[fio.git] / eta.c

/*
 * Status and ETA code
 */
#include <unistd.h>
#include <fcntl.h>
#include <string.h>

#include "fio.h"
#include "lib/pow2.h"

static char __run_str[REAL_MAX_JOBS + 1];
static char run_str[__THREAD_RUNSTR_SZ(REAL_MAX_JOBS)];

static void update_condensed_str(char *rstr, char *run_str_condensed)
{
	if (*rstr) {
		while (*rstr) {
			int nr = 1;

			*run_str_condensed++ = *rstr++;
			while (*(rstr - 1) == *rstr) {
				rstr++;
				nr++;
			}
			run_str_condensed += sprintf(run_str_condensed, "(%u),", nr);
		}
		run_str_condensed--;
	}
	*run_str_condensed = '\0';
}

/*
 * Sets the status of the 'td' in the printed status map.
 */
static void check_str_update(struct thread_data *td)
{
	char c = __run_str[td->thread_number - 1];

	switch (td->runstate) {
	case TD_REAPED:
		if (td->error)
			c = 'X';
		else if (td->sig)
			c = 'K';
		else
			c = '_';
		break;
	case TD_EXITED:
		c = 'E';
		break;
	case TD_RAMP:
		c = '/';
		break;
	case TD_RUNNING:
		if (td_rw(td)) {
			if (td_random(td)) {
				if (td->o.rwmix[DDIR_READ] == 100)
					c = 'r';
				else if (td->o.rwmix[DDIR_WRITE] == 100)
					c = 'w';
				else
					c = 'm';
			} else {
				if (td->o.rwmix[DDIR_READ] == 100)
					c = 'R';
				else if (td->o.rwmix[DDIR_WRITE] == 100)
					c = 'W';
				else
					c = 'M';
			}
		} else if (td_read(td)) {
			if (td_random(td))
				c = 'r';
			else
				c = 'R';
		} else if (td_write(td)) {
			if (td_random(td))
				c = 'w';
			else
				c = 'W';
		} else {
			if (td_random(td))
				c = 'd';
			else
				c = 'D';
		}
		break;
	case TD_PRE_READING:
		c = 'p';
		break;
	case TD_VERIFYING:
		c = 'V';
		break;
	case TD_FSYNCING:
		c = 'F';
		break;
	case TD_FINISHING:
		c = 'f';
		break;
	case TD_CREATED:
		c = 'C';
		break;
	case TD_INITIALIZED:
	case TD_SETTING_UP:
		c = 'I';
		break;
	case TD_NOT_CREATED:
		c = 'P';
		break;
	default:
		log_err("state %d\n", td->runstate);
	}

	__run_str[td->thread_number - 1] = c;
	update_condensed_str(__run_str, run_str);
}

/*
 * Convert seconds to a printable string.
 */
void eta_to_str(char *str, unsigned long eta_sec)
{
	unsigned int d, h, m, s;
	int disp_hour = 0;

	if (eta_sec == -1) {
		sprintf(str, "--");
		return;
	}

	s = eta_sec % 60;
	eta_sec /= 60;
	m = eta_sec % 60;
	eta_sec /= 60;
	h = eta_sec % 24;
	eta_sec /= 24;
	d = eta_sec;

	if (d) {
		disp_hour = 1;
		str += sprintf(str, "%02ud:", d);
	}

	if (h || disp_hour)
		str += sprintf(str, "%02uh:", h);

	str += sprintf(str, "%02um:", m);
	str += sprintf(str, "%02us", s);
}

/*
 * Best effort calculation of the estimated pending runtime of a job.
 */
static unsigned long thread_eta(struct thread_data *td)
{
	unsigned long long bytes_total, bytes_done;
	unsigned long eta_sec = 0;
	unsigned long elapsed;
	uint64_t timeout;

	elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
	timeout = td->o.timeout / 1000000UL;

	bytes_total = td->total_io_size;

	if (td->flags & TD_F_NO_PROGRESS)
		return -1;

	if (td->o.fill_device && td->o.size  == -1ULL) {
		if (!td->fill_device_size || td->fill_device_size == -1ULL)
			return 0;

		bytes_total = td->fill_device_size;
	}

	if (td->o.zone_size && td->o.zone_skip && bytes_total) {
		unsigned int nr_zones;
		uint64_t zone_bytes;

		zone_bytes = bytes_total + td->o.zone_size + td->o.zone_skip;
		nr_zones = (zone_bytes - 1) / (td->o.zone_size + td->o.zone_skip);
		bytes_total -= nr_zones * td->o.zone_skip;
	}

	/*
	 * if writing and verifying afterwards, bytes_total will be twice the
	 * size. In a mixed workload, verify phase will be the size of the
	 * first stage writes.
	 */
	if (td->o.do_verify && td->o.verify && td_write(td)) {
		if (td_rw(td)) {
			unsigned int perc = 50;

			if (td->o.rwmix[DDIR_WRITE])
				perc = td->o.rwmix[DDIR_WRITE];

			bytes_total += (bytes_total * perc) / 100;
		} else
			bytes_total <<= 1;
	}

	if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
		double perc, perc_t;

		bytes_done = ddir_rw_sum(td->io_bytes);

		if (bytes_total) {
			perc = (double) bytes_done / (double) bytes_total;
			if (perc > 1.0)
				perc = 1.0;
		} else
			perc = 0.0;

		if (td->o.time_based) {
			if (timeout) {
				perc_t = (double) elapsed / (double) timeout;
				if (perc_t < perc)
					perc = perc_t;
			} else {
				/*
				 * Will never hit, we can't have time_based
				 * without a timeout set.
				 */
				perc = 0.0;
			}
		}

		eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;

		if (td->o.timeout &&
		    eta_sec > (timeout + done_secs - elapsed))
			eta_sec = timeout + done_secs - elapsed;
	} else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
			|| td->runstate == TD_INITIALIZED
			|| td->runstate == TD_SETTING_UP
			|| td->runstate == TD_RAMP
			|| td->runstate == TD_PRE_READING) {
		int t_eta = 0, r_eta = 0;
		unsigned long long rate_bytes;

		/*
		 * We can only guess - assume it'll run the full timeout
		 * if given, otherwise assume it'll run at the specified rate.
		 */
		if (td->o.timeout) {
			uint64_t __timeout = td->o.timeout;
			uint64_t start_delay = td->o.start_delay;
			uint64_t ramp_time = td->o.ramp_time;

			t_eta = __timeout + start_delay + ramp_time;
			t_eta /= 1000000ULL;

			if (in_ramp_time(td)) {
				unsigned long ramp_left;

				ramp_left = mtime_since_now(&td->epoch);
				ramp_left = (ramp_left + 999) / 1000;
				if (ramp_left <= t_eta)
					t_eta -= ramp_left;
			}
		}
		rate_bytes = ddir_rw_sum(td->o.rate);
		if (rate_bytes) {
			r_eta = (bytes_total / 1024) / rate_bytes;
			r_eta += (td->o.start_delay / 1000000ULL);
		}

		if (r_eta && t_eta)
			eta_sec = min(r_eta, t_eta);
		else if (r_eta)
			eta_sec = r_eta;
		else if (t_eta)
			eta_sec = t_eta;
		else
			eta_sec = 0;
	} else {
		/*
		 * thread is already done or waiting for fsync
		 */
		eta_sec = 0;
	}

	return eta_sec;
}

static void calc_rate(int unified_rw_rep, unsigned long mtime,
		      unsigned long long *io_bytes,
		      unsigned long long *prev_io_bytes, unsigned int *rate)
{
	int i;

	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
		unsigned long long diff, this_rate;

		diff = io_bytes[i] - prev_io_bytes[i];
		if (mtime)
			this_rate = ((1000 * diff) / mtime) / 1024;
		else
			this_rate = 0;

		if (unified_rw_rep) {
			rate[i] = 0;
			rate[0] += this_rate;
		} else
			rate[i] = this_rate;

		prev_io_bytes[i] = io_bytes[i];
	}
}

static void calc_iops(int unified_rw_rep, unsigned long mtime,
		      unsigned long long *io_iops,
		      unsigned long long *prev_io_iops, unsigned int *iops)
{
	int i;

	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
		unsigned long long diff;

		diff = io_iops[i] - prev_io_iops[i];
		if (unified_rw_rep) {
			iops[i] = 0;
			iops[0] += (diff * 1000) / mtime;
		} else
			iops[i] = (diff * 1000) / mtime;

		prev_io_iops[i] = io_iops[i];
	}
}

/*
 * Print status of the jobs we know about. This includes rate estimates,
 * ETA, thread state, etc.
 */
int calc_thread_status(struct jobs_eta *je, int force)
{
	struct thread_data *td;
	int i, unified_rw_rep;
	unsigned long rate_time, disp_time, bw_avg_time, *eta_secs;
	unsigned long long io_bytes[DDIR_RWDIR_CNT];
	unsigned long long io_iops[DDIR_RWDIR_CNT];
	struct timeval now;

	static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
	static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
	static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
	static struct timeval rate_prev_time, disp_prev_time;

	if (!force) {
		if (output_format != FIO_OUTPUT_NORMAL &&
		    f_out == stdout)
			return 0;
		if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
			return 0;

		if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
			return 0;
	}

	if (!ddir_rw_sum(rate_io_bytes))
		fill_start_time(&rate_prev_time);
	if (!ddir_rw_sum(disp_io_bytes))
		fill_start_time(&disp_prev_time);

	eta_secs = malloc(thread_number * sizeof(unsigned long));
	memset(eta_secs, 0, thread_number * sizeof(unsigned long));

	je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;

	io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
	io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
	bw_avg_time = ULONG_MAX;
	unified_rw_rep = 0;
	for_each_td(td, i) {
		unified_rw_rep += td->o.unified_rw_rep;
		if (is_power_of_2(td->o.kb_base))
			je->is_pow2 = 1;
		je->unit_base = td->o.unit_base;
		if (td->o.bw_avg_time < bw_avg_time)
			bw_avg_time = td->o.bw_avg_time;
		if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
		    || td->runstate == TD_FSYNCING
		    || td->runstate == TD_PRE_READING
		    || td->runstate == TD_FINISHING) {
			je->nr_running++;
			if (td_read(td)) {
				je->t_rate[0] += td->o.rate[DDIR_READ];
				je->t_iops[0] += td->o.rate_iops[DDIR_READ];
				je->m_rate[0] += td->o.ratemin[DDIR_READ];
				je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
			}
			if (td_write(td)) {
				je->t_rate[1] += td->o.rate[DDIR_WRITE];
				je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
				je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
				je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
			}
			if (td_trim(td)) {
				je->t_rate[2] += td->o.rate[DDIR_TRIM];
				je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
				je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
				je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
			}

			je->files_open += td->nr_open_files;
		} else if (td->runstate == TD_RAMP) {
			je->nr_running++;
			je->nr_ramp++;
		} else if (td->runstate == TD_SETTING_UP)
			je->nr_setting_up++;
		else if (td->runstate < TD_RUNNING)
			je->nr_pending++;

		if (je->elapsed_sec >= 3)
			eta_secs[i] = thread_eta(td);
		else
			eta_secs[i] = INT_MAX;

		check_str_update(td);

		if (td->runstate > TD_SETTING_UP) {
			int ddir;

			for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
				if (unified_rw_rep) {
					io_bytes[0] += td->io_bytes[ddir];
					io_iops[0] += td->io_blocks[ddir];
				} else {
					io_bytes[ddir] += td->io_bytes[ddir];
					io_iops[ddir] += td->io_blocks[ddir];
				}
			}
		}
	}

	if (exitall_on_terminate)
		je->eta_sec = INT_MAX;
	else
		je->eta_sec = 0;

	for_each_td(td, i) {
		if (exitall_on_terminate) {
			if (eta_secs[i] < je->eta_sec)
				je->eta_sec = eta_secs[i];
		} else {
			if (eta_secs[i] > je->eta_sec)
				je->eta_sec = eta_secs[i];
		}
	}

	free(eta_secs);

	fio_gettime(&now, NULL);
	rate_time = mtime_since(&rate_prev_time, &now);

	if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
		calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
				je->rate);
		memcpy(&rate_prev_time, &now, sizeof(now));
		add_agg_sample(je->rate[DDIR_READ], DDIR_READ, 0);
		add_agg_sample(je->rate[DDIR_WRITE], DDIR_WRITE, 0);
		add_agg_sample(je->rate[DDIR_TRIM], DDIR_TRIM, 0);
	}

	disp_time = mtime_since(&disp_prev_time, &now);

	/*
	 * Allow a little slack, the target is to print it every 1000 msecs
	 */
	if (!force && disp_time < 900)
		return 0;

	calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
	calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);

	memcpy(&disp_prev_time, &now, sizeof(now));

	if (!force && !je->nr_running && !je->nr_pending)
		return 0;

	je->nr_threads = thread_number;
	update_condensed_str(__run_str, run_str);
	memcpy(je->run_str, run_str, strlen(run_str));
	return 1;
}

void display_thread_status(struct jobs_eta *je)
{
	static struct timeval disp_eta_new_line;
	static int eta_new_line_init, eta_new_line_pending;
	static int linelen_last;
	static int eta_good;
	char output[REAL_MAX_JOBS + 512], *p = output;
	char eta_str[128];
	double perc = 0.0;

	if (je->eta_sec != INT_MAX && je->elapsed_sec) {
		perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
		eta_to_str(eta_str, je->eta_sec);
	}

	if (eta_new_line_pending) {
		eta_new_line_pending = 0;
		p += sprintf(p, "\n");
	}

	p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
	if (je->m_rate[0] || je->m_rate[1] || je->t_rate[0] || je->t_rate[1]) {
		char *tr, *mr;

		mr = num2str(je->m_rate[0] + je->m_rate[1], 4, 0, je->is_pow2, 8);
		tr = num2str(je->t_rate[0] + je->t_rate[1], 4, 0, je->is_pow2, 8);
		p += sprintf(p, ", CR=%s/%s KB/s", tr, mr);
		free(tr);
		free(mr);
	} else if (je->m_iops[0] || je->m_iops[1] || je->t_iops[0] || je->t_iops[1]) {
		p += sprintf(p, ", CR=%d/%d IOPS",
					je->t_iops[0] + je->t_iops[1],
					je->m_iops[0] + je->m_iops[1]);
	}
	if (je->eta_sec != INT_MAX && je->nr_running) {
		char perc_str[32];
		char *iops_str[DDIR_RWDIR_CNT];
		char *rate_str[DDIR_RWDIR_CNT];
		size_t left;
		int l;
		int ddir;

		if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running ||
		    je->eta_sec == -1)
			strcpy(perc_str, "-.-% done");
		else {
			double mult = 100.0;

			if (je->nr_setting_up && je->nr_running)
				mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);

			eta_good = 1;
			perc *= mult;
			sprintf(perc_str, "%3.1f%% done", perc);
		}

		for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
			rate_str[ddir] = num2str(je->rate[ddir], 5,
						1024, je->is_pow2, je->unit_base);
			iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, 0);
		}

		left = sizeof(output) - (p - output) - 1;

		l = snprintf(p, left, ": [%s] [%s] [%s/%s/%s /s] [%s/%s/%s iops] [eta %s]",
				je->run_str, perc_str, rate_str[DDIR_READ],
				rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
				iops_str[DDIR_READ], iops_str[DDIR_WRITE],
				iops_str[DDIR_TRIM], eta_str);
		p += l;
		if (l >= 0 && l < linelen_last)
			p += sprintf(p, "%*s", linelen_last - l, "");
		linelen_last = l;

		for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
			free(rate_str[ddir]);
			free(iops_str[ddir]);
		}
	}
	p += sprintf(p, "\r");

	printf("%s", output);

	if (!eta_new_line_init) {
		fio_gettime(&disp_eta_new_line, NULL);
		eta_new_line_init = 1;
	} else if (eta_new_line && mtime_since_now(&disp_eta_new_line) > eta_new_line) {
		fio_gettime(&disp_eta_new_line, NULL);
		eta_new_line_pending = 1;
	}

	fflush(stdout);
}

struct jobs_eta *get_jobs_eta(int force, size_t *size)
{
	struct jobs_eta *je;

	if (!thread_number)
		return NULL;

	*size = sizeof(*je) + THREAD_RUNSTR_SZ;
	je = malloc(*size);
	if (!je)
		return NULL;
	memset(je, 0, *size);

	if (!calc_thread_status(je, force)) {
		free(je);
		return NULL;
	}

	*size = sizeof(*je) + strlen((char *) je->run_str) + 1;
	return je;
}

void print_thread_status(void)
{
	struct jobs_eta *je;
	size_t size;

	je = get_jobs_eta(0, &size);
	if (je)
		display_thread_status(je);

	free(je);
}

void print_status_init(int thr_number)
{
	__run_str[thr_number] = 'P';
	update_condensed_str(__run_str, run_str);
}
Commit	Line	Data
	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	#include "lib/pow2.h"
	10
	11	static char __run_str[REAL_MAX_JOBS + 1];
	12	static char run_str[__THREAD_RUNSTR_SZ(REAL_MAX_JOBS)];
	13
	14	static void update_condensed_str(char rstr, char run_str_condensed)
	15	{
	16	if (*rstr) {
	17	while (*rstr) {
	18	int nr = 1;
	19
	20	run_str_condensed++ = rstr++;
	21	while ((rstr - 1) == rstr) {
	22	rstr++;
	23	nr++;
	24	}
	25	run_str_condensed += sprintf(run_str_condensed, "(%u),", nr);
	26	}
	27	run_str_condensed--;
	28	}
	29	*run_str_condensed = '\0';
	30	}
	31
	32	/*
	33	* Sets the status of the 'td' in the printed status map.
	34	*/
	35	static void check_str_update(struct thread_data *td)
	36	{
	37	char c = __run_str[td->thread_number - 1];
	38
	39	switch (td->runstate) {
	40	case TD_REAPED:
	41	if (td->error)
	42	c = 'X';
	43	else if (td->sig)
	44	c = 'K';
	45	else
	46	c = '_';
	47	break;
	48	case TD_EXITED:
	49	c = 'E';
	50	break;
	51	case TD_RAMP:
	52	c = '/';
	53	break;
	54	case TD_RUNNING:
	55	if (td_rw(td)) {
	56	if (td_random(td)) {
	57	if (td->o.rwmix[DDIR_READ] == 100)
	58	c = 'r';
	59	else if (td->o.rwmix[DDIR_WRITE] == 100)
	60	c = 'w';
	61	else
	62	c = 'm';
	63	} else {
	64	if (td->o.rwmix[DDIR_READ] == 100)
	65	c = 'R';
	66	else if (td->o.rwmix[DDIR_WRITE] == 100)
	67	c = 'W';
	68	else
	69	c = 'M';
	70	}
	71	} else if (td_read(td)) {
	72	if (td_random(td))
	73	c = 'r';
	74	else
	75	c = 'R';
	76	} else if (td_write(td)) {
	77	if (td_random(td))
	78	c = 'w';
	79	else
	80	c = 'W';
	81	} else {
	82	if (td_random(td))
	83	c = 'd';
	84	else
	85	c = 'D';
	86	}
	87	break;
	88	case TD_PRE_READING:
	89	c = 'p';
	90	break;
	91	case TD_VERIFYING:
	92	c = 'V';
	93	break;
	94	case TD_FSYNCING:
	95	c = 'F';
	96	break;
	97	case TD_FINISHING:
	98	c = 'f';
	99	break;
	100	case TD_CREATED:
	101	c = 'C';
	102	break;
	103	case TD_INITIALIZED:
	104	case TD_SETTING_UP:
	105	c = 'I';
	106	break;
	107	case TD_NOT_CREATED:
	108	c = 'P';
	109	break;
	110	default:
	111	log_err("state %d\n", td->runstate);
	112	}
	113
	114	__run_str[td->thread_number - 1] = c;
	115	update_condensed_str(__run_str, run_str);
	116	}
	117
	118	/*
	119	* Convert seconds to a printable string.
	120	*/
	121	void eta_to_str(char *str, unsigned long eta_sec)
	122	{
	123	unsigned int d, h, m, s;
	124	int disp_hour = 0;
	125
	126	if (eta_sec == -1) {
	127	sprintf(str, "--");
	128	return;
	129	}
	130
	131	s = eta_sec % 60;
	132	eta_sec /= 60;
	133	m = eta_sec % 60;
	134	eta_sec /= 60;
	135	h = eta_sec % 24;
	136	eta_sec /= 24;
	137	d = eta_sec;
	138
	139	if (d) {
	140	disp_hour = 1;
	141	str += sprintf(str, "%02ud:", d);
	142	}
	143
	144	if (h \|\| disp_hour)
	145	str += sprintf(str, "%02uh:", h);
	146
	147	str += sprintf(str, "%02um:", m);
	148	str += sprintf(str, "%02us", s);
	149	}
	150
	151	/*
	152	* Best effort calculation of the estimated pending runtime of a job.
	153	*/
	154	static unsigned long thread_eta(struct thread_data *td)
	155	{
	156	unsigned long long bytes_total, bytes_done;
	157	unsigned long eta_sec = 0;
	158	unsigned long elapsed;
	159	uint64_t timeout;
	160
	161	elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
	162	timeout = td->o.timeout / 1000000UL;
	163
	164	bytes_total = td->total_io_size;
	165
	166	if (td->flags & TD_F_NO_PROGRESS)
	167	return -1;
	168
	169	if (td->o.fill_device && td->o.size == -1ULL) {
	170	if (!td->fill_device_size \|\| td->fill_device_size == -1ULL)
	171	return 0;
	172
	173	bytes_total = td->fill_device_size;
	174	}
	175
	176	if (td->o.zone_size && td->o.zone_skip && bytes_total) {
	177	unsigned int nr_zones;
	178	uint64_t zone_bytes;
	179
	180	zone_bytes = bytes_total + td->o.zone_size + td->o.zone_skip;
	181	nr_zones = (zone_bytes - 1) / (td->o.zone_size + td->o.zone_skip);
	182	bytes_total -= nr_zones * td->o.zone_skip;
	183	}
	184
	185	/*
	186	* if writing and verifying afterwards, bytes_total will be twice the
	187	* size. In a mixed workload, verify phase will be the size of the
	188	* first stage writes.
	189	*/
	190	if (td->o.do_verify && td->o.verify && td_write(td)) {
	191	if (td_rw(td)) {
	192	unsigned int perc = 50;
	193
	194	if (td->o.rwmix[DDIR_WRITE])
	195	perc = td->o.rwmix[DDIR_WRITE];
	196
	197	bytes_total += (bytes_total * perc) / 100;
	198	} else
	199	bytes_total <<= 1;
	200	}
	201
	202	if (td->runstate == TD_RUNNING \|\| td->runstate == TD_VERIFYING) {
	203	double perc, perc_t;
	204
	205	bytes_done = ddir_rw_sum(td->io_bytes);
	206
	207	if (bytes_total) {
	208	perc = (double) bytes_done / (double) bytes_total;
	209	if (perc > 1.0)
	210	perc = 1.0;
	211	} else
	212	perc = 0.0;
	213
	214	if (td->o.time_based) {
	215	if (timeout) {
	216	perc_t = (double) elapsed / (double) timeout;
	217	if (perc_t < perc)
	218	perc = perc_t;
	219	} else {
	220	/*
	221	* Will never hit, we can't have time_based
	222	* without a timeout set.
	223	*/
	224	perc = 0.0;
	225	}
	226	}
	227
	228	eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
	229
	230	if (td->o.timeout &&
	231	eta_sec > (timeout + done_secs - elapsed))
	232	eta_sec = timeout + done_secs - elapsed;
	233	} else if (td->runstate == TD_NOT_CREATED \|\| td->runstate == TD_CREATED
	234	\|\| td->runstate == TD_INITIALIZED
	235	\|\| td->runstate == TD_SETTING_UP
	236	\|\| td->runstate == TD_RAMP
	237	\|\| td->runstate == TD_PRE_READING) {
	238	int t_eta = 0, r_eta = 0;
	239	unsigned long long rate_bytes;
	240
	241	/*
	242	* We can only guess - assume it'll run the full timeout
	243	* if given, otherwise assume it'll run at the specified rate.
	244	*/
	245	if (td->o.timeout) {
	246	uint64_t __timeout = td->o.timeout;
	247	uint64_t start_delay = td->o.start_delay;
	248	uint64_t ramp_time = td->o.ramp_time;
	249
	250	t_eta = __timeout + start_delay + ramp_time;
	251	t_eta /= 1000000ULL;
	252
	253	if (in_ramp_time(td)) {
	254	unsigned long ramp_left;
	255
	256	ramp_left = mtime_since_now(&td->epoch);
	257	ramp_left = (ramp_left + 999) / 1000;
	258	if (ramp_left <= t_eta)
	259	t_eta -= ramp_left;
	260	}
	261	}
	262	rate_bytes = ddir_rw_sum(td->o.rate);
	263	if (rate_bytes) {
	264	r_eta = (bytes_total / 1024) / rate_bytes;
	265	r_eta += (td->o.start_delay / 1000000ULL);
	266	}
	267
	268	if (r_eta && t_eta)
	269	eta_sec = min(r_eta, t_eta);
	270	else if (r_eta)
	271	eta_sec = r_eta;
	272	else if (t_eta)
	273	eta_sec = t_eta;
	274	else
	275	eta_sec = 0;
	276	} else {
	277	/*
	278	* thread is already done or waiting for fsync
	279	*/
	280	eta_sec = 0;
	281	}
	282
	283	return eta_sec;
	284	}
	285
	286	static void calc_rate(int unified_rw_rep, unsigned long mtime,
	287	unsigned long long *io_bytes,
	288	unsigned long long prev_io_bytes, unsigned int rate)
	289	{
	290	int i;
	291
	292	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	293	unsigned long long diff, this_rate;
	294
	295	diff = io_bytes[i] - prev_io_bytes[i];
	296	if (mtime)
	297	this_rate = ((1000 * diff) / mtime) / 1024;
	298	else
	299	this_rate = 0;
	300
	301	if (unified_rw_rep) {
	302	rate[i] = 0;
	303	rate[0] += this_rate;
	304	} else
	305	rate[i] = this_rate;
	306
	307	prev_io_bytes[i] = io_bytes[i];
	308	}
	309	}
	310
	311	static void calc_iops(int unified_rw_rep, unsigned long mtime,
	312	unsigned long long *io_iops,
	313	unsigned long long prev_io_iops, unsigned int iops)
	314	{
	315	int i;
	316
	317	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	318	unsigned long long diff;
	319
	320	diff = io_iops[i] - prev_io_iops[i];
	321	if (unified_rw_rep) {
	322	iops[i] = 0;
	323	iops[0] += (diff * 1000) / mtime;
	324	} else
	325	iops[i] = (diff * 1000) / mtime;
	326
	327	prev_io_iops[i] = io_iops[i];
	328	}
	329	}
	330
	331	/*
	332	* Print status of the jobs we know about. This includes rate estimates,
	333	* ETA, thread state, etc.
	334	*/
	335	int calc_thread_status(struct jobs_eta *je, int force)
	336	{
	337	struct thread_data *td;
	338	int i, unified_rw_rep;
	339	unsigned long rate_time, disp_time, bw_avg_time, *eta_secs;
	340	unsigned long long io_bytes[DDIR_RWDIR_CNT];
	341	unsigned long long io_iops[DDIR_RWDIR_CNT];
	342	struct timeval now;
	343
	344	static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
	345	static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
	346	static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
	347	static struct timeval rate_prev_time, disp_prev_time;
	348
	349	if (!force) {
	350	if (output_format != FIO_OUTPUT_NORMAL &&
	351	f_out == stdout)
	352	return 0;
	353	if (temp_stall_ts \|\| eta_print == FIO_ETA_NEVER)
	354	return 0;
	355
	356	if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
	357	return 0;
	358	}
	359
	360	if (!ddir_rw_sum(rate_io_bytes))
	361	fill_start_time(&rate_prev_time);
	362	if (!ddir_rw_sum(disp_io_bytes))
	363	fill_start_time(&disp_prev_time);
	364
	365	eta_secs = malloc(thread_number * sizeof(unsigned long));
	366	memset(eta_secs, 0, thread_number * sizeof(unsigned long));
	367
	368	je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
	369
	370	io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
	371	io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
	372	bw_avg_time = ULONG_MAX;
	373	unified_rw_rep = 0;
	374	for_each_td(td, i) {
	375	unified_rw_rep += td->o.unified_rw_rep;
	376	if (is_power_of_2(td->o.kb_base))
	377	je->is_pow2 = 1;
	378	je->unit_base = td->o.unit_base;
	379	if (td->o.bw_avg_time < bw_avg_time)
	380	bw_avg_time = td->o.bw_avg_time;
	381	if (td->runstate == TD_RUNNING \|\| td->runstate == TD_VERIFYING
	382	\|\| td->runstate == TD_FSYNCING
	383	\|\| td->runstate == TD_PRE_READING
	384	\|\| td->runstate == TD_FINISHING) {
	385	je->nr_running++;
	386	if (td_read(td)) {
	387	je->t_rate[0] += td->o.rate[DDIR_READ];
	388	je->t_iops[0] += td->o.rate_iops[DDIR_READ];
	389	je->m_rate[0] += td->o.ratemin[DDIR_READ];
	390	je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
	391	}
	392	if (td_write(td)) {
	393	je->t_rate[1] += td->o.rate[DDIR_WRITE];
	394	je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
	395	je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
	396	je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
	397	}
	398	if (td_trim(td)) {
	399	je->t_rate[2] += td->o.rate[DDIR_TRIM];
	400	je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
	401	je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
	402	je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
	403	}
	404
	405	je->files_open += td->nr_open_files;
	406	} else if (td->runstate == TD_RAMP) {
	407	je->nr_running++;
	408	je->nr_ramp++;
	409	} else if (td->runstate == TD_SETTING_UP)
	410	je->nr_setting_up++;
	411	else if (td->runstate < TD_RUNNING)
	412	je->nr_pending++;
	413
	414	if (je->elapsed_sec >= 3)
	415	eta_secs[i] = thread_eta(td);
	416	else
	417	eta_secs[i] = INT_MAX;
	418
	419	check_str_update(td);
	420
	421	if (td->runstate > TD_SETTING_UP) {
	422	int ddir;
	423
	424	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
	425	if (unified_rw_rep) {
	426	io_bytes[0] += td->io_bytes[ddir];
	427	io_iops[0] += td->io_blocks[ddir];
	428	} else {
	429	io_bytes[ddir] += td->io_bytes[ddir];
	430	io_iops[ddir] += td->io_blocks[ddir];
	431	}
	432	}
	433	}
	434	}
	435
	436	if (exitall_on_terminate)
	437	je->eta_sec = INT_MAX;
	438	else
	439	je->eta_sec = 0;
	440
	441	for_each_td(td, i) {
	442	if (exitall_on_terminate) {
	443	if (eta_secs[i] < je->eta_sec)
	444	je->eta_sec = eta_secs[i];
	445	} else {
	446	if (eta_secs[i] > je->eta_sec)
	447	je->eta_sec = eta_secs[i];
	448	}
	449	}
	450
	451	free(eta_secs);
	452
	453	fio_gettime(&now, NULL);
	454	rate_time = mtime_since(&rate_prev_time, &now);
	455
	456	if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
	457	calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
	458	je->rate);
	459	memcpy(&rate_prev_time, &now, sizeof(now));
	460	add_agg_sample(je->rate[DDIR_READ], DDIR_READ, 0);
	461	add_agg_sample(je->rate[DDIR_WRITE], DDIR_WRITE, 0);
	462	add_agg_sample(je->rate[DDIR_TRIM], DDIR_TRIM, 0);
	463	}
	464
	465	disp_time = mtime_since(&disp_prev_time, &now);
	466
	467	/*
	468	* Allow a little slack, the target is to print it every 1000 msecs
	469	*/
	470	if (!force && disp_time < 900)
	471	return 0;
	472
	473	calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
	474	calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);
	475
	476	memcpy(&disp_prev_time, &now, sizeof(now));
	477
	478	if (!force && !je->nr_running && !je->nr_pending)
	479	return 0;
	480
	481	je->nr_threads = thread_number;
	482	update_condensed_str(__run_str, run_str);
	483	memcpy(je->run_str, run_str, strlen(run_str));
	484	return 1;
	485	}
	486
	487	void display_thread_status(struct jobs_eta *je)
	488	{
	489	static struct timeval disp_eta_new_line;
	490	static int eta_new_line_init, eta_new_line_pending;
	491	static int linelen_last;
	492	static int eta_good;
	493	char output[REAL_MAX_JOBS + 512], *p = output;
	494	char eta_str[128];
	495	double perc = 0.0;
	496
	497	if (je->eta_sec != INT_MAX && je->elapsed_sec) {
	498	perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
	499	eta_to_str(eta_str, je->eta_sec);
	500	}
	501
	502	if (eta_new_line_pending) {
	503	eta_new_line_pending = 0;
	504	p += sprintf(p, "\n");
	505	}
	506
	507	p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
	508	if (je->m_rate[0] \|\| je->m_rate[1] \|\| je->t_rate[0] \|\| je->t_rate[1]) {
	509	char tr, mr;
	510
	511	mr = num2str(je->m_rate[0] + je->m_rate[1], 4, 0, je->is_pow2, 8);
	512	tr = num2str(je->t_rate[0] + je->t_rate[1], 4, 0, je->is_pow2, 8);
	513	p += sprintf(p, ", CR=%s/%s KB/s", tr, mr);
	514	free(tr);
	515	free(mr);
	516	} else if (je->m_iops[0] \|\| je->m_iops[1] \|\| je->t_iops[0] \|\| je->t_iops[1]) {
	517	p += sprintf(p, ", CR=%d/%d IOPS",
	518	je->t_iops[0] + je->t_iops[1],
	519	je->m_iops[0] + je->m_iops[1]);
	520	}
	521	if (je->eta_sec != INT_MAX && je->nr_running) {
	522	char perc_str[32];
	523	char *iops_str[DDIR_RWDIR_CNT];
	524	char *rate_str[DDIR_RWDIR_CNT];
	525	size_t left;
	526	int l;
	527	int ddir;
	528
	529	if ((!je->eta_sec && !eta_good) \|\| je->nr_ramp == je->nr_running \|\|
	530	je->eta_sec == -1)
	531	strcpy(perc_str, "-.-% done");
	532	else {
	533	double mult = 100.0;
	534
	535	if (je->nr_setting_up && je->nr_running)
	536	mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);
	537
	538	eta_good = 1;
	539	perc *= mult;
	540	sprintf(perc_str, "%3.1f%% done", perc);
	541	}
	542
	543	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
	544	rate_str[ddir] = num2str(je->rate[ddir], 5,
	545	1024, je->is_pow2, je->unit_base);
	546	iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, 0);
	547	}
	548
	549	left = sizeof(output) - (p - output) - 1;
	550
	551	l = snprintf(p, left, ": [%s] [%s] [%s/%s/%s /s] [%s/%s/%s iops] [eta %s]",
	552	je->run_str, perc_str, rate_str[DDIR_READ],
	553	rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
	554	iops_str[DDIR_READ], iops_str[DDIR_WRITE],
	555	iops_str[DDIR_TRIM], eta_str);
	556	p += l;
	557	if (l >= 0 && l < linelen_last)
	558	p += sprintf(p, "%*s", linelen_last - l, "");
	559	linelen_last = l;
	560
	561	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
	562	free(rate_str[ddir]);
	563	free(iops_str[ddir]);
	564	}
	565	}
	566	p += sprintf(p, "\r");
	567
	568	printf("%s", output);
	569
	570	if (!eta_new_line_init) {
	571	fio_gettime(&disp_eta_new_line, NULL);
	572	eta_new_line_init = 1;
	573	} else if (eta_new_line && mtime_since_now(&disp_eta_new_line) > eta_new_line) {
	574	fio_gettime(&disp_eta_new_line, NULL);
	575	eta_new_line_pending = 1;
	576	}
	577
	578	fflush(stdout);
	579	}
	580
	581	struct jobs_eta get_jobs_eta(int force, size_t size)
	582	{
	583	struct jobs_eta *je;
	584
	585	if (!thread_number)
	586	return NULL;
	587
	588	size = sizeof(je) + THREAD_RUNSTR_SZ;
	589	je = malloc(*size);
	590	if (!je)
	591	return NULL;
	592	memset(je, 0, *size);
	593
	594	if (!calc_thread_status(je, force)) {
	595	free(je);
	596	return NULL;
	597	}
	598
	599	size = sizeof(je) + strlen((char *) je->run_str) + 1;
	600	return je;
	601	}
	602
	603	void print_thread_status(void)
	604	{
	605	struct jobs_eta *je;
	606	size_t size;
	607
	608	je = get_jobs_eta(0, &size);
	609	if (je)
	610	display_thread_status(je);
	611
	612	free(je);
	613	}
	614
	615	void print_status_init(int thr_number)
	616	{
	617	__run_str[thr_number] = 'P';
	618	update_condensed_str(__run_str, run_str);
	619	}