simulator-agent.cc 13.9 KB
Newer Older
1 2
/*
 * EMULAB-COPYRIGHT
3
 * Copyright (c) 2004-2011 University of Utah and the Flux Group.
4 5 6 7 8 9 10 11 12 13 14 15
 * All rights reserved.
 */

/**
 * @file simulator-agent.cc
 *
 * Implementation file for the SIMULATOR agent.
 */

#include "config.h"

#include <stdlib.h>
16 17
#include <errno.h>
#include <string.h>
18 19 20 21

#include "popenf.h"
#include "systemf.h"
#include "rpc.h"
22
#include "log.h"
23 24
#include "simulator-agent.h"

25 26
using namespace emulab;

27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80
/**
 * A "looper" function for the simulator agent that dequeues and processes
 * events destined for the Simulator object.  This function will be passed to
 * pthread_create when a new thread needs to be created to handle events.
 *
 * @param arg The simulator agent object to handle events for.
 * @return NULL
 *
 * @see send_report
 * @see add_report_data
 * @see local_agent_queue
 * @see local_agent_dequeue
 */
static void *simulator_agent_looper(void *arg);

simulator_agent_t create_simulator_agent(void)
{
	simulator_agent_t sa, retval;

	if ((sa = (simulator_agent_t)
	     malloc(sizeof(struct _simulator_agent))) == NULL) {
		retval = NULL;
		errno = ENOMEM;
	}
	else if (local_agent_init(&sa->sa_local_agent) != 0) {
		retval = NULL;
	}
	else {
		int lpc;
		
		sa->sa_local_agent.la_looper = simulator_agent_looper;
		lnNewList(&sa->sa_error_records);
		for (lpc = 0; lpc < SA_RDK_MAX; lpc++)
			sa->sa_report_data[lpc] = NULL;
		
		retval = sa;
		sa = NULL;
	}

	free(sa);
	sa = NULL;

	return retval;
}

int simulator_agent_invariant(simulator_agent_t sa)
{
	assert(sa != NULL);
	assert(local_agent_invariant(&sa->sa_local_agent));
	lnCheck(&sa->sa_error_records);
	
	return 1;
}

81 82
int add_report_data(simulator_agent_t sa,
		    sa_report_data_kind_t rdk,
83 84
		    char *data,
		    unsigned long flags)
85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108
{
	char *new_data;
	int retval;

	assert(sa != NULL);
	assert(rdk >= 0);
	assert(rdk < SA_RDK_MAX);
	assert(data != NULL);

	if ((new_data = (char *)realloc(sa->sa_report_data[rdk],
					((sa->sa_report_data[rdk] != NULL) ?
					 strlen(sa->sa_report_data[rdk]) : 0) +
					strlen(data) +
					1 +
					1)) == NULL) {
		retval = -1;
		errno = ENOMEM;
	}
	else {
		if (sa->sa_report_data[rdk] == NULL)
			new_data[0] = '\0'; // Need to clear the fresh malloc.
		sa->sa_report_data[rdk] = new_data;
		
		strcat(sa->sa_report_data[rdk], data);
109 110
		if ((flags & SA_RDF_NEWLINE) &&
		    (data[strlen(data) - 1] != '\n')) {
111
			strcat(sa->sa_report_data[rdk], "\n");
112 113
		}

114 115 116 117 118 119
		retval = 0;
	}
	
	return retval;
}

120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
static int remap_experiment(simulator_agent_t sa, int token)
{
	char nsfile[BUFSIZ];
	EmulabResponse er;
	int retval;

	snprintf(nsfile, sizeof(nsfile),
		 "/proj/%s/exp/%s/tbdata/%s-modify.ns",
		 pid, eid, eid);
	if (access(nsfile, R_OK) == -1) {
		snprintf(nsfile, sizeof(nsfile),
			 "/proj/%s/exp/%s/tbdata/%s.ns",
			 pid, eid, eid);
	}
	RPC_grab();
	retval = RPC_invoke("experiment.modify",
136 137 138 139 140 141 142 143
			    &er,
			    SPA_String, "proj", pid,
			    SPA_String, "exp", eid,
			    SPA_Boolean, "wait", true,
			    SPA_Boolean, "reboot", true,
			    SPA_Boolean, "restart_eventsys", true,
			    SPA_String, "nsfilepath", nsfile,
			    SPA_TAG_DONE);
144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167
	RPC_drop();

	if (retval != 0) {
		rename("tbdata/feedback_data.tcl",
		       "tbdata/feedback_data_failed.tcl");
		rename("tbdata/feedback_data_old.tcl",
		       "tbdata/feedback_data.tcl");
	}

	return retval;
}

static int do_modify(simulator_agent_t sa, int token, char *args)
{
	int rc, retval = 0;
	char *mode;

	assert(sa != NULL);
	assert(args != NULL);
	
	if ((rc = event_arg_get(args, "MODE", &mode)) <= 0) {
		error("no mode specified\n");
	}
	else if (strncasecmp("stabilize", mode, rc) == 0) {
168 169
		rename("tbdata/feedback_data.tcl",
		       "tbdata/feedback_data_old.tcl");
170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196
		if (systemf("loghole --port=%d --quiet sync",
			    DEFAULT_RPC_PORT) != 0) {
			error("failed to sync log holes\n");
		}
		else if (systemf("feedbacklogs %s %s", pid, eid) != 0) {
			if (sa->sa_flags & SAF_STABLE) {
				/* XXX log error */
				warning("unstabilized!\n");
			}
			else {
				retval = remap_experiment(sa, token);
			}
		}
		else {
			info("stabilized\n");
			sa->sa_flags |= SAF_STABLE;
		}
	}
	else {
		warning("unknown mode %s\n", mode);
	}

	return retval;
}

static void dump_report_data(FILE *file,
			     simulator_agent_t sa,
197 198
			     sa_report_data_kind_t srdk,
			     int clear)
199 200 201 202 203 204 205 206 207
{
	assert(file != NULL);
	assert(sa != NULL);
	assert(srdk >= 0);
	assert(srdk < SA_RDK_MAX);

	if ((sa->sa_report_data[srdk] != NULL) &&
	    (strlen(sa->sa_report_data[srdk]) > 0)) {
		fprintf(file, "\n%s\n", sa->sa_report_data[srdk]);
208 209 210 211
		if (clear) {
			free(sa->sa_report_data[srdk]);
			sa->sa_report_data[srdk] = NULL;
		}
212 213 214
	}
}

215
static int real_send_report(simulator_agent_t sa, char *args, bool error_report)
216 217
{
	struct lnList error_records;
218 219
	char loghole_name[BUFSIZ];
	int rc, retval;
220
	FILE *file;
221
	bool archive;
222
	char * tmp;
223
	const char * report_file_name;
224 225

	assert(sa != NULL);
226
	assert(args != NULL);
227 228 229 230 231 232 233 234 235 236 237 238
	
	/*
	 * Atomically move the error records from the agent object onto our
	 * local list and make the agent's list empty.
	 */
	if (pthread_mutex_lock(&sa->sa_local_agent.la_mutex) != 0)
		assert(0);
	lnMoveList(&error_records, &sa->sa_error_records);

	assert(lnEmptyList(&sa->sa_error_records));
	if (pthread_mutex_unlock(&sa->sa_local_agent.la_mutex) != 0)
		assert(0);
239 240 241 242 243 244
	
	if (debug) {
		char time_buf[24];
		struct timeval now;
		gettimeofday(&now, NULL);
		make_timestamp(time_buf, &now);
245 246
		info("Sending %sReport at %s with args \"%s\"\n",
		     error_report ? "Error " : "", time_buf, args);
247
	}
248 249 250 251 252

	/*
	 * Get the logs off the nodes so we can generate summaries from the
	 * error records.
	 */
253 254 255
	if ((rc = systemf("loghole --port=%d -vvv sync", DEFAULT_RPC_PORT))
	    != 0) {
		errorc("failed to sync log holes %d\n", rc);
256
	}
257

258 259 260 261 262 263
	report_file_name = error_report 
		? "logs/error-report.mail" 
		: "logs/report.mail";

	if ((file = fopen(report_file_name, "w")) == NULL) {
		errorc("could not create %s\n", report_file_name);
264 265 266
		retval = -1;
	}
	else {
267
		int rc, lpc, error_count;
268
		char *digester;
269
		FILE *vfile;
270 271 272
		
		retval = 0;

273 274 275 276 277 278 279 280 281 282
		error_count = lnCountNodes(&error_records);
		if (error_count > 0) {
			fprintf(file,
				"\n"
				"  *** %d error(s) were detected!\n"
				"      Details should be below in the logs.\n"
				"\n",
				error_count);
		}

283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305
		if ((vfile = popenf("loghole --port=%d validate",
				    "r",
				    DEFAULT_RPC_PORT)) == NULL) {
			fprintf(file, "[unable to validate logs]\n");
		}
		else {
			char buf[BUFSIZ];
			int total = 0;
			
			while ((rc = fread(buf,
					   1,
					   sizeof(buf),
					   vfile)) > 0) {
				fwrite(buf, 1, rc, file);
				total += rc;
			}
			pclose(vfile);
			vfile = NULL;

			if (total > 0)
				fprintf(file, "\n");
		}
		
306
		/* Dump user supplied stuff first, */
307
		dump_report_data(file, sa, SA_RDK_MESSAGE, 1);
308 309 310

		/* ... run the user-specified log digester, then */
		if ((rc = event_arg_get(args, "DIGESTER", &digester)) > 0) {
311 312
			FILE *dfile;
			
313 314 315 316 317
			digester[rc] = '\0';

			if ((dfile = popenf("%s | tee logs/digest.out",
					    "r",
					    digester)) == NULL) {
318
				fprintf(file,
319 320
					"[failed to run digester %s]\n",
					digester);
321
			}
322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337
			else {
				char buf[BUFSIZ];
				
				while ((rc = fread(buf,
						   1,
						   sizeof(buf),
						   dfile)) > 0) {
					fwrite(buf, 1, rc, file);
				}
				pclose(dfile);
				dfile = NULL;
			}

			fprintf(file, "\n");
		}
		
338
		fprintf(file, "Configuration:\n");
339
		dump_report_data(file, sa, SA_RDK_CONFIG, 0);
340

341
		fprintf(file, "\nLog:\n");
342
		dump_report_data(file, sa, SA_RDK_LOG, 1);
343
		
344 345 346 347 348 349
		/* ... dump the error records. */
		if (dump_error_records(&error_records, file) != 0) {
			errorc("dump_error_records failed");
			retval = -1;
		}
		
350
		fclose(file);
351 352
		file = NULL;
	}
353
	
354 355
	if (error_report) {
		archive = false;
356
	} else if ((rc = event_arg_get(args, "ARCHIVE", &tmp)) > 0) {
357 358 359 360 361
		if (rc == 4 && strncmp(tmp, "true", 4) == 0) {
			archive = true;
		} else if (rc == 5 && strncmp(tmp, "false", 5) == 0) {
			archive = false;
		} else {
362 363
			error("ARCHIVE must be \"true\" or \"false\", assuming \"false\"\n");
			archive = false;
364
		}
365 366 367 368
	} else {
		archive = true;
	}
	
369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385
	if (archive) {
		if ((file = popenf("loghole --port=%d --quiet archive --delete", 
				   "r", DEFAULT_RPC_PORT)) == NULL) {
			strcpy(loghole_name, eid);
			error("failed to archive log holes\n");
		}
		else {
			int len;
			
			fgets(loghole_name, sizeof(loghole_name), file);
			pclose(file);
			file = NULL;
			
			len = strlen(loghole_name);
			if (loghole_name[len - 1] == '\n')
				loghole_name[len - 1] = '\0';
		}
386 387
	}
	else {
388
		loghole_name[0] = '\0';
389
	}
390

391 392
	if (systemf("mail -s \"%s: %s/%s experiment %sreport\" %s "
		    "< %s",
393 394
		    OURDOMAIN,
		    pid,
395
		    loghole_name[0] ? loghole_name : eid,
396 397 398
		    error_report ? "error " : "",
		    getenv("USER"),
		    report_file_name) != 0) {
399 400 401 402
		errorc("could not execute send report\n");
		retval = -1;
	}
	
403 404 405 406 407 408 409
	delete_error_records(&error_records);

	assert(lnEmptyList(&error_records));
	
	return retval;
}

410 411 412 413 414 415 416 417
int send_report(simulator_agent_t sa, char *args) {
	return real_send_report(sa, args, false);
}

int send_error_report(simulator_agent_t sa, char *args) {
	return real_send_report(sa, args, true);
}

418 419 420 421 422 423 424
static int do_reset(simulator_agent_t sa, char *args)
{
	int retval = 0;

	assert(sa != NULL);
	assert(args != NULL);

425
	if (systemf("loghole --port=%d --quiet clean --force",
426 427 428 429 430 431 432
		    DEFAULT_RPC_PORT) != 0) {
		error("failed to clean log holes\n");
	}

	return retval;
}

433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452
static int do_snapshot(simulator_agent_t sa, char *args)
{
	char *loghole_args;
	int retval = 0;

	assert(sa != NULL);
	assert(args != NULL);
	
	if (event_arg_get(args, "LOGHOLE_ARGS", &loghole_args) <= 0) {
		loghole_args = "";
	}
	
	if (systemf("loghole --port=%d sync %s",
		    DEFAULT_RPC_PORT, loghole_args) != 0) {
		error("failed to sync log holes\n");
	}

	return retval;
}

453
static int do_stoprun(simulator_agent_t sa, int token, char *args)
454 455 456 457 458
{
	int retval = 0;

	assert(sa != NULL);
	assert(args != NULL);
459 460 461 462 463 464

	/*
	 * Not allowed to use waitmode; will deadlock the event system!
	 */
	if (systemf("template_stoprun -t %d -p %s -e %s",
		    token, pid, eid) != 0) {
465 466 467 468 469 470 471
		error("failed to stop current run\n");
		retval = -1;
	}

	return retval;
}

472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520
static int strreltime(char *buf, size_t buflen, time_t secs)
{
    int hours, mins, retval = 0;
    char *signage = "";
    
    assert(buf != NULL);
    
    if (secs < 0) {
	secs = abs(secs);
	signage = "-";
    }
    
    hours = secs / (60 * 60);
    mins = ((secs - (hours * 60 * 60)) / 60) % 60;
    secs = secs % 60;

    if (hours)
	snprintf(buf, buflen, "%s%dh%dm%ds", signage, hours, mins, secs);
    else if (mins)
	snprintf(buf, buflen, "%s%dm%ds", signage, mins, secs);
    else
	snprintf(buf, buflen, "%s%ds", signage, secs);
    return retval;
}

static int do_log(simulator_agent_t sa, char *message)
{
	time_t current_time;
	int retval = -1;
	char date[256];
	
	current_time = time(NULL);
	if (sa->sa_flags & SAF_TIME_STARTED) {
		strreltime(date,
			   sizeof(date),
			   current_time - sa->sa_time_start.tv_sec);
	}
	else {
		ctime_r(&current_time, date);
		date[strlen(date) - 1] = '\0';
	}
	strcat(date, " - ");
	fprintf(stderr, "log: %s%s\n", date, message);
	add_report_data(sa, SA_RDK_LOG, date, 0);
	add_report_data(sa, SA_RDK_LOG, message, SA_RDF_NEWLINE);

	return retval;
}

521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543
static void *simulator_agent_looper(void *arg)
{
	simulator_agent_t sa = (simulator_agent_t)arg;
	event_handle_t handle;
	void *retval = NULL;
	sched_event_t se;
	
	assert(arg != NULL);

	handle = sa->sa_local_agent.la_handle;

	while (local_agent_dequeue(&sa->sa_local_agent, 0, &se) == 0) {
		char evtype[TBDB_FLEN_EVEVENTTYPE];
		char argsbuf[BUFSIZ];

		assert(se.length == 1);
		
		if (!event_notification_get_eventtype(
			handle, se.notification, evtype, sizeof(evtype))) {
			error("couldn't get event type from notification %p\n",
			      se.notification);
		}
		else {
544 545
			int rc = 0;
			int32_t token = ~0;
546 547 548 549 550 551 552 553
			
			event_notification_get_arguments(handle,
							 se.notification,
							 argsbuf,
							 sizeof(argsbuf));
			event_notification_get_int32(handle,
						     se.notification,
						     "TOKEN",
554
						     &token);
555 556
			argsbuf[sizeof(argsbuf) - 1] = '\0';

557 558 559
			/* Strictly for the event viewer */
			event_notify(handle, se.notification);

560
			if (strcmp(evtype, TBDB_EVENTTYPE_SWAPOUT) == 0) {
561 562 563 564 565 566 567 568 569 570 571 572 573
				EmulabResponse er;
				
				RPC_grab();
				RPC_invoke("experiment.swapexp",
					   &er,
					   SPA_String, "proj", pid,
					   SPA_String, "exp", eid,
					   SPA_String, "direction", "out",
					   SPA_TAG_DONE);
				RPC_drop();
			}
			else if (strcmp(evtype, TBDB_EVENTTYPE_MODIFY) == 0) {
				do_modify(sa, token, argsbuf);
574 575 576 577 578 579 580 581 582
			}
			else if (strcmp(evtype, TBDB_EVENTTYPE_HALT) == 0) {
				rc = systemf("endexp %s %s", pid, eid);
			}
			else if (strcmp(evtype, TBDB_EVENTTYPE_DEBUG) == 0) {
				fprintf(stderr, "debug: %s\n", argsbuf);
			}
			else if (strcmp(evtype, TBDB_EVENTTYPE_MESSAGE) == 0) {
				fprintf(stderr, "msg: %s\n", argsbuf);
583 584 585 586
				add_report_data(sa,
						SA_RDK_MESSAGE,
						argsbuf,
						SA_RDF_NEWLINE);
587 588
			}
			else if (strcmp(evtype, TBDB_EVENTTYPE_LOG) == 0) {
589
				do_log(sa, argsbuf);
590 591
			}
			else if (strcmp(evtype, TBDB_EVENTTYPE_REPORT) == 0) {
592
				do_log(sa, "Sending report\n");
593
				send_report(sa, argsbuf);
594
			}
595 596 597
			else if (strcmp(evtype, TBDB_EVENTTYPE_RESET) == 0) {
				do_reset(sa, argsbuf);
			}
598 599 600
			else if (strcmp(evtype, TBDB_EVENTTYPE_SNAPSHOT) == 0){
				do_snapshot(sa, argsbuf);
			}
601
			else if (strcmp(evtype, TBDB_EVENTTYPE_STOPRUN) == 0){
602
				do_stoprun(sa, token, argsbuf);
603
			}
604 605 606 607
			else {
				error("cannot handle SIMULATOR event %s.",
				      evtype);
			}
608 609
			if (strcmp(evtype, TBDB_EVENTTYPE_RESET) == 0 ||
			    strcmp(evtype, TBDB_EVENTTYPE_REPORT) == 0 ||
610
			    strcmp(evtype, TBDB_EVENTTYPE_LOG) == 0 ||
611
			    strcmp(evtype, TBDB_EVENTTYPE_SNAPSHOT) == 0 ||
612 613 614 615 616 617 618 619 620 621 622
			    strcmp(evtype, TBDB_EVENTTYPE_MODIFY) == 0) {
				event_do(handle,
					 EA_Experiment, pideid,
					 EA_Type, TBDB_OBJECTTYPE_SIMULATOR,
					 EA_Name,
					 sa->sa_local_agent.la_link.ln_Name,
					 EA_Event, TBDB_EVENTTYPE_COMPLETE,
					 EA_ArgInteger, "ERROR", rc != 0,
					 EA_ArgInteger, "CTOKEN", token,
					 EA_TAG_DONE);
			}
623 624 625 626 627 628 629
		}

		sched_event_free(handle, &se);
	}

	return retval;
}