[fio.git] / eta.c

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

#include "fio.h"

static char run_str[REAL_MAX_JOBS + 1];

/*
 * 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_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;
}

/*
 * 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;

	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 int thread_eta(struct thread_data *td)
{
	unsigned long long bytes_total, bytes_done;
	unsigned long eta_sec = 0;
	unsigned long elapsed;

	elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;

	bytes_total = td->total_io_size;

	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);
		perc = (double) bytes_done / (double) bytes_total;
		if (perc > 1.0)
			perc = 1.0;

		if (td->o.time_based) {
			perc_t = (double) elapsed / (double) td->o.timeout;
			if (perc_t < perc)
				perc = perc_t;
		}

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

		if (td->o.timeout &&
		    eta_sec > (td->o.timeout + done_secs - elapsed))
			eta_sec = td->o.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) {
			t_eta = td->o.timeout + td->o.start_delay +
					td->o.ramp_time;

			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;
		}

		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;

		diff = io_bytes[i] - prev_io_bytes[i];
		if (unified_rw_rep) {
			rate[i] = 0;
			rate[0] += ((1000 * diff) / mtime) / 1024;
		} else
			rate[i] = ((1000 * diff) / mtime) / 1024;

		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) {
			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_running++;
			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;
	memcpy(je->run_str, run_str, thread_number * sizeof(char));
	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)
			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 * 1000) {
		fio_gettime(&disp_eta_new_line, NULL);
		eta_new_line_pending = 1;
	}

	fflush(stdout);
}

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

	if (!thread_number)
		return;

	size = sizeof(*je) + thread_number * sizeof(char) + 1;
	je = malloc(size);
	memset(je, 0, size);

	if (calc_thread_status(je, 0))
		display_thread_status(je);

	free(je);
}

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