m", "alembic", "current"],
cwd=ROOT / "backend",
env=env,
check=True,
timeout=120,
)
if args.operation == "upgrade":
subprocess.run(
[str(binary), "-m", "alembic", "upgrade", "head"],
cwd=ROOT / "backend",
env=env,
check=True,
timeout=None,
)
subprocess.run(
[str(binary), "-m", "alembic", "current"],
cwd=ROOT / "backend",
env=env,
check=True,
timeout=120,
)
def main() -> int:
"""Dispatch local development commands and return their exit status."""
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--cluster-name", default="onyx-dev")
parser.add_argument("--namespace", default="onyx")
parser.add_argument("--sandbox-namespace", default="onyx-sandboxes")
commands = parser.add_subparsers(dest="command", required=True)
image_parser = commands.add_parser("images")
image_parser.add_argument("--only", choices=list(IMAGES))
action = image_parser.add_mutually_exclusive_group()
action.add_argument("--check", action="store_true")
action.add_argument("--force", action="store_true")
action.add_argument("--reuse", action="store_true")
image_parser.add_argument("--load-only", action="store_true")
commands.add_parser("disk")
commands.add_parser("doctor")
commands.add_parser("connect")
intercept = commands.add_parser("intercept")
intercept.add_argument("--ownership-file", type=Path)
ready = commands.add_parser("ready")
ready.add_argument(
"--stage", choices=("infrastructure", "application"), default="application"
)
ready.add_argument("--timeout", type=float, default=600)
ready.add_argument(
"--frontend-url",
default=os.environ.get("CRAFT_FRONTEND_URL", "http://localhost:3000"),
)
cleanup = commands.add_parser("clean")
cleanup.add_argument("--apply", action="store_true")
cleanup.add_argument("--build-cache", action="store_true")
commands.add_parser("pause")
commands.add_parser("resume")
migration = commands.add_parser("migrate")
migration.add_argument("operation", choices=["status", "upgrade"])
migration.add_argument("--env-file", type=Path, help="Defaults to .vscode/.env.k8s")
args = parser.parse_args()
try:
if args.command == "images":
return images(args)
if args.command == "doctor":
return doctor(args)
if args.command == "connect":
cluster_runtime(args).connect()
elif args.command == "intercept":
cluster_runtime(args).intercept(args.ownership_file)
elif args.command == "ready":
readiness(args)
elif args.command == "disk":
disk(args)
elif args.command == "clean":
clean(args)
elif args.command == "migrate":
migrate(args)
else:
pause_resume(args)
except (
ValueError,
OSError,
ImportError,
subprocess.CalledProcessError,
subprocess.TimeoutExpired,
) as exc:
print(f"error: {exc}", file=sys.stderr)
return 1
return 0
if __name__ == "__main__":
raise SystemExit(main())