CCF/tests/infra/remote.py

856 строки
29 KiB
Python
Исходник Обычный вид История

2019-04-26 18:27:27 +03:00
# Copyright (c) Microsoft Corporation. All rights reserved.
# Licensed under the Apache 2.0 License.
import os
import time
from enum import Enum, auto
2019-04-26 18:27:27 +03:00
import paramiko
import subprocess
from contextlib import contextmanager
import infra.path
import ctypes
import signal
2019-08-08 18:11:46 +03:00
import re
2020-06-15 17:58:45 +03:00
import stat
import shutil
from collections import deque
2019-04-26 18:27:27 +03:00
from loguru import logger as LOG
DBG = os.getenv("DBG", "cgdb")
FILE_TIMEOUT = 60
2019-04-26 18:27:27 +03:00
_libc = ctypes.CDLL("libc.so.6")
def _term_on_pdeathsig():
# usr/include/linux/prctl.h: #define PR_SET_PDEATHSIG 1
_libc.prctl(1, signal.SIGTERM)
def popen(*args, **kwargs):
kwargs["preexec_fn"] = _term_on_pdeathsig
return subprocess.Popen(*args, **kwargs)
2019-04-26 18:27:27 +03:00
@contextmanager
def sftp_session(hostname):
client = paramiko.SSHClient()
client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
client.connect(hostname)
try:
session = client.open_sftp()
try:
yield session
finally:
session.close()
finally:
client.close()
def log_errors(out_path, err_path):
2019-08-28 12:57:45 +03:00
error_filter = ["[fail ]", "[fatal]"]
error_lines = []
2019-04-26 18:27:27 +03:00
try:
tail_lines = deque(maxlen=10)
2021-08-23 12:59:42 +03:00
with open(out_path, "r", errors="replace", encoding="utf-8") as lines:
2019-04-26 18:27:27 +03:00
for line in lines:
stripped_line = line.rstrip()
tail_lines.append(stripped_line)
if any(x in stripped_line for x in error_filter):
LOG.error("{}: {}".format(out_path, stripped_line))
error_lines.append(stripped_line)
if error_lines:
LOG.info(
"{} errors found, printing end of output for context:", len(error_lines)
)
for line in tail_lines:
LOG.info(line)
2019-04-26 18:27:27 +03:00
except IOError:
LOG.exception("Could not check output {} for errors".format(out_path))
fatal_error_lines = []
try:
2021-08-23 12:59:42 +03:00
with open(err_path, "r", errors="replace", encoding="utf-8") as lines:
fatal_error_lines = [
line
for line in lines.readlines()
if not line.startswith("[get_qpl_handle ")
]
if fatal_error_lines:
LOG.error(f"Contents of {err_path}:\n{''.join(fatal_error_lines)}")
except IOError:
LOG.exception("Could not read err output {}".format(err_path))
2021-01-14 22:42:40 +03:00
# See https://github.com/microsoft/CCF/issues/1701
ignore_fatal_errors = False
for line in fatal_error_lines:
if line.startswith("Tracer caught signal 11"):
ignore_fatal_errors = True
if ignore_fatal_errors:
fatal_error_lines = []
return error_lines, fatal_error_lines
2019-04-26 18:27:27 +03:00
class CmdMixin(object):
def set_perf(self):
self.cmd = [
"perf",
"record",
"--freq=1000",
"--call-graph=dwarf",
"-s",
] + self.cmd
def _get_perf(self, lines):
pattern = "=> (.*)tx/s"
2019-08-08 18:11:46 +03:00
for line in lines:
LOG.debug(line.decode())
res = re.search(pattern, line.decode())
if res:
return float(res.group(1))
raise ValueError(f"No performance result found (pattern is {pattern})")
2019-08-08 18:11:46 +03:00
2019-04-26 18:27:27 +03:00
class SSHRemote(CmdMixin):
def __init__(
2019-08-22 13:08:15 +03:00
self,
name,
hostname,
exe_files,
data_files,
cmd,
workspace,
label,
common_dir,
2019-08-22 13:08:15 +03:00
env=None,
log_format_json=None,
):
2019-04-26 18:27:27 +03:00
"""
Runs a command on a remote host, through an SSH connection. A temporary
directory is created, and some files can be shipped over. The command is
run out of that directory.
Note that the name matters, since the temporary directory that will be first
deleted, then created and populated is workspace/label_name. There is deliberately no
2019-04-26 18:27:27 +03:00
cleanup on shutdown, to make debugging/inspection possible.
setup() connects, creates the directory and ships over the files
start() runs the specified command
stop() disconnects, which shuts down the command via SIGHUP
"""
self.hostname = hostname
self.exe_files = exe_files
self.data_files = data_files
2019-04-26 18:27:27 +03:00
self.cmd = cmd
self.client = paramiko.SSHClient()
# this client (proc_client) is used to execute commands on the remote host since the main client uses pty
self.proc_client = paramiko.SSHClient()
2019-04-26 18:27:27 +03:00
self.client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
self.proc_client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
self.common_dir = common_dir
self.root = os.path.join(workspace, f"{label}_{name}")
2019-05-21 11:27:22 +03:00
self.name = name
self.env = env or {}
2019-08-22 13:08:15 +03:00
self.out = os.path.join(self.root, "out")
self.err = os.path.join(self.root, "err")
self.suspension_proc = None
self.pid_file = f"{os.path.basename(self.cmd[0])}.pid"
2020-03-06 15:56:13 +03:00
self._pid = None
2019-04-26 18:27:27 +03:00
def _rc(self, cmd):
LOG.info("[{}] {}".format(self.hostname, cmd))
_, stdout, _ = self.client.exec_command(cmd)
return stdout.channel.recv_exit_status()
def _connect(self):
LOG.debug("[{}] connect".format(self.hostname))
self.client.connect(self.hostname)
self.proc_client.connect(self.hostname)
2019-04-26 18:27:27 +03:00
def _setup_files(self):
assert self._rc("rm -rf {}".format(self.root)) == 0
assert self._rc("mkdir -p {}".format(self.root)) == 0
# For SSHRemote, both executable files (host and enclave) and data
# files (ledger, secrets) are copied to the remote
2019-04-26 18:27:27 +03:00
session = self.client.open_sftp()
for path in self.exe_files:
tgt_path = os.path.join(self.root, os.path.basename(path))
LOG.info("[{}] copy {} from {}".format(self.hostname, tgt_path, path))
session.put(path, tgt_path)
2020-04-01 11:06:55 +03:00
stat = os.stat(path)
session.chmod(tgt_path, stat.st_mode)
for path in self.data_files:
tgt_path = os.path.join(self.root, os.path.basename(path))
2020-06-15 17:58:45 +03:00
if os.path.isdir(path):
session.mkdir(tgt_path)
for f in os.listdir(path):
session.put(os.path.join(path, f), os.path.join(tgt_path, f))
else:
session.put(path, tgt_path)
LOG.info("[{}] copy {} from {}".format(self.hostname, tgt_path, path))
2019-04-26 18:27:27 +03:00
session.close()
def get(
self,
file_name,
dst_path,
timeout=FILE_TIMEOUT,
target_name=None,
pre_condition_func=lambda src_dir, _: True,
):
2019-04-26 18:27:27 +03:00
"""
Get file called `file_name` under the root of the remote. If the
2019-04-26 18:27:27 +03:00
file is missing, wait for timeout, and raise an exception.
2019-06-03 17:16:32 +03:00
If the file is present, it is copied to the CWD on the caller's
host, as `target_name` if it is set.
2019-04-26 18:27:27 +03:00
This call spins up a separate client because we don't want to interrupt
the main cmd that may be running.
"""
with sftp_session(self.hostname) as session:
2020-03-27 16:12:40 +03:00
end_time = time.time() + timeout
start_time = time.time()
while time.time() < end_time:
2019-04-26 18:27:27 +03:00
try:
target_name = target_name or file_name
2020-06-15 17:58:45 +03:00
fileattr = session.lstat(os.path.join(self.root, file_name))
if stat.S_ISDIR(fileattr.st_mode):
src_dir = os.path.join(self.root, file_name)
dst_dir = os.path.join(dst_path, file_name)
if os.path.exists(dst_dir):
shutil.rmtree(dst_dir)
os.makedirs(dst_dir)
if not pre_condition_func(src_dir, session.listdir):
raise RuntimeError(
"Pre-condition for getting remote files failed"
)
2020-06-15 17:58:45 +03:00
for f in session.listdir(src_dir):
session.get(
2020-08-26 18:52:41 +03:00
os.path.join(src_dir, f),
os.path.join(dst_dir, f),
2020-06-15 17:58:45 +03:00
)
else:
session.get(
os.path.join(self.root, file_name),
os.path.join(dst_path, target_name),
)
2019-04-26 18:27:27 +03:00
LOG.debug(
"[{}] found {} after {}s".format(
2020-03-27 16:12:40 +03:00
self.hostname, file_name, int(time.time() - start_time)
2019-04-26 18:27:27 +03:00
)
)
break
except FileNotFoundError:
2020-03-26 18:30:06 +03:00
time.sleep(0.1)
2019-04-26 18:27:27 +03:00
else:
raise ValueError(file_name)
2019-04-26 18:27:27 +03:00
def list_files(self, timeout=FILE_TIMEOUT):
2019-04-26 18:27:27 +03:00
files = []
with sftp_session(self.hostname) as session:
2020-03-27 16:12:40 +03:00
end_time = time.time() + timeout
while time.time() < end_time:
2019-04-26 18:27:27 +03:00
try:
files = session.listdir(self.root)
break
except Exception:
2020-03-26 18:30:06 +03:00
time.sleep(0.1)
2019-04-26 18:27:27 +03:00
else:
raise ValueError(self.root)
return files
def get_logs(self):
with sftp_session(self.hostname) as session:
2019-08-22 13:08:15 +03:00
for filepath in (self.err, self.out):
2019-04-26 18:27:27 +03:00
try:
local_file_name = "{}_{}_{}".format(
2020-08-26 18:52:41 +03:00
self.hostname,
self.name,
os.path.basename(filepath),
2019-05-21 11:27:22 +03:00
)
dst_path = os.path.join(self.common_dir, local_file_name)
session.get(filepath, dst_path)
LOG.info("Downloaded {}".format(dst_path))
except FileNotFoundError:
2019-04-26 18:27:27 +03:00
LOG.warning(
"Failed to download {} to {} (host: {})".format(
filepath, dst_path, self.hostname
)
2019-04-26 18:27:27 +03:00
)
2019-08-28 17:32:40 +03:00
def start(self):
2019-04-26 18:27:27 +03:00
"""
Start cmd on the remote host. stdout and err are captured to file locally.
We create a pty on the remote host under which to run the command, so as to
2019-04-26 18:27:27 +03:00
get a SIGHUP on disconnection.
"""
2020-04-17 14:53:01 +03:00
cmd = self.get_cmd()
2019-04-26 18:27:27 +03:00
LOG.info("[{}] {}".format(self.hostname, cmd))
2020-03-06 15:56:13 +03:00
self.client.exec_command(cmd, get_pty=True)
self.pid()
def pid(self):
if self._pid is None:
pid_path = os.path.join(self.root, self.pid_file)
2020-03-06 15:56:13 +03:00
time_left = 3
while time_left > 0:
_, stdout, _ = self.proc_client.exec_command(f'cat "{pid_path}"')
res = stdout.read().strip()
if res:
self._pid = int(res)
2020-03-06 15:56:13 +03:00
break
time_left = max(time_left - 0.1, 0)
if not time_left:
raise TimeoutError("Failed to read PID from file")
time.sleep(0.1)
return self._pid
def suspend(self):
2020-03-06 15:56:13 +03:00
_, stdout, _ = self.proc_client.exec_command(f"kill -STOP {self.pid()}")
2020-05-26 16:11:49 +03:00
if stdout.channel.recv_exit_status() != 0:
raise RuntimeError(f"Remote {self.name} could not be suspended")
def resume(self):
2020-03-06 15:56:13 +03:00
_, stdout, _ = self.proc_client.exec_command(f"kill -CONT {self.pid()}")
if stdout.channel.recv_exit_status() != 0:
2020-05-26 16:11:49 +03:00
raise RuntimeError(f"Could not resume remote {self.name} from suspension!")
2019-04-26 18:27:27 +03:00
def stop(self):
"""
Disconnect the client, and therefore shut down the command as well.
"""
LOG.info("[{}] closing".format(self.hostname))
self.get_logs()
errors, fatal_errors = log_errors(
os.path.join(self.common_dir, "{}_{}_out".format(self.hostname, self.name)),
os.path.join(self.common_dir, "{}_{}_err".format(self.hostname, self.name)),
2019-05-21 11:27:22 +03:00
)
2019-04-26 18:27:27 +03:00
self.client.close()
self.proc_client.close()
return errors, fatal_errors
2019-04-26 18:27:27 +03:00
def setup(self):
"""
Connect to the remote host, empty the temporary directory if it exsits,
and populate it with the initial set of files.
"""
self._connect()
self._setup_files()
2020-04-17 14:53:01 +03:00
def get_cmd(self):
env = " ".join(f"{key}={value}" for key, value in self.env.items())
cmd = " ".join(self.cmd)
return f"cd {self.root} && {env} {cmd} 1> {self.out} 2> {self.err} 0< /dev/null"
2019-04-26 18:27:27 +03:00
2020-04-21 10:19:05 +03:00
def debug_node_cmd(self):
2019-08-22 13:08:15 +03:00
cmd = " ".join(self.cmd)
return f"cd {self.root} && {DBG} --args {cmd}"
2019-04-26 18:27:27 +03:00
2019-06-07 18:35:22 +03:00
def _connect_new(self):
2019-04-26 18:27:27 +03:00
client = paramiko.SSHClient()
client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
client.connect(self.hostname)
2019-06-07 18:35:22 +03:00
return client
def check_done(self):
2019-06-07 18:35:22 +03:00
client = self._connect_new()
2019-04-26 18:27:27 +03:00
try:
_, stdout, _ = client.exec_command(f"ps -p {self.pid()}")
return stdout.channel.recv_exit_status() == 1
2019-04-26 18:27:27 +03:00
finally:
client.close()
def get_result(self, line_count):
2019-06-07 18:35:22 +03:00
client = self._connect_new()
try:
_, stdout, _ = client.exec_command(f"tail -{line_count} {self.out}")
2019-06-07 18:35:22 +03:00
if stdout.channel.recv_exit_status() == 0:
lines = stdout.read().splitlines()
result = lines[-line_count:]
return self._get_perf(result)
2019-06-07 18:35:22 +03:00
finally:
client.close()
2019-04-26 18:27:27 +03:00
class LocalRemote(CmdMixin):
def __init__(
2019-08-22 13:08:15 +03:00
self,
name,
hostname,
exe_files,
data_files,
cmd,
workspace,
label,
common_dir,
2019-08-22 13:08:15 +03:00
env=None,
log_format_json=None,
):
2019-04-26 18:27:27 +03:00
"""
Local Equivalent to the SSHRemote
"""
self.hostname = hostname
self.exe_files = exe_files
self.data_files = data_files
2019-04-26 18:27:27 +03:00
self.cmd = cmd
self.root = os.path.join(workspace, f"{label}_{name}")
self.common_dir = common_dir
2019-04-26 18:27:27 +03:00
self.proc = None
self.stdout = None
self.stderr = None
self.env = env
2019-06-07 18:35:22 +03:00
self.name = name
2019-08-22 13:08:15 +03:00
self.out = os.path.join(self.root, "out")
self.err = os.path.join(self.root, "err")
2019-04-26 18:27:27 +03:00
def _rc(self, cmd):
LOG.info("[{}] {}".format(self.hostname, cmd))
return subprocess.call(cmd, shell=True)
2020-06-15 17:58:45 +03:00
def _cp(self, src_path, dst_path):
if os.path.isdir(src_path):
assert self._rc("rm -rf {}".format(os.path.join(dst_path))) == 0
2020-06-15 17:58:45 +03:00
assert self._rc("cp -r {} {}".format(src_path, dst_path)) == 0
else:
assert self._rc("cp {} {}".format(src_path, dst_path)) == 0
2019-04-26 18:27:27 +03:00
def _setup_files(self):
assert self._rc("rm -rf {}".format(self.root)) == 0
assert self._rc("mkdir -p {}".format(self.root)) == 0
for path in self.exe_files:
2020-03-03 20:43:12 +03:00
dst_path = os.path.normpath(os.path.join(self.root, os.path.basename(path)))
src_path = os.path.normpath(os.path.join(os.getcwd(), path))
assert self._rc("ln -s {} {}".format(src_path, dst_path)) == 0
for path in self.data_files:
dst_path = os.path.join(self.root, os.path.basename(path))
2020-06-15 17:58:45 +03:00
self._cp(path, dst_path)
def get(
self,
src_path,
dst_path,
timeout=FILE_TIMEOUT,
target_name=None,
pre_condition_func=lambda src_dir, _: True,
):
path = os.path.join(self.root, src_path)
2020-03-27 16:12:40 +03:00
end_time = time.time() + timeout
while time.time() < end_time:
2019-04-26 18:27:27 +03:00
if os.path.exists(path):
break
2020-03-26 18:30:06 +03:00
time.sleep(0.1)
2019-04-26 18:27:27 +03:00
else:
raise ValueError(path)
if not pre_condition_func(path, os.listdir):
raise RuntimeError("Pre-condition for getting remote files failed")
target_name = target_name or os.path.basename(src_path)
self._cp(path, os.path.join(dst_path, target_name))
2019-04-26 18:27:27 +03:00
def list_files(self):
return os.listdir(self.root)
def start(self):
2019-04-26 18:27:27 +03:00
"""
Start cmd. stdout and err are captured to file locally.
"""
2020-04-17 14:53:01 +03:00
cmd = self.get_cmd()
LOG.info(f"[{self.hostname}] {cmd} (env: {self.env})")
2019-08-22 13:08:15 +03:00
self.stdout = open(self.out, "wb")
self.stderr = open(self.err, "wb")
self.proc = popen(
self.cmd,
cwd=self.root,
stdout=self.stdout,
stderr=self.stderr,
env=self.env,
2019-04-26 18:27:27 +03:00
)
def suspend(self):
self.proc.send_signal(signal.SIGSTOP)
def resume(self):
self.proc.send_signal(signal.SIGCONT)
2019-04-26 18:27:27 +03:00
def stop(self):
"""
Disconnect the client, and therefore shut down the command as well.
"""
LOG.info("[{}] closing".format(self.hostname))
if self.proc:
self.proc.terminate()
2020-01-30 16:28:38 +03:00
self.proc.wait(10)
2019-04-26 18:27:27 +03:00
if self.stdout:
self.stdout.close()
if self.stderr:
self.stderr.close()
return log_errors(self.out, self.err)
2019-04-26 18:27:27 +03:00
def setup(self):
"""
Empty the temporary directory if it exists,
and populate it with the initial set of files.
"""
self._setup_files()
2020-04-17 14:53:01 +03:00
def get_cmd(self):
2019-08-22 13:08:15 +03:00
cmd = " ".join(self.cmd)
return f"cd {self.root} && {cmd} 1> {self.out} 2> {self.err}"
2019-04-26 18:27:27 +03:00
2020-04-21 10:19:05 +03:00
def debug_node_cmd(self):
2019-08-22 13:08:15 +03:00
cmd = " ".join(self.cmd)
return f"cd {self.root} && {DBG} --args {cmd}"
2019-04-26 18:27:27 +03:00
def check_done(self):
return self.proc.poll() is not None
2019-10-11 17:14:31 +03:00
def get_result(self, line_count):
2019-08-22 13:08:15 +03:00
with open(self.out, "rb") as out:
2019-06-07 18:35:22 +03:00
lines = out.read().splitlines()
result = lines[-line_count:]
return self._get_perf(result)
2019-06-07 18:35:22 +03:00
2019-04-26 18:27:27 +03:00
CCF_TO_OE_LOG_LEVEL = {
"trace": "VERBOSE",
"debug": "INFO",
"info": "WARNING",
"fail": "ERROR",
"fatal": "FATAL",
}
def make_address(host, port=None):
if port is not None:
return f"{host}:{port}"
else:
return f"{host}:0"
2019-04-26 18:27:27 +03:00
class CCFRemote(object):
BIN = "cchost"
DEPS = []
def __init__(
self,
2019-08-28 12:57:45 +03:00
start_type,
2019-04-26 18:27:27 +03:00
lib_path,
local_node_id,
2019-04-26 18:27:27 +03:00
host,
pubhost,
2019-08-06 19:38:19 +03:00
node_port,
rpc_port,
node_client_host,
2019-04-26 18:27:27 +03:00
remote_class,
enclave_type,
workspace,
label,
common_dir,
2019-08-28 12:57:45 +03:00
target_rpc_address=None,
members_info=None,
2020-08-27 14:13:40 +03:00
snapshot_dir=None,
join_timer=None,
2019-08-28 12:57:45 +03:00
host_log_level="info",
2020-08-27 10:17:42 +03:00
sig_tx_interval=5000,
sig_ms_interval=1000,
raft_election_timeout_ms=1000,
bft_view_change_timeout_ms=5000,
2020-09-09 12:59:57 +03:00
consensus="cft",
worker_threads=0,
memory_reserve_startup=0,
2021-03-22 22:21:44 +03:00
constitution=None,
2020-06-15 17:58:45 +03:00
ledger_dir=None,
read_only_ledger_dir=None, # Read-only ledger dir to copy to node directory
common_read_only_ledger_dir=None, # Read-only ledger dir for all nodes
log_format_json=None,
binary_dir=".",
2020-08-27 10:17:42 +03:00
ledger_chunk_bytes=(5 * 1000 * 1000),
san=None,
2020-08-27 10:17:42 +03:00
snapshot_tx_interval=None,
max_open_sessions=None,
max_open_sessions_hard=None,
jwt_key_refresh_interval_s=None,
curve_id=None,
client_connection_timeout_ms=None,
additional_raw_node_args=None,
2019-04-26 18:27:27 +03:00
):
"""
Run a ccf binary on a remote host.
"""
2019-08-28 12:57:45 +03:00
self.start_type = start_type
self.local_node_id = local_node_id
self.pem = f"{local_node_id}.pem"
self.node_address_path = f"{local_node_id}.node_address"
self.rpc_address_path = f"{local_node_id}.rpc_address"
self.BIN = infra.path.build_bin_path(
self.BIN, enclave_type, binary_dir=binary_dir
)
self.common_dir = common_dir
2020-06-15 17:58:45 +03:00
self.ledger_dir = os.path.normpath(ledger_dir) if ledger_dir else None
2020-06-15 17:58:45 +03:00
self.ledger_dir_name = (
os.path.basename(self.ledger_dir)
if self.ledger_dir
else f"{local_node_id}.ledger"
2019-04-26 18:27:27 +03:00
)
self.read_only_ledger_dir = read_only_ledger_dir
2021-06-04 16:30:19 +03:00
self.read_only_ledger_dir_name = (
os.path.basename(self.read_only_ledger_dir)
if self.read_only_ledger_dir
else None
)
self.common_read_only_ledger_dir = common_read_only_ledger_dir
2020-08-27 14:13:40 +03:00
self.snapshot_dir = os.path.normpath(snapshot_dir) if snapshot_dir else None
self.snapshot_dir_name = (
os.path.basename(self.snapshot_dir)
if self.snapshot_dir
else f"{local_node_id}.snapshots"
2020-08-27 14:13:40 +03:00
)
2019-04-26 18:27:27 +03:00
2019-08-28 12:57:45 +03:00
exe_files = [self.BIN, lib_path] + self.DEPS
2020-06-15 17:58:45 +03:00
data_files = [self.ledger_dir] if self.ledger_dir else []
2020-09-29 14:37:15 +03:00
data_files += [self.snapshot_dir] if self.snapshot_dir else []
2019-08-28 12:57:45 +03:00
2020-04-01 11:06:55 +03:00
# exe_files may be relative or absolute. The remote implementation should
2020-02-07 17:29:02 +03:00
# copy (or symlink) to the target workspace, and then node will be able
# to reference the destination file locally in the target workspace.
2020-04-01 11:06:55 +03:00
bin_path = os.path.join(".", os.path.basename(self.BIN))
2020-02-07 17:29:02 +03:00
enclave_path = os.path.join(".", os.path.basename(lib_path))
2020-03-09 15:52:43 +03:00
election_timeout_arg = (
f"--bft-view-change-timeout-ms={bft_view_change_timeout_ms}"
2020-09-09 12:59:57 +03:00
if consensus == "bft"
else f"--raft-election-timeout-ms={raft_election_timeout_ms}"
2020-03-09 15:52:43 +03:00
)
2019-08-28 17:32:40 +03:00
cmd = [
2020-04-01 11:06:55 +03:00
bin_path,
2020-02-07 17:29:02 +03:00
f"--enclave-file={enclave_path}",
2019-08-28 17:32:40 +03:00
f"--enclave-type={enclave_type}",
f"--node-address={make_address(host, node_port)}",
f"--node-address-file={self.node_address_path}",
f"--rpc-address={make_address(host, rpc_port)}",
f"--rpc-address-file={self.rpc_address_path}",
f"--public-rpc-address={make_address(pubhost, rpc_port)}",
2020-06-15 17:58:45 +03:00
f"--ledger-dir={self.ledger_dir_name}",
f"--snapshot-dir={self.snapshot_dir_name}",
2019-08-28 17:32:40 +03:00
f"--node-cert-file={self.pem}",
f"--host-log-level={host_log_level}",
2020-03-09 15:52:43 +03:00
election_timeout_arg,
f"--consensus={consensus}",
2020-03-27 16:12:40 +03:00
f"--worker-threads={worker_threads}",
2019-08-28 17:32:40 +03:00
]
if node_client_host:
cmd += [f"--node-client-interface={node_client_host}"]
if log_format_json:
cmd += ["--log-format-json"]
2019-09-04 14:03:33 +03:00
2020-08-27 10:17:42 +03:00
if sig_tx_interval:
cmd += [f"--sig-tx-interval={sig_tx_interval}"]
2019-08-28 17:32:40 +03:00
2020-08-27 10:17:42 +03:00
if sig_ms_interval:
cmd += [f"--sig-ms-interval={sig_ms_interval}"]
2019-08-28 17:32:40 +03:00
if memory_reserve_startup:
cmd += [f"--memory-reserve-startup={memory_reserve_startup}"]
2020-08-27 10:17:42 +03:00
if ledger_chunk_bytes:
cmd += [f"--ledger-chunk-bytes={ledger_chunk_bytes}"]
2020-06-15 17:58:45 +03:00
if san:
cmd += [f"--san={s}" for s in san]
2020-08-27 10:17:42 +03:00
if snapshot_tx_interval:
cmd += [f"--snapshot-tx-interval={snapshot_tx_interval}"]
if max_open_sessions:
cmd += [f"--max-open-sessions={max_open_sessions}"]
if max_open_sessions_hard:
cmd += [f"--max-open-sessions-hard={max_open_sessions_hard}"]
if jwt_key_refresh_interval_s:
cmd += [f"--jwt-key-refresh-interval-s={jwt_key_refresh_interval_s}"]
if self.read_only_ledger_dir is not None:
2021-06-04 16:30:19 +03:00
cmd += [f"--read-only-ledger-dir={self.read_only_ledger_dir_name}"]
data_files += [os.path.join(self.common_dir, self.read_only_ledger_dir)]
if self.common_read_only_ledger_dir is not None:
cmd += [f"--read-only-ledger-dir={self.common_read_only_ledger_dir}"]
if curve_id is not None:
cmd += [f"--curve-id={curve_id.name}"]
if client_connection_timeout_ms:
cmd += [f"--client-connection-timeout-ms={client_connection_timeout_ms}"]
if additional_raw_node_args:
for s in additional_raw_node_args:
cmd += [str(s)]
if start_type == StartType.new:
2019-08-28 17:32:40 +03:00
cmd += [
"start",
"--network-cert-file=networkcert.pem",
2019-08-28 17:32:40 +03:00
]
for fragment in constitution:
cmd.append(f"--constitution={os.path.basename(fragment)}")
data_files += [
os.path.join(self.common_dir, os.path.basename(fragment))
]
if members_info is None:
raise ValueError(
2020-11-10 18:34:58 +03:00
"Starting node should be given at least one member info"
)
2020-10-09 19:10:37 +03:00
for mi in members_info:
2020-11-10 18:34:58 +03:00
member_info_cmd = f"--member-info={mi.certificate_file}"
if mi.encryption_pub_key_file is not None:
member_info_cmd += f",{mi.encryption_pub_key_file}"
elif mi.member_data_file is not None:
member_info_cmd += ","
if mi.member_data_file is not None:
member_info_cmd += f",{mi.member_data_file}"
2020-10-09 19:10:37 +03:00
for mf in mi:
2020-11-10 18:34:58 +03:00
if mf is not None:
data_files.append(os.path.join(self.common_dir, mf))
cmd += [member_info_cmd]
2019-08-28 17:32:40 +03:00
elif start_type == StartType.join:
cmd += [
"join",
"--network-cert-file=networkcert.pem",
f"--target-rpc-address={target_rpc_address}",
f"--join-timer={join_timer}",
2019-08-28 17:32:40 +03:00
]
data_files += [os.path.join(self.common_dir, "networkcert.pem")]
2020-11-10 18:34:58 +03:00
2019-08-28 17:32:40 +03:00
elif start_type == StartType.recover:
cmd += ["recover", "--network-cert-file=networkcert.pem"]
2020-11-10 18:34:58 +03:00
2019-08-28 17:32:40 +03:00
else:
raise ValueError(
f"Unexpected CCFRemote start type {start_type}. Should be start, join or recover"
)
env = {}
2019-08-09 15:53:46 +03:00
if enclave_type == "virtual":
env["UBSAN_OPTIONS"] = "print_stacktrace=1"
2019-05-28 18:33:57 +03:00
2019-08-28 12:57:45 +03:00
oe_log_level = CCF_TO_OE_LOG_LEVEL.get(host_log_level)
if oe_log_level:
env["OE_LOG_LEVEL"] = oe_log_level
2019-04-26 18:27:27 +03:00
self.remote = remote_class(
2019-08-29 18:01:57 +03:00
local_node_id,
host,
exe_files,
data_files,
cmd,
workspace,
label,
common_dir,
2019-08-29 18:01:57 +03:00
env,
log_format_json,
2019-04-26 18:27:27 +03:00
)
def setup(self):
self.remote.setup()
def start(self):
2019-08-28 17:32:40 +03:00
self.remote.start()
def suspend(self):
return self.remote.suspend()
def resume(self):
self.remote.resume()
def get_startup_files(self, dst_path):
self.remote.get(self.pem, dst_path)
self.remote.get(self.node_address_path, dst_path)
self.remote.get(self.rpc_address_path, dst_path)
if self.start_type in {StartType.new, StartType.recover}:
self.remote.get("networkcert.pem", dst_path)
2019-04-26 18:27:27 +03:00
def debug_node_cmd(self):
2020-04-17 14:53:01 +03:00
return self.remote.debug_node_cmd()
2019-04-26 18:27:27 +03:00
def stop(self):
errors, fatal_errors = [], []
2019-04-26 18:27:27 +03:00
try:
errors, fatal_errors = self.remote.stop()
2019-04-26 18:27:27 +03:00
except Exception:
LOG.exception("Failed to shut down {} cleanly".format(self.local_node_id))
return errors, fatal_errors
2019-04-26 18:27:27 +03:00
def check_done(self):
return self.remote.check_done()
2019-04-26 18:27:27 +03:00
def set_perf(self):
self.remote.set_perf()
2020-10-14 18:49:34 +03:00
# For now, it makes sense to default include_read_only_dirs to False
# but when nodes started from snapshots are fully supported in the test
# suite, this argument will probably default to True (or be deleted entirely)
def get_ledger(self, ledger_dir_name, include_read_only_dirs=False):
self.remote.get(
self.ledger_dir_name,
self.common_dir,
target_name=ledger_dir_name,
)
read_only_ledger_dirs = []
if include_read_only_dirs and self.read_only_ledger_dir is not None:
read_only_ledger_dir_name = (
f"{ledger_dir_name}.ro"
if ledger_dir_name
else self.read_only_ledger_dir
)
self.remote.get(
os.path.basename(self.read_only_ledger_dir),
self.common_dir,
target_name=read_only_ledger_dir_name,
)
read_only_ledger_dirs.append(
os.path.join(self.common_dir, read_only_ledger_dir_name)
)
return (
os.path.join(self.common_dir, ledger_dir_name),
read_only_ledger_dirs,
)
2019-04-26 18:27:27 +03:00
2020-12-03 20:49:54 +03:00
def get_snapshots(self):
self.remote.get(self.snapshot_dir_name, self.common_dir)
return os.path.join(self.common_dir, self.snapshot_dir_name)
def get_committed_snapshots(self, pre_condition_func=lambda src_dir, _: True):
self.remote.get(
self.snapshot_dir_name,
self.common_dir,
pre_condition_func=pre_condition_func,
)
2020-08-27 14:13:40 +03:00
return os.path.join(self.common_dir, self.snapshot_dir_name)
def log_path(self):
return self.remote.out
def ledger_paths(self):
paths = [os.path.join(self.remote.root, self.ledger_dir_name)]
2021-06-04 16:30:19 +03:00
if self.read_only_ledger_dir_name is not None:
paths += [os.path.join(self.remote.root, self.read_only_ledger_dir_name)]
return paths
2019-04-26 18:27:27 +03:00
2019-08-28 12:57:45 +03:00
class StartType(Enum):
new = auto()
join = auto()
recover = auto()