Пример #1
0
int main(int argc, char *argv[])
{
	log_options_t logopt = LOG_OPTS_STDERR_ONLY;
	job_desc_msg_t desc;
	resource_allocation_response_msg_t *alloc;
	time_t before, after;
	allocation_msg_thread_t *msg_thr;
	char **env = NULL;
	int status = 0;
	int retries = 0;
	pid_t pid  = getpid();
	pid_t tpgid = 0;
	pid_t rc_pid = 0;
	int i, rc = 0;
	static char *msg = "Slurm job queue full, sleeping and retrying.";
	slurm_allocation_callbacks_t callbacks;

	log_init(xbasename(argv[0]), logopt, 0, NULL);
	_set_exit_code();

	if (spank_init_allocator() < 0) {
		error("Failed to initialize plugin stack");
		exit(error_exit);
	}

	/* Be sure to call spank_fini when salloc exits
	 */
	if (atexit((void (*) (void)) spank_fini) < 0)
		error("Failed to register atexit handler for plugins: %m");


	if (initialize_and_process_args(argc, argv) < 0) {
		error("salloc parameter parsing");
		exit(error_exit);
	}
	/* reinit log with new verbosity (if changed by command line) */
	if (opt.verbose || opt.quiet) {
		logopt.stderr_level += opt.verbose;
		logopt.stderr_level -= opt.quiet;
		logopt.prefix_level = 1;
		log_alter(logopt, 0, NULL);
	}

	if (spank_init_post_opt() < 0) {
		error("Plugin stack post-option processing failed");
		exit(error_exit);
	}

	_set_spank_env();
	_set_submit_dir_env();
	if (opt.cwd && chdir(opt.cwd)) {
		error("chdir(%s): %m", opt.cwd);
		exit(error_exit);
	}

	if (opt.get_user_env_time >= 0) {
		char *user = uid_to_string(opt.uid);
		if (strcmp(user, "nobody") == 0) {
			error("Invalid user id %u: %m", (uint32_t)opt.uid);
			exit(error_exit);
		}
		env = env_array_user_default(user,
					     opt.get_user_env_time,
					     opt.get_user_env_mode);
		xfree(user);
		if (env == NULL)
			exit(error_exit);    /* error already logged */
		_set_rlimits(env);
	}

	/*
	 * Job control for interactive salloc sessions: only if ...
	 *
	 * a) input is from a terminal (stdin has valid termios attributes),
	 * b) controlling terminal exists (non-negative tpgid),
	 * c) salloc is not run in allocation-only (--no-shell) mode,
	 * NOTE: d and e below are configuration dependent
	 * d) salloc runs in its own process group (true in interactive
	 *    shells that support job control),
	 * e) salloc has been configured at compile-time to support background
	 *    execution and is not currently in the background process group.
	 */
	if (tcgetattr(STDIN_FILENO, &saved_tty_attributes) < 0) {
		/*
		 * Test existence of controlling terminal (tpgid > 0)
		 * after first making sure stdin is not redirected.
		 */
	} else if ((tpgid = tcgetpgrp(STDIN_FILENO)) < 0) {
		if (!opt.no_shell) {
			error("no controlling terminal: please set --no-shell");
			exit(error_exit);
		}
#ifdef SALLOC_RUN_FOREGROUND
	} else if ((!opt.no_shell) && (pid == getpgrp())) {
		if (tpgid == pid)
			is_interactive = true;
		while (tcgetpgrp(STDIN_FILENO) != pid) {
			if (!is_interactive) {
				error("Waiting for program to be placed in "
				      "the foreground");
				is_interactive = true;
			}
			killpg(pid, SIGTTIN);
		}
	}
#else
	} else if ((!opt.no_shell) && (getpgrp() == tcgetpgrp(STDIN_FILENO))) {
Пример #2
0
int main(int argc, char *argv[])
{
	log_options_t logopt = LOG_OPTS_STDERR_ONLY;
	job_desc_msg_t desc;
	resource_allocation_response_msg_t *alloc;
	time_t before, after;
	allocation_msg_thread_t *msg_thr;
	char **env = NULL;
	int status = 0;
	int retries = 0;
	pid_t pid  = getpid();
	pid_t tpgid = 0;
	pid_t rc_pid = 0;
	int i, rc = 0;
	static char *msg = "Slurm job queue full, sleeping and retrying.";
	slurm_allocation_callbacks_t callbacks;

	log_init(xbasename(argv[0]), logopt, 0, NULL);
	_set_exit_code();

	if (spank_init_allocator() < 0) {
		error("Failed to initialize plugin stack");
		exit(error_exit);
	}

	/* Be sure to call spank_fini when salloc exits
	 */
	if (atexit((void (*) (void)) spank_fini) < 0)
		error("Failed to register atexit handler for plugins: %m");


	if (initialize_and_process_args(argc, argv) < 0) {
		error("salloc parameter parsing");
		exit(error_exit);
	}
	/* reinit log with new verbosity (if changed by command line) */
	if (opt.verbose || opt.quiet) {
		logopt.stderr_level += opt.verbose;
		logopt.stderr_level -= opt.quiet;
		logopt.prefix_level = 1;
		log_alter(logopt, 0, NULL);
	}

	if (spank_init_post_opt() < 0) {
		error("Plugin stack post-option processing failed");
		exit(error_exit);
	}

	_set_spank_env();
	_set_submit_dir_env();
	if (opt.cwd && chdir(opt.cwd)) {
		error("chdir(%s): %m", opt.cwd);
		exit(error_exit);
	}

	if (opt.get_user_env_time >= 0) {
		char *user = uid_to_string(opt.uid);
		if (strcmp(user, "nobody") == 0) {
			error("Invalid user id %u: %m", (uint32_t)opt.uid);
			exit(error_exit);
		}
		env = env_array_user_default(user,
					     opt.get_user_env_time,
					     opt.get_user_env_mode);
		xfree(user);
		if (env == NULL)
			exit(error_exit);    /* error already logged */
		_set_rlimits(env);
	}

	/*
	 * Job control for interactive salloc sessions: only if ...
	 *
	 * a) input is from a terminal (stdin has valid termios attributes),
	 * b) controlling terminal exists (non-negative tpgid),
	 * c) salloc is not run in allocation-only (--no-shell) mode,
	 * d) salloc runs in its own process group (true in interactive
	 *    shells that support job control),
	 * e) salloc has been configured at compile-time to support background
	 *    execution and is not currently in the background process group.
	 */
	if (tcgetattr(STDIN_FILENO, &saved_tty_attributes) < 0) {
		/*
		 * Test existence of controlling terminal (tpgid > 0)
		 * after first making sure stdin is not redirected.
		 */
	} else if ((tpgid = tcgetpgrp(STDIN_FILENO)) < 0) {
		if (!opt.no_shell) {
			error("no controlling terminal: please set --no-shell");
			exit(error_exit);
		}
	} else if ((!opt.no_shell) && (pid == getpgrp())) {
		if (tpgid == pid)
			is_interactive = true;
#ifdef SALLOC_RUN_FOREGROUND
		while (tcgetpgrp(STDIN_FILENO) != pid) {
			if (!is_interactive) {
				error("Waiting for program to be placed in "
				      "the foreground");
				is_interactive = true;
			}
			killpg(pid, SIGTTIN);
		}
#endif
	}
	/*
	 * Reset saved tty attributes at exit, in case a child
	 * process died before properly resetting terminal.
	 */
	if (is_interactive)
		atexit (_reset_input_mode);

	/*
	 * Request a job allocation
	 */
	slurm_init_job_desc_msg(&desc);
	if (_fill_job_desc_from_opts(&desc) == -1) {
		exit(error_exit);
	}
	if (opt.gid != (gid_t) -1) {
		if (setgid(opt.gid) < 0) {
			error("setgid: %m");
			exit(error_exit);
		}
	}

	callbacks.ping = _ping_handler;
	callbacks.timeout = _timeout_handler;
	callbacks.job_complete = _job_complete_handler;
	callbacks.user_msg = _user_msg_handler;
	callbacks.node_fail = _node_fail_handler;
	/* create message thread to handle pings and such from slurmctld */
	msg_thr = slurm_allocation_msg_thr_create(&desc.other_port,
						  &callbacks);

	/* NOTE: Do not process signals in separate pthread. The signal will
	 * cause slurm_allocate_resources_blocking() to exit immediately. */
	for (i = 0; sig_array[i]; i++)
		xsignal(sig_array[i], _signal_while_allocating);

	before = time(NULL);
	while ((alloc = slurm_allocate_resources_blocking(&desc, opt.immediate,
					_pending_callback)) == NULL) {
		if ((errno != ESLURM_ERROR_ON_DESC_TO_RECORD_COPY) ||
		    (retries >= MAX_RETRIES))
			break;
		if (retries == 0)
			error("%s", msg);
		else
			debug("%s", msg);
		sleep (++retries);
	}

	/* become the user after the allocation has been requested. */
	if (opt.uid != (uid_t) -1) {
		if (setuid(opt.uid) < 0) {
			error("setuid: %m");
			exit(error_exit);
		}
	}
	if (alloc == NULL) {
		if (allocation_interrupted) {
			/* cancelled by signal */
			info("Job aborted due to signal");
		} else if (errno == EINTR) {
			error("Interrupted by signal."
			      "  Allocation request rescinded.");
		} else if (opt.immediate &&
			   ((errno == ETIMEDOUT) ||
			    (errno == ESLURM_NOT_TOP_PRIORITY) ||
			    (errno == ESLURM_NODES_BUSY))) {
			error("Unable to allocate resources: %m");
			error_exit = immediate_exit;
		} else {
			error("Failed to allocate resources: %m");
		}
		slurm_allocation_msg_thr_destroy(msg_thr);
		exit(error_exit);
	} else if (!allocation_interrupted) {
		/*
		 * Allocation granted!
		 */
		info("Granted job allocation %u", alloc->job_id);
		pending_job_id = alloc->job_id;
#ifdef HAVE_BG
		if (!_wait_bluegene_block_ready(alloc)) {
			if(!allocation_interrupted)
				error("Something is wrong with the "
				      "boot of the block.");
			goto relinquish;
		}
#else
		if (!_wait_nodes_ready(alloc)) {
			if(!allocation_interrupted)
				error("Something is wrong with the "
				      "boot of the nodes.");
			goto relinquish;
		}
#endif
	}

	after = time(NULL);
	if (opt.bell == BELL_ALWAYS
	    || (opt.bell == BELL_AFTER_DELAY
		&& ((after - before) > DEFAULT_BELL_DELAY))) {
		_ring_terminal_bell();
	}
	if (opt.no_shell)
		exit(0);
	if (allocation_interrupted) {
		/* salloc process received a signal after
		 * slurm_allocate_resources_blocking returned with the
		 * allocation, but before the new signal handlers were
		 * registered.
		 */
		goto relinquish;
	}

	/*
	 * Run the user's command.
	 */
	if (env_array_for_job(&env, alloc, &desc) != SLURM_SUCCESS)
		goto relinquish;

	/* Add default task count for srun, if not already set */
	if (opt.ntasks_set) {
		env_array_append_fmt(&env, "SLURM_NTASKS", "%d", opt.ntasks);
		/* keep around for old scripts */
		env_array_append_fmt(&env, "SLURM_NPROCS", "%d", opt.ntasks);
	}
	if (opt.cpus_per_task > 1) {
		env_array_append_fmt(&env, "SLURM_CPUS_PER_TASK", "%d",
				     opt.cpus_per_task);
	}
	if (opt.overcommit) {
		env_array_append_fmt(&env, "SLURM_OVERCOMMIT", "%d",
			opt.overcommit);
	}
	if (opt.acctg_freq >= 0) {
		env_array_append_fmt(&env, "SLURM_ACCTG_FREQ", "%d",
			opt.acctg_freq);
	}
	if (opt.network)
		env_array_append_fmt(&env, "SLURM_NETWORK", "%s", opt.network);

	env_array_set_environment(env);
	env_array_free(env);
	pthread_mutex_lock(&allocation_state_lock);
	if (allocation_state == REVOKED) {
		error("Allocation was revoked for job %u before command could "
		      "be run", alloc->job_id);
		pthread_mutex_unlock(&allocation_state_lock);
		if (slurm_complete_job(alloc->job_id, status) != 0) {
			error("Unable to clean up allocation for job %u: %m",
			      alloc->job_id);
		}
		return 1;
 	}
	allocation_state = GRANTED;
	pthread_mutex_unlock(&allocation_state_lock);

	/*  Ensure that salloc has initial terminal foreground control.  */
	if (is_interactive) {
		/*
		 * Ignore remaining job-control signals (other than those in
		 * sig_array, which at this state act like SIG_IGN).
		 */
		xsignal(SIGTSTP, SIG_IGN);
		xsignal(SIGTTIN, SIG_IGN);
		xsignal(SIGTTOU, SIG_IGN);

		pid = getpid();
		setpgid(pid, pid);

		tcsetpgrp(STDIN_FILENO, pid);
	}

	command_pid = _fork_command(command_argv);
	/*
	 * Wait for command to exit, OR for waitpid to be interrupted by a
	 * signal.  Either way, we are going to release the allocation next.
	 */
	if (command_pid > 0) {
		setpgid(command_pid, command_pid);
		if (is_interactive)
			tcsetpgrp(STDIN_FILENO, command_pid);

		/* NOTE: Do not process signals in separate pthread.
		 * The signal will cause waitpid() to exit immediately. */
		xsignal(SIGHUP,  _exit_on_signal);

		/* Use WUNTRACED to treat stopped children like terminated ones */
		do {
			rc_pid = waitpid(command_pid, &status, WUNTRACED);
		} while ((rc_pid == -1) && (!exit_flag));

		if ((rc_pid == -1) && (errno != EINTR))
			error("waitpid for %s failed: %m", command_argv[0]);
	}

	if (is_interactive)
		tcsetpgrp(STDIN_FILENO, pid);

	/*
	 * Relinquish the job allocation (if not already revoked).
	 */
relinquish:
	pthread_mutex_lock(&allocation_state_lock);
	if (allocation_state != REVOKED) {
		pthread_mutex_unlock(&allocation_state_lock);

		info("Relinquishing job allocation %d", alloc->job_id);
		if ((slurm_complete_job(alloc->job_id, status) != 0) &&
		    (slurm_get_errno() != ESLURM_ALREADY_DONE))
			error("Unable to clean up job allocation %d: %m",
			      alloc->job_id);
		pthread_mutex_lock(&allocation_state_lock);
		allocation_state = REVOKED;
	}
	pthread_mutex_unlock(&allocation_state_lock);

	slurm_free_resource_allocation_response_msg(alloc);
	slurm_allocation_msg_thr_destroy(msg_thr);

	/*
	 * Figure out what return code we should use.  If the user's command
	 * exited normally, return the user's return code.
	 */
	rc = 1;
	if (rc_pid != -1) {

		if (WIFEXITED(status)) {
			rc = WEXITSTATUS(status);
		} else if (WIFSIGNALED(status)) {
			verbose("Command \"%s\" was terminated by signal %d",
				command_argv[0], WTERMSIG(status));
			/* if we get these signals we return a normal
			 * exit since this was most likely sent from the
			 * user */
			switch(WTERMSIG(status)) {
			case SIGHUP:
			case SIGINT:
			case SIGQUIT:
			case SIGKILL:
				rc = 0;
				break;
			default:
				break;
			}
		}
	}
	return rc;
}