• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Status and ETA code
3  */
4 #include <unistd.h>
5 #include <fcntl.h>
6 #include <string.h>
7 
8 #include "fio.h"
9 
10 static char run_str[REAL_MAX_JOBS + 1];
11 
12 /*
13  * Sets the status of the 'td' in the printed status map.
14  */
check_str_update(struct thread_data * td)15 static void check_str_update(struct thread_data *td)
16 {
17 	char c = run_str[td->thread_number - 1];
18 
19 	switch (td->runstate) {
20 	case TD_REAPED:
21 		if (td->error)
22 			c = 'X';
23 		else if (td->sig)
24 			c = 'K';
25 		else
26 			c = '_';
27 		break;
28 	case TD_EXITED:
29 		c = 'E';
30 		break;
31 	case TD_RAMP:
32 		c = '/';
33 		break;
34 	case TD_RUNNING:
35 		if (td_rw(td)) {
36 			if (td_random(td)) {
37 				if (td->o.rwmix[DDIR_READ] == 100)
38 					c = 'r';
39 				else if (td->o.rwmix[DDIR_WRITE] == 100)
40 					c = 'w';
41 				else
42 					c = 'm';
43 			} else {
44 				if (td->o.rwmix[DDIR_READ] == 100)
45 					c = 'R';
46 				else if (td->o.rwmix[DDIR_WRITE] == 100)
47 					c = 'W';
48 				else
49 					c = 'M';
50 			}
51 		} else if (td_read(td)) {
52 			if (td_random(td))
53 				c = 'r';
54 			else
55 				c = 'R';
56 		} else if (td_write(td)) {
57 			if (td_random(td))
58 				c = 'w';
59 			else
60 				c = 'W';
61 		} else {
62 			if (td_random(td))
63 				c = 'd';
64 			else
65 				c = 'D';
66 		}
67 		break;
68 	case TD_PRE_READING:
69 		c = 'p';
70 		break;
71 	case TD_VERIFYING:
72 		c = 'V';
73 		break;
74 	case TD_FSYNCING:
75 		c = 'F';
76 		break;
77 	case TD_FINISHING:
78 		c = 'f';
79 		break;
80 	case TD_CREATED:
81 		c = 'C';
82 		break;
83 	case TD_INITIALIZED:
84 	case TD_SETTING_UP:
85 		c = 'I';
86 		break;
87 	case TD_NOT_CREATED:
88 		c = 'P';
89 		break;
90 	default:
91 		log_err("state %d\n", td->runstate);
92 	}
93 
94 	run_str[td->thread_number - 1] = c;
95 }
96 
97 /*
98  * Convert seconds to a printable string.
99  */
eta_to_str(char * str,unsigned long eta_sec)100 void eta_to_str(char *str, unsigned long eta_sec)
101 {
102 	unsigned int d, h, m, s;
103 	int disp_hour = 0;
104 
105 	s = eta_sec % 60;
106 	eta_sec /= 60;
107 	m = eta_sec % 60;
108 	eta_sec /= 60;
109 	h = eta_sec % 24;
110 	eta_sec /= 24;
111 	d = eta_sec;
112 
113 	if (d) {
114 		disp_hour = 1;
115 		str += sprintf(str, "%02ud:", d);
116 	}
117 
118 	if (h || disp_hour)
119 		str += sprintf(str, "%02uh:", h);
120 
121 	str += sprintf(str, "%02um:", m);
122 	str += sprintf(str, "%02us", s);
123 }
124 
125 /*
126  * Best effort calculation of the estimated pending runtime of a job.
127  */
thread_eta(struct thread_data * td)128 static int thread_eta(struct thread_data *td)
129 {
130 	unsigned long long bytes_total, bytes_done;
131 	unsigned long eta_sec = 0;
132 	unsigned long elapsed;
133 	uint64_t timeout;
134 
135 	elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
136 	timeout = td->o.timeout / 1000000UL;
137 
138 	bytes_total = td->total_io_size;
139 
140 	if (td->o.fill_device && td->o.size  == -1ULL) {
141 		if (!td->fill_device_size || td->fill_device_size == -1ULL)
142 			return 0;
143 
144 		bytes_total = td->fill_device_size;
145 	}
146 
147 	if (td->o.zone_size && td->o.zone_skip && bytes_total) {
148 		unsigned int nr_zones;
149 		uint64_t zone_bytes;
150 
151 		zone_bytes = bytes_total + td->o.zone_size + td->o.zone_skip;
152 		nr_zones = (zone_bytes - 1) / (td->o.zone_size + td->o.zone_skip);
153 		bytes_total -= nr_zones * td->o.zone_skip;
154 	}
155 
156 	/*
157 	 * if writing and verifying afterwards, bytes_total will be twice the
158 	 * size. In a mixed workload, verify phase will be the size of the
159 	 * first stage writes.
160 	 */
161 	if (td->o.do_verify && td->o.verify && td_write(td)) {
162 		if (td_rw(td)) {
163 			unsigned int perc = 50;
164 
165 			if (td->o.rwmix[DDIR_WRITE])
166 				perc = td->o.rwmix[DDIR_WRITE];
167 
168 			bytes_total += (bytes_total * perc) / 100;
169 		} else
170 			bytes_total <<= 1;
171 	}
172 
173 	if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
174 		double perc, perc_t;
175 
176 		bytes_done = ddir_rw_sum(td->io_bytes);
177 
178 		if (bytes_total) {
179 			perc = (double) bytes_done / (double) bytes_total;
180 			if (perc > 1.0)
181 				perc = 1.0;
182 		} else
183 			perc = 0.0;
184 
185 		if (td->o.time_based) {
186 			if (timeout) {
187 				perc_t = (double) elapsed / (double) timeout;
188 				if (perc_t < perc)
189 					perc = perc_t;
190 			} else {
191 				/*
192 				 * Will never hit, we can't have time_based
193 				 * without a timeout set.
194 				 */
195 				perc = 0.0;
196 			}
197 		}
198 
199 		eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
200 
201 		if (td->o.timeout &&
202 		    eta_sec > (timeout + done_secs - elapsed))
203 			eta_sec = timeout + done_secs - elapsed;
204 	} else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
205 			|| td->runstate == TD_INITIALIZED
206 			|| td->runstate == TD_SETTING_UP
207 			|| td->runstate == TD_RAMP
208 			|| td->runstate == TD_PRE_READING) {
209 		int t_eta = 0, r_eta = 0;
210 		unsigned long long rate_bytes;
211 
212 		/*
213 		 * We can only guess - assume it'll run the full timeout
214 		 * if given, otherwise assume it'll run at the specified rate.
215 		 */
216 		if (td->o.timeout) {
217 			uint64_t timeout = td->o.timeout;
218 			uint64_t start_delay = td->o.start_delay;
219 			uint64_t ramp_time = td->o.ramp_time;
220 
221 			t_eta = timeout + start_delay + ramp_time;
222 			t_eta /= 1000000ULL;
223 
224 			if (in_ramp_time(td)) {
225 				unsigned long ramp_left;
226 
227 				ramp_left = mtime_since_now(&td->epoch);
228 				ramp_left = (ramp_left + 999) / 1000;
229 				if (ramp_left <= t_eta)
230 					t_eta -= ramp_left;
231 			}
232 		}
233 		rate_bytes = ddir_rw_sum(td->o.rate);
234 		if (rate_bytes) {
235 			r_eta = (bytes_total / 1024) / rate_bytes;
236 			r_eta += (td->o.start_delay / 1000000ULL);
237 		}
238 
239 		if (r_eta && t_eta)
240 			eta_sec = min(r_eta, t_eta);
241 		else if (r_eta)
242 			eta_sec = r_eta;
243 		else if (t_eta)
244 			eta_sec = t_eta;
245 		else
246 			eta_sec = 0;
247 	} else {
248 		/*
249 		 * thread is already done or waiting for fsync
250 		 */
251 		eta_sec = 0;
252 	}
253 
254 	return eta_sec;
255 }
256 
calc_rate(int unified_rw_rep,unsigned long mtime,unsigned long long * io_bytes,unsigned long long * prev_io_bytes,unsigned int * rate)257 static void calc_rate(int unified_rw_rep, unsigned long mtime,
258 		      unsigned long long *io_bytes,
259 		      unsigned long long *prev_io_bytes, unsigned int *rate)
260 {
261 	int i;
262 
263 	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
264 		unsigned long long diff;
265 
266 		diff = io_bytes[i] - prev_io_bytes[i];
267 		if (unified_rw_rep) {
268 			rate[i] = 0;
269 			rate[0] += ((1000 * diff) / mtime) / 1024;
270 		} else
271 			rate[i] = ((1000 * diff) / mtime) / 1024;
272 
273 		prev_io_bytes[i] = io_bytes[i];
274 	}
275 }
276 
calc_iops(int unified_rw_rep,unsigned long mtime,unsigned long long * io_iops,unsigned long long * prev_io_iops,unsigned int * iops)277 static void calc_iops(int unified_rw_rep, unsigned long mtime,
278 		      unsigned long long *io_iops,
279 		      unsigned long long *prev_io_iops, unsigned int *iops)
280 {
281 	int i;
282 
283 	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
284 		unsigned long long diff;
285 
286 		diff = io_iops[i] - prev_io_iops[i];
287 		if (unified_rw_rep) {
288 			iops[i] = 0;
289 			iops[0] += (diff * 1000) / mtime;
290 		} else
291 			iops[i] = (diff * 1000) / mtime;
292 
293 		prev_io_iops[i] = io_iops[i];
294 	}
295 }
296 
297 /*
298  * Print status of the jobs we know about. This includes rate estimates,
299  * ETA, thread state, etc.
300  */
calc_thread_status(struct jobs_eta * je,int force)301 int calc_thread_status(struct jobs_eta *je, int force)
302 {
303 	struct thread_data *td;
304 	int i, unified_rw_rep;
305 	unsigned long rate_time, disp_time, bw_avg_time, *eta_secs;
306 	unsigned long long io_bytes[DDIR_RWDIR_CNT];
307 	unsigned long long io_iops[DDIR_RWDIR_CNT];
308 	struct timeval now;
309 
310 	static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
311 	static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
312 	static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
313 	static struct timeval rate_prev_time, disp_prev_time;
314 
315 	if (!force) {
316 		if (output_format != FIO_OUTPUT_NORMAL &&
317 		    f_out == stdout)
318 			return 0;
319 		if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
320 			return 0;
321 
322 		if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
323 			return 0;
324 	}
325 
326 	if (!ddir_rw_sum(rate_io_bytes))
327 		fill_start_time(&rate_prev_time);
328 	if (!ddir_rw_sum(disp_io_bytes))
329 		fill_start_time(&disp_prev_time);
330 
331 	eta_secs = malloc(thread_number * sizeof(unsigned long));
332 	memset(eta_secs, 0, thread_number * sizeof(unsigned long));
333 
334 	je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
335 
336 	io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
337 	io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
338 	bw_avg_time = ULONG_MAX;
339 	unified_rw_rep = 0;
340 	for_each_td(td, i) {
341 		unified_rw_rep += td->o.unified_rw_rep;
342 		if (is_power_of_2(td->o.kb_base))
343 			je->is_pow2 = 1;
344 		je->unit_base = td->o.unit_base;
345 		if (td->o.bw_avg_time < bw_avg_time)
346 			bw_avg_time = td->o.bw_avg_time;
347 		if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
348 		    || td->runstate == TD_FSYNCING
349 		    || td->runstate == TD_PRE_READING
350 		    || td->runstate == TD_FINISHING) {
351 			je->nr_running++;
352 			if (td_read(td)) {
353 				je->t_rate[0] += td->o.rate[DDIR_READ];
354 				je->t_iops[0] += td->o.rate_iops[DDIR_READ];
355 				je->m_rate[0] += td->o.ratemin[DDIR_READ];
356 				je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
357 			}
358 			if (td_write(td)) {
359 				je->t_rate[1] += td->o.rate[DDIR_WRITE];
360 				je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
361 				je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
362 				je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
363 			}
364 			if (td_trim(td)) {
365 				je->t_rate[2] += td->o.rate[DDIR_TRIM];
366 				je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
367 				je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
368 				je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
369 			}
370 
371 			je->files_open += td->nr_open_files;
372 		} else if (td->runstate == TD_RAMP) {
373 			je->nr_running++;
374 			je->nr_ramp++;
375 		} else if (td->runstate == TD_SETTING_UP) {
376 			je->nr_running++;
377 			je->nr_setting_up++;
378 		} else if (td->runstate < TD_RUNNING)
379 			je->nr_pending++;
380 
381 		if (je->elapsed_sec >= 3)
382 			eta_secs[i] = thread_eta(td);
383 		else
384 			eta_secs[i] = INT_MAX;
385 
386 		check_str_update(td);
387 
388 		if (td->runstate > TD_SETTING_UP) {
389 			int ddir;
390 
391 			for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
392 				if (unified_rw_rep) {
393 					io_bytes[0] += td->io_bytes[ddir];
394 					io_iops[0] += td->io_blocks[ddir];
395 				} else {
396 					io_bytes[ddir] += td->io_bytes[ddir];
397 					io_iops[ddir] += td->io_blocks[ddir];
398 				}
399 			}
400 		}
401 	}
402 
403 	if (exitall_on_terminate)
404 		je->eta_sec = INT_MAX;
405 	else
406 		je->eta_sec = 0;
407 
408 	for_each_td(td, i) {
409 		if (exitall_on_terminate) {
410 			if (eta_secs[i] < je->eta_sec)
411 				je->eta_sec = eta_secs[i];
412 		} else {
413 			if (eta_secs[i] > je->eta_sec)
414 				je->eta_sec = eta_secs[i];
415 		}
416 	}
417 
418 	free(eta_secs);
419 
420 	fio_gettime(&now, NULL);
421 	rate_time = mtime_since(&rate_prev_time, &now);
422 
423 	if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
424 		calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
425 				je->rate);
426 		memcpy(&rate_prev_time, &now, sizeof(now));
427 		add_agg_sample(je->rate[DDIR_READ], DDIR_READ, 0);
428 		add_agg_sample(je->rate[DDIR_WRITE], DDIR_WRITE, 0);
429 		add_agg_sample(je->rate[DDIR_TRIM], DDIR_TRIM, 0);
430 	}
431 
432 	disp_time = mtime_since(&disp_prev_time, &now);
433 
434 	/*
435 	 * Allow a little slack, the target is to print it every 1000 msecs
436 	 */
437 	if (!force && disp_time < 900)
438 		return 0;
439 
440 	calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
441 	calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);
442 
443 	memcpy(&disp_prev_time, &now, sizeof(now));
444 
445 	if (!force && !je->nr_running && !je->nr_pending)
446 		return 0;
447 
448 	je->nr_threads = thread_number;
449 	memcpy(je->run_str, run_str, thread_number * sizeof(char));
450 	return 1;
451 }
452 
display_thread_status(struct jobs_eta * je)453 void display_thread_status(struct jobs_eta *je)
454 {
455 	static struct timeval disp_eta_new_line;
456 	static int eta_new_line_init, eta_new_line_pending;
457 	static int linelen_last;
458 	static int eta_good;
459 	char output[REAL_MAX_JOBS + 512], *p = output;
460 	char eta_str[128];
461 	double perc = 0.0;
462 
463 	if (je->eta_sec != INT_MAX && je->elapsed_sec) {
464 		perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
465 		eta_to_str(eta_str, je->eta_sec);
466 	}
467 
468 	if (eta_new_line_pending) {
469 		eta_new_line_pending = 0;
470 		p += sprintf(p, "\n");
471 	}
472 
473 	p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
474 	if (je->m_rate[0] || je->m_rate[1] || je->t_rate[0] || je->t_rate[1]) {
475 		char *tr, *mr;
476 
477 		mr = num2str(je->m_rate[0] + je->m_rate[1], 4, 0, je->is_pow2, 8);
478 		tr = num2str(je->t_rate[0] + je->t_rate[1], 4, 0, je->is_pow2, 8);
479 		p += sprintf(p, ", CR=%s/%s KB/s", tr, mr);
480 		free(tr);
481 		free(mr);
482 	} else if (je->m_iops[0] || je->m_iops[1] || je->t_iops[0] || je->t_iops[1]) {
483 		p += sprintf(p, ", CR=%d/%d IOPS",
484 					je->t_iops[0] + je->t_iops[1],
485 					je->m_iops[0] + je->m_iops[1]);
486 	}
487 	if (je->eta_sec != INT_MAX && je->nr_running) {
488 		char perc_str[32];
489 		char *iops_str[DDIR_RWDIR_CNT];
490 		char *rate_str[DDIR_RWDIR_CNT];
491 		size_t left;
492 		int l;
493 		int ddir;
494 
495 		if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running)
496 			strcpy(perc_str, "-.-% done");
497 		else {
498 			double mult = 100.0;
499 
500 			if (je->nr_setting_up && je->nr_running)
501 				mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);
502 
503 			eta_good = 1;
504 			perc *= mult;
505 			sprintf(perc_str, "%3.1f%% done", perc);
506 		}
507 
508 		for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
509 			rate_str[ddir] = num2str(je->rate[ddir], 5,
510 						1024, je->is_pow2, je->unit_base);
511 			iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, 0);
512 		}
513 
514 		left = sizeof(output) - (p - output) - 1;
515 
516 		l = snprintf(p, left, ": [%s] [%s] [%s/%s/%s /s] [%s/%s/%s iops] [eta %s]",
517 				je->run_str, perc_str, rate_str[DDIR_READ],
518 				rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
519 				iops_str[DDIR_READ], iops_str[DDIR_WRITE],
520 				iops_str[DDIR_TRIM], eta_str);
521 		p += l;
522 		if (l >= 0 && l < linelen_last)
523 			p += sprintf(p, "%*s", linelen_last - l, "");
524 		linelen_last = l;
525 
526 		for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
527 			free(rate_str[ddir]);
528 			free(iops_str[ddir]);
529 		}
530 	}
531 	p += sprintf(p, "\r");
532 
533 	printf("%s", output);
534 
535 	if (!eta_new_line_init) {
536 		fio_gettime(&disp_eta_new_line, NULL);
537 		eta_new_line_init = 1;
538 	} else if (eta_new_line &&
539 		   mtime_since_now(&disp_eta_new_line) > eta_new_line * 1000) {
540 		fio_gettime(&disp_eta_new_line, NULL);
541 		eta_new_line_pending = 1;
542 	}
543 
544 	fflush(stdout);
545 }
546 
print_thread_status(void)547 void print_thread_status(void)
548 {
549 	struct jobs_eta *je;
550 	size_t size;
551 
552 	if (!thread_number)
553 		return;
554 
555 	size = sizeof(*je) + thread_number * sizeof(char) + 1;
556 	je = malloc(size);
557 	memset(je, 0, size);
558 
559 	if (calc_thread_status(je, 0))
560 		display_thread_status(je);
561 
562 	free(je);
563 }
564 
print_status_init(int thr_number)565 void print_status_init(int thr_number)
566 {
567 	run_str[thr_number] = 'P';
568 }
569