clan-cli: Rename Host -> Remote move to clan_lib and mark as frozen

This commit is contained in:
Qubasa
2025-05-22 14:08:27 +02:00
parent 7bcb0afae9
commit cff5d61f26
31 changed files with 453 additions and 429 deletions

View File

@@ -103,26 +103,23 @@ def generate_machine_hardware_info(opts: HardwareGenerateOptions) -> HardwareCon
"--show-hardware-config",
]
with machine.target_host() as host:
host.ssh_options["StrictHostKeyChecking"] = "accept-new"
host.ssh_options["UserKnownHostsFile"] = "/dev/null"
if opts.password:
host.password = opts.password
host = opts.machine.target_host()
out = host.run(config_command, become_root=True, opts=RunOpts(check=False))
if out.returncode != 0:
if "nixos-facter" in out.stderr and "not found" in out.stderr:
machine.error(str(out.stderr))
msg = (
"Please use our custom nixos install images from https://github.com/nix-community/nixos-images/releases/tag/nixos-unstable. "
"nixos-factor only works on nixos / clan systems currently."
)
raise ClanError(msg)
machine.error(str(out))
msg = f"Failed to inspect {opts.machine}. Address: {host.target}"
with host.ssh_control_master() as ssh:
out = ssh.run(config_command, become_root=True, opts=RunOpts(check=False))
if out.returncode != 0:
if "nixos-facter" in out.stderr and "not found" in out.stderr:
machine.error(str(out.stderr))
msg = (
"Please use our custom nixos install images from https://github.com/nix-community/nixos-images/releases/tag/nixos-unstable. "
"nixos-factor only works on nixos / clan systems currently."
)
raise ClanError(msg)
machine.error(str(out))
msg = f"Failed to inspect {opts.machine}. Address: {host.target}"
raise ClanError(msg)
backup_file = None
if hw_file.exists():
backup_file = hw_file.with_suffix(".bak")

View File

@@ -57,9 +57,9 @@ def install_machine(opts: InstallOptions) -> None:
generate_facts([machine])
generate_vars([machine])
host = machine.target_host()
with (
TemporaryDirectory(prefix="nixos-install-") as _base_directory,
machine.target_host() as host,
):
base_directory = Path(_base_directory).resolve()
activation_secrets = base_directory / "activation_secrets"

View File

@@ -2,8 +2,6 @@ import importlib
import json
import logging
import re
from collections.abc import Iterator
from contextlib import contextmanager
from dataclasses import dataclass, field
from functools import cached_property
from pathlib import Path
@@ -12,12 +10,11 @@ from typing import TYPE_CHECKING, Any
from clan_lib.errors import ClanCmdError, ClanError
from clan_lib.flake import Flake
from clan_lib.nix import nix_config, nix_test_store
from clan_lib.ssh.remote import Remote
from clan_cli.facts import public_modules as facts_public_modules
from clan_cli.facts import secret_modules as facts_secret_modules
from clan_cli.ssh.host import Host
from clan_cli.ssh.host_key import HostKeyCheck
from clan_cli.ssh.parse import parse_deployment_address
from clan_cli.vars._types import StoreBase
log = logging.getLogger(__name__)
@@ -146,37 +143,31 @@ class Machine:
def flake_dir(self) -> Path:
return self.flake.path
@contextmanager
def target_host(self) -> Iterator[Host]:
with parse_deployment_address(
self.name,
self.target_host_address,
self.host_key_check,
def target_host(self) -> Remote:
return Remote.from_deployment_address(
machine_name=self.name,
address=self.target_host_address,
host_key_check=self.host_key_check,
private_key=self.private_key,
meta={"machine": self},
) as target_host:
yield target_host
)
@contextmanager
def build_host(self) -> Iterator[Host | None]:
def build_host(self) -> Remote | None:
"""
The host where the machine is built and deployed from.
Can be the same as the target host.
"""
build_host = self.override_build_host or self.deployment.get("buildHost")
if build_host is None:
yield None
return
address = self.override_build_host or self.deployment.get("buildHost")
if address is None:
return None
# enable ssh agent forwarding to allow the build host to access the target host
with parse_deployment_address(
self.name,
build_host,
self.host_key_check,
host = Remote.from_deployment_address(
machine_name=self.name,
address=address,
host_key_check=self.host_key_check,
forward_agent=True,
private_key=self.private_key,
meta={"machine": self},
) as build_host:
yield build_host
)
return host
def nix(
self,

View File

@@ -5,7 +5,6 @@ import os
import re
import shlex
import sys
from contextlib import ExitStack
from clan_lib.api import API
from clan_lib.async_run import AsyncContext, AsyncOpts, AsyncRuntime, is_async_cancelled
@@ -13,6 +12,7 @@ from clan_lib.cmd import Log, MsgColor, RunOpts, run
from clan_lib.colors import AnsiColor
from clan_lib.errors import ClanError
from clan_lib.nix import nix_command, nix_config, nix_metadata
from clan_lib.ssh.remote import HostKeyCheck, Remote
from clan_cli.completions import (
add_dynamic_completer,
@@ -22,7 +22,6 @@ from clan_cli.facts.generate import generate_facts
from clan_cli.facts.upload import upload_secrets
from clan_cli.machines.list import list_machines
from clan_cli.machines.machines import Machine
from clan_cli.ssh.host import Host, HostKeyCheck
from clan_cli.vars.generate import generate_vars
from clan_cli.vars.upload import upload_secret_vars
@@ -44,161 +43,160 @@ def is_local_input(node: dict[str, dict[str, str]]) -> bool:
return local
def upload_sources(machine: Machine, host: Host) -> str:
env = host.nix_ssh_env(os.environ.copy())
def upload_sources(machine: Machine, host: Remote) -> str:
with host.ssh_control_master() as ssh:
env = ssh.nix_ssh_env(os.environ.copy())
flake_url = (
str(machine.flake.path) if machine.flake.is_local else machine.flake.identifier
)
flake_data = nix_metadata(flake_url)
has_path_inputs = any(
is_local_input(node) for node in flake_data["locks"]["nodes"].values()
)
flake_url = (
str(machine.flake.path)
if machine.flake.is_local
else machine.flake.identifier
)
flake_data = nix_metadata(flake_url)
has_path_inputs = any(
is_local_input(node) for node in flake_data["locks"]["nodes"].values()
)
if not has_path_inputs:
# Just copy the flake to the remote machine, we can substitute other inputs there.
path = flake_data["path"]
remote_url = f"ssh-ng://{host.target}"
# MacOS doesn't come with a proper login shell for ssh and therefore doesn't have nix in $PATH as it doesn't source /etc/profile
if machine._class_ == "darwin":
remote_url += "?remote-program=bash -lc 'exec nix-daemon --stdio'"
if not has_path_inputs:
# Just copy the flake to the remote machine, we can substitute other inputs there.
path = flake_data["path"]
cmd = nix_command(
[
"copy",
"--to",
f"ssh://{host.target}",
"--no-check-sigs",
path,
]
)
run(
cmd,
RunOpts(
env=env,
needs_user_terminal=True,
error_msg="failed to upload sources",
prefix=machine.name,
),
)
return path
# Slow path: we need to upload all sources to the remote machine
cmd = nix_command(
[
"copy",
"flake",
"archive",
"--to",
remote_url,
"--no-check-sigs",
path,
f"ssh://{host.target}",
"--json",
flake_url,
]
)
run(
proc = run(
cmd,
RunOpts(
env=env,
needs_user_terminal=True,
error_msg="failed to upload sources",
prefix=machine.name,
env=env, needs_user_terminal=True, error_msg="failed to upload sources"
),
)
return path
# Slow path: we need to upload all sources to the remote machine
cmd = nix_command(
[
"flake",
"archive",
"--to",
f"ssh://{host.target}",
"--json",
flake_url,
]
)
proc = run(
cmd,
RunOpts(
env=env, needs_user_terminal=True, error_msg="failed to upload sources"
),
)
try:
return json.loads(proc.stdout)["path"]
except (json.JSONDecodeError, OSError) as e:
msg = f"failed to parse output of {shlex.join(cmd)}: {e}\nGot: {proc.stdout}"
raise ClanError(msg) from e
try:
return json.loads(proc.stdout)["path"]
except (json.JSONDecodeError, OSError) as e:
msg = (
f"failed to parse output of {shlex.join(cmd)}: {e}\nGot: {proc.stdout}"
)
raise ClanError(msg) from e
@API.register
def deploy_machine(machine: Machine) -> None:
with ExitStack() as stack:
target_host = stack.enter_context(machine.target_host())
build_host = stack.enter_context(machine.build_host())
target_host = machine.target_host()
build_host = machine.build_host()
host = build_host or target_host
host = build_host or target_host
generate_facts([machine], service=None, regenerate=False)
generate_vars([machine], generator_name=None, regenerate=False)
generate_facts([machine], service=None, regenerate=False)
generate_vars([machine], generator_name=None, regenerate=False)
upload_secrets(machine)
upload_secret_vars(machine, target_host)
upload_secrets(machine)
upload_secret_vars(machine, target_host)
path = upload_sources(machine, host)
path = upload_sources(machine, host)
nix_options = [
"--show-trace",
"--option",
"keep-going",
"true",
"--option",
"accept-flake-config",
"true",
"-L",
*machine.nix_options,
"--flake",
f"{path}#{machine.name}",
nix_options = [
"--show-trace",
"--option",
"keep-going",
"true",
"--option",
"accept-flake-config",
"true",
"-L",
*machine.nix_options,
"--flake",
f"{path}#{machine.name}",
]
become_root = True
if machine._class_ == "nixos":
nix_options += [
"--fast",
"--build-host",
"",
]
become_root = True
if build_host:
become_root = False
nix_options += ["--target-host", target_host.target]
if machine._class_ == "nixos":
nix_options += [
"--fast",
"--build-host",
"",
]
if target_host.user != "root":
nix_options += ["--use-remote-sudo"]
switch_cmd = ["nixos-rebuild", "switch", *nix_options]
elif machine._class_ == "darwin":
# use absolute path to darwin-rebuild
switch_cmd = [
"/run/current-system/sw/bin/darwin-rebuild",
"switch",
*nix_options,
]
if build_host:
become_root = False
nix_options += ["--target-host", target_host.target]
remote_env = host.nix_ssh_env(control_master=False)
ret = host.run(
switch_cmd,
RunOpts(
check=False,
log=Log.BOTH,
msg_color=MsgColor(stderr=AnsiColor.DEFAULT),
needs_user_terminal=True,
),
extra_env=remote_env,
become_root=become_root,
control_master=False,
)
if target_host.user != "root":
nix_options += ["--use-remote-sudo"]
switch_cmd = ["nixos-rebuild", "switch", *nix_options]
elif machine._class_ == "darwin":
# use absolute path to darwin-rebuild
switch_cmd = [
"/run/current-system/sw/bin/darwin-rebuild",
"switch",
*nix_options,
]
if is_async_cancelled():
return
remote_env = host.nix_ssh_env(None, local_ssh=False)
# retry nixos-rebuild switch if the first attempt failed
if ret.returncode != 0:
is_mobile = machine.deployment.get("nixosMobileWorkaround", False)
# if the machine is mobile, we retry to deploy with the mobile workaround method
if is_mobile:
machine.info(
"Mobile machine detected, applying workaround deployment method"
)
ret = host.run(
switch_cmd,
["nixos--rebuild", "test", *nix_options] if is_mobile else switch_cmd,
RunOpts(
check=False,
log=Log.BOTH,
msg_color=MsgColor(stderr=AnsiColor.DEFAULT),
needs_user_terminal=True,
),
extra_env=remote_env,
become_root=become_root,
control_master=False,
)
if is_async_cancelled():
return
# retry nixos-rebuild switch if the first attempt failed
if ret.returncode != 0:
is_mobile = machine.deployment.get("nixosMobileWorkaround", False)
# if the machine is mobile, we retry to deploy with the mobile workaround method
if is_mobile:
machine.info(
"Mobile machine detected, applying workaround deployment method"
)
ret = host.run(
["nixos--rebuild", "test", *nix_options] if is_mobile else switch_cmd,
RunOpts(
log=Log.BOTH,
msg_color=MsgColor(stderr=AnsiColor.DEFAULT),
needs_user_terminal=True,
),
extra_env=remote_env,
become_root=become_root,
)
def deploy_machines(machines: list[Machine]) -> None:
"""