def _run_helper(self,
final_cmd,
with_output=False,
exit_on_fail=False,
silent=False):
"""Run a command that was already setup with SSH and `bash` settings.
Args:
cmd (List[str]):
Full command to run. Should include SSH options and other
processing that we do.
with_output (bool):
If `with_output` is `True`, command stdout and stderr
will be captured and returned.
exit_on_fail (bool):
If `exit_on_fail` is `True`, the process will exit
if the command fails (exits with a code other than 0).
Raises:
ProcessRunnerError if using new log style and disabled
login shells.
click.ClickException if using login shells.
"""
try:
# For now, if the output is needed we just skip the new logic.
# In the future we could update the new logic to support
# capturing output, but it is probably not needed.
if not cli_logger.old_style and not with_output:
return run_cmd_redirected(
final_cmd,
process_runner=self.process_runner,
silent=silent,
use_login_shells=is_using_login_shells())
if with_output:
return self.process_runner.check_output(final_cmd)
else:
return self.process_runner.check_call(final_cmd)
except subprocess.CalledProcessError as e:
quoted_cmd = " ".join(final_cmd[:-1] + [quote(final_cmd[-1])])
if not cli_logger.old_style and not is_using_login_shells():
raise ProcessRunnerError("Command failed",
"ssh_command_failed",
code=e.returncode,
command=quoted_cmd)
if exit_on_fail:
raise click.ClickException(
"Command failed:\n\n {}\n".format(quoted_cmd)) from None
else:
fail_msg = "SSH command failed."
if is_output_redirected():
fail_msg += " See above for the output from the failure."
raise click.ClickException(fail_msg) from None
def run(self):
cli_logger.old_info(logger, "{}Updating to {}", self.log_prefix,
self.runtime_hash)
if cmd_output_util.does_allow_interactive(
) and cmd_output_util.is_output_redirected():
# this is most probably a bug since the user has no control
# over these settings
msg = ("Output was redirected for an interactive command. "
"Either do not pass `--redirect-command-output` "
"or also pass in `--use-normal-shells`.")
cli_logger.abort(msg)
raise click.ClickException(msg)
try:
with LogTimer(self.log_prefix +
"Applied config {}".format(self.runtime_hash)):
self.do_update()
except Exception as e:
error_str = str(e)
if hasattr(e, "cmd"):
error_str = "(Exit Status {}) {}".format(
e.returncode, " ".join(e.cmd))
self.provider.set_node_tags(
self.node_id, {TAG_RAY_NODE_STATUS: STATUS_UPDATE_FAILED})
cli_logger.error("New status: {}", cf.bold(STATUS_UPDATE_FAILED))
cli_logger.old_error(logger, "{}Error executing: {}\n",
self.log_prefix, error_str)
cli_logger.error("!!!")
if hasattr(e, "cmd"):
cli_logger.error(
"Setup command `{}` failed with exit code {}. stderr:",
cf.bold(e.cmd), e.returncode)
else:
cli_logger.verbose_error("{}", str(vars(e)))
# todo: handle this better somehow?
cli_logger.error("{}", str(e))
# todo: print stderr here
cli_logger.error("!!!")
cli_logger.newline()
if isinstance(e, click.ClickException):
# todo: why do we ignore this here
return
raise
tags_to_set = {
TAG_RAY_NODE_STATUS: STATUS_UP_TO_DATE,
TAG_RAY_RUNTIME_CONFIG: self.runtime_hash,
}
if self.file_mounts_contents_hash is not None:
tags_to_set[
TAG_RAY_FILE_MOUNTS_CONTENTS] = self.file_mounts_contents_hash
self.provider.set_node_tags(self.node_id, tags_to_set)
cli_logger.labeled_value("New status", STATUS_UP_TO_DATE)
self.exitcode = 0
def do_update(self):
self.provider.set_node_tags(
self.node_id, {TAG_RAY_NODE_STATUS: STATUS_WAITING_FOR_SSH})
cli_logger.labeled_value("New status", STATUS_WAITING_FOR_SSH)
deadline = time.time() + NODE_START_WAIT_S
self.wait_ready(deadline)
node_tags = self.provider.node_tags(self.node_id)
logger.debug("Node tags: {}".format(str(node_tags)))
# runtime_hash will only change whenever the user restarts
# or updates their cluster with `get_or_create_head_node`
if node_tags.get(TAG_RAY_RUNTIME_CONFIG) == self.runtime_hash and (
self.file_mounts_contents_hash is None
or node_tags.get(TAG_RAY_FILE_MOUNTS_CONTENTS)
== self.file_mounts_contents_hash):
# todo: we lie in the confirmation message since
# full setup might be cancelled here
cli_logger.print(
"Configuration already up to date, "
"skipping file mounts, initalization and setup commands.",
_numbered=("[]", "2-5", 6))
cli_logger.old_info(logger,
"{}{} already up-to-date, skip to ray start",
self.log_prefix, self.node_id)
# When resuming from a stopped instance the runtime_hash may be the
# same, but the container will not be started.
self.cmd_runner.run_init(as_head=self.is_head_node,
file_mounts=self.file_mounts)
else:
cli_logger.print("Updating cluster configuration.",
_tags=dict(hash=self.runtime_hash))
self.provider.set_node_tags(
self.node_id, {TAG_RAY_NODE_STATUS: STATUS_SYNCING_FILES})
cli_logger.labeled_value("New status", STATUS_SYNCING_FILES)
self.sync_file_mounts(self.rsync_up, step_numbers=(2, 6))
# Only run setup commands if runtime_hash has changed because
# we don't want to run setup_commands every time the head node
# file_mounts folders have changed.
if node_tags.get(TAG_RAY_RUNTIME_CONFIG) != self.runtime_hash:
# Run init commands
self.provider.set_node_tags(
self.node_id, {TAG_RAY_NODE_STATUS: STATUS_SETTING_UP})
cli_logger.labeled_value("New status", STATUS_SETTING_UP)
if self.initialization_commands:
with cli_logger.group("Running initialization commands",
_numbered=("[]", 3, 5)):
with LogTimer(self.log_prefix +
"Initialization commands",
show_status=True):
for cmd in self.initialization_commands:
try:
# Overriding the existing SSHOptions class
# with a new SSHOptions class that uses
# this ssh_private_key as its only __init__
# argument.
# Run outside docker.
self.cmd_runner.run(
cmd,
ssh_options_override_ssh_key=self.
auth_config.get("ssh_private_key"),
run_env="host")
except ProcessRunnerError as e:
if e.msg_type == "ssh_command_failed":
cli_logger.error("Failed.")
cli_logger.error(
"See above for stderr.")
raise click.ClickException(
"Initialization command failed."
) from None
else:
cli_logger.print("No initialization commands to run.",
_numbered=("[]", 3, 6))
self.cmd_runner.run_init(as_head=self.is_head_node,
file_mounts=self.file_mounts)
if self.setup_commands:
with cli_logger.group(
"Running setup commands",
# todo: fix command numbering
_numbered=("[]", 4, 6)):
with LogTimer(self.log_prefix + "Setup commands",
show_status=True):
total = len(self.setup_commands)
for i, cmd in enumerate(self.setup_commands):
if cli_logger.verbosity == 0 and len(cmd) > 30:
cmd_to_print = cf.bold(cmd[:30]) + "..."
else:
cmd_to_print = cf.bold(cmd)
cli_logger.print("{}",
cmd_to_print,
_numbered=("()", i, total))
try:
# Runs in the container if docker is in use
self.cmd_runner.run(cmd, run_env="auto")
except ProcessRunnerError as e:
if e.msg_type == "ssh_command_failed":
cli_logger.error("Failed.")
cli_logger.error(
"See above for stderr.")
raise click.ClickException(
"Setup command failed.")
else:
cli_logger.print("No setup commands to run.",
_numbered=("[]", 4, 6))
with cli_logger.group("Starting the Ray runtime",
_numbered=("[]", 6, 6)):
with LogTimer(self.log_prefix + "Ray start commands",
show_status=True):
for cmd in self.ray_start_commands:
if self.node_resources:
env_vars = {
ray_constants.RESOURCES_ENVIRONMENT_VARIABLE:
self.node_resources
}
else:
env_vars = {}
try:
old_redirected = cmd_output_util.is_output_redirected()
cmd_output_util.set_output_redirected(False)
# Runs in the container if docker is in use
self.cmd_runner.run(cmd,
environment_variables=env_vars,
run_env="auto")
cmd_output_util.set_output_redirected(old_redirected)
except ProcessRunnerError as e:
if e.msg_type == "ssh_command_failed":
cli_logger.error("Failed.")
cli_logger.error("See above for stderr.")
raise click.ClickException("Start command failed.")
return self.process_runner.check_call(final_cmd)
except subprocess.CalledProcessError as e:
quoted_cmd = " ".join(final_cmd[:-1] + [quote(final_cmd[-1])])
if not cli_logger.old_style and not is_using_login_shells():
raise ProcessRunnerError(
"Command failed",
"ssh_command_failed",
code=e.returncode,
command=quoted_cmd)
if exit_on_fail:
raise click.ClickException(
"Command failed:\n\n {}\n".format(quoted_cmd)) from None
else:
fail_msg = "SSH command failed."
if is_output_redirected():
fail_msg += " See above for the output from the failure."
raise click.ClickException(fail_msg) from None
def run(
self,
cmd,
timeout=120,
exit_on_fail=False,
port_forward=None,
with_output=False,
environment_variables: Dict[str, object] = None,
run_env="auto", # Unused argument.
ssh_options_override_ssh_key="",
shutdown_after_run=False,
):
