[fio.git] / eta.c

/*
 * Status and ETA code
 */
#include <unistd.h>
#include <fcntl.h>
#include <string.h>
#ifdef CONFIG_VALGRIND_DEV
#include <valgrind/drd.h>
#else
#define DRD_IGNORE_VAR(x) do { } while (0)
#endif

#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) + 1];

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;
			}
		}

		if (perc == 0.0) {
			eta_sec = timeout;
		} else {
			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) {
		int64_t 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;
			if (!td->ramp_time_over) {
				t_eta += ramp_time;
			}
			t_eta /= 1000000ULL;

			if ((td->runstate == TD_RAMP) && 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 = 0;
		if (td_read(td))
			rate_bytes  = td->o.rate[DDIR_READ];
		if (td_write(td))
			rate_bytes += td->o.rate[DDIR_WRITE];
		if (td_trim(td))
			rate_bytes += td->o.rate[DDIR_TRIM];

		if (rate_bytes) {
			r_eta = bytes_total / 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, uint64_t *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; /* KiB/s */
		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, this_iops;

		diff = io_iops[i] - prev_io_iops[i];
		if (mtime)
			this_iops = (diff * 1000) / mtime;
		else
			this_iops = 0;

		if (unified_rw_rep) {
			iops[i] = 0;
			iops[0] += this_iops;
		} else
			iops[i] = this_iops;

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

/*
 * Allow a little slack - if we're within 95% of the time, allow ETA.
 */
bool eta_time_within_slack(unsigned int time)
{
	return time > ((eta_interval_msec * 95) / 100);
}

/*
 * Print status of the jobs we know about. This includes rate estimates,
 * ETA, thread state, etc.
 */
bool calc_thread_status(struct jobs_eta *je, int force)
{
	struct thread_data *td;
	int i, unified_rw_rep;
	uint64_t 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 timespec 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 timespec rate_prev_time, disp_prev_time;

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

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

	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(uint64_t));
	memset(eta_secs, 0, thread_number * sizeof(uint64_t));

	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 = 0; 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;
		for_each_td(td, i) {
			if (eta_secs[i] < je->eta_sec)
				je->eta_sec = eta_secs[i];
		}
	} else {
		unsigned long eta_stone = 0;

		je->eta_sec = 0;
		for_each_td(td, i) {
			if ((td->runstate == TD_NOT_CREATED) && td->o.stonewall)
				eta_stone += eta_secs[i];
			else {
				if (eta_secs[i] > je->eta_sec)
					je->eta_sec = eta_secs[i];
			}
		}
		je->eta_sec += eta_stone;
	}

	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(sample_val(je->rate[DDIR_READ]), DDIR_READ, 0);
		add_agg_sample(sample_val(je->rate[DDIR_WRITE]), DDIR_WRITE, 0);
		add_agg_sample(sample_val(je->rate[DDIR_TRIM]), DDIR_TRIM, 0);
	}

	disp_time = mtime_since(&disp_prev_time, &now);

	if (!force && !eta_time_within_slack(disp_time))
		return false;

	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 false;

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

void display_thread_status(struct jobs_eta *je)
{
	static struct timespec disp_eta_new_line;
	static int eta_new_line_init, eta_new_line_pending;
	static int linelen_last;
	static int eta_good;
	char output[__THREAD_RUNSTR_SZ(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;
		linelen_last = 0;
		p += sprintf(p, "\n");
	}

	p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);

	/* rate limits, if any */
	if (je->m_rate[0] || je->m_rate[1] || je->m_rate[2] ||
	    je->t_rate[0] || je->t_rate[1] || je->t_rate[2]) {
		char *tr, *mr;

		mr = num2str(je->m_rate[0] + je->m_rate[1] + je->m_rate[2],
				je->sig_figs, 0, je->is_pow2, N2S_BYTEPERSEC);
		tr = num2str(je->t_rate[0] + je->t_rate[1] + je->t_rate[2],
				je->sig_figs, 0, je->is_pow2, N2S_BYTEPERSEC);

		p += sprintf(p, ", %s-%s", mr, tr);
		free(tr);
		free(mr);
	} else if (je->m_iops[0] || je->m_iops[1] || je->m_iops[2] ||
		   je->t_iops[0] || je->t_iops[1] || je->t_iops[2]) {
		p += sprintf(p, ", %d-%d IOPS",
					je->m_iops[0] + je->m_iops[1] + je->m_iops[2],
					je->t_iops[0] + je->t_iops[1] + je->t_iops[2]);
	}

	/* current run string, % done, bandwidth, iops, eta */
	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;
		int linelen;

		if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running ||
		    je->eta_sec == -1)
			strcpy(perc_str, "-.-%");
		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%%", perc);
		}

		for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
			rate_str[ddir] = num2str(je->rate[ddir], 4,
						1024, je->is_pow2, je->unit_base);
			iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, N2S_NONE);
		}

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

		if (je->rate[DDIR_TRIM] || je->iops[DDIR_TRIM])
			l = snprintf(p, left,
				": [%s][%s][r=%s,w=%s,t=%s][r=%s,w=%s,t=%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);
		else
			l = snprintf(p, left,
				": [%s][%s][r=%s,w=%s][r=%s,w=%s IOPS][eta %s]",
				je->run_str, perc_str,
				rate_str[DDIR_READ], rate_str[DDIR_WRITE],
				iops_str[DDIR_READ], iops_str[DDIR_WRITE],
				eta_str);
		/* If truncation occurred adjust l so p is on the null */
		if (l >= left)
			l = left - 1;
		p += l;
		linelen = p - output;
		if (l >= 0 && linelen < linelen_last)
			p += sprintf(p, "%*s", linelen_last - linelen, "");
		linelen_last = linelen;

		for (ddir = 0; 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(bool force, size_t *size)
{
	struct jobs_eta *je;

	if (!thread_number)
		return NULL;

	*size = sizeof(*je) + THREAD_RUNSTR_SZ + 8;
	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(false, &size);
	if (je)
		display_thread_status(je);

	free(je);
}

void print_status_init(int thr_number)
{
	DRD_IGNORE_VAR(__run_str);
	__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	#ifdef CONFIG_VALGRIND_DEV
	8	#include <valgrind/drd.h>
	9	#else
	10	#define DRD_IGNORE_VAR(x) do { } while (0)
	11	#endif
	12
	13	#include "fio.h"
	14	#include "lib/pow2.h"
	15
	16	static char __run_str[REAL_MAX_JOBS + 1];
	17	static char run_str[__THREAD_RUNSTR_SZ(REAL_MAX_JOBS) + 1];
	18
	19	static void update_condensed_str(char rstr, char run_str_condensed)
	20	{
	21	if (*rstr) {
	22	while (*rstr) {
	23	int nr = 1;
	24
	25	run_str_condensed++ = rstr++;
	26	while ((rstr - 1) == rstr) {
	27	rstr++;
	28	nr++;
	29	}
	30	run_str_condensed += sprintf(run_str_condensed, "(%u),", nr);
	31	}
	32	run_str_condensed--;
	33	}
	34	*run_str_condensed = '\0';
	35	}
	36
	37	/*
	38	* Sets the status of the 'td' in the printed status map.
	39	*/
	40	static void check_str_update(struct thread_data *td)
	41	{
	42	char c = __run_str[td->thread_number - 1];
	43
	44	switch (td->runstate) {
	45	case TD_REAPED:
	46	if (td->error)
	47	c = 'X';
	48	else if (td->sig)
	49	c = 'K';
	50	else
	51	c = '_';
	52	break;
	53	case TD_EXITED:
	54	c = 'E';
	55	break;
	56	case TD_RAMP:
	57	c = '/';
	58	break;
	59	case TD_RUNNING:
	60	if (td_rw(td)) {
	61	if (td_random(td)) {
	62	if (td->o.rwmix[DDIR_READ] == 100)
	63	c = 'r';
	64	else if (td->o.rwmix[DDIR_WRITE] == 100)
	65	c = 'w';
	66	else
	67	c = 'm';
	68	} else {
	69	if (td->o.rwmix[DDIR_READ] == 100)
	70	c = 'R';
	71	else if (td->o.rwmix[DDIR_WRITE] == 100)
	72	c = 'W';
	73	else
	74	c = 'M';
	75	}
	76	} else if (td_read(td)) {
	77	if (td_random(td))
	78	c = 'r';
	79	else
	80	c = 'R';
	81	} else if (td_write(td)) {
	82	if (td_random(td))
	83	c = 'w';
	84	else
	85	c = 'W';
	86	} else {
	87	if (td_random(td))
	88	c = 'd';
	89	else
	90	c = 'D';
	91	}
	92	break;
	93	case TD_PRE_READING:
	94	c = 'p';
	95	break;
	96	case TD_VERIFYING:
	97	c = 'V';
	98	break;
	99	case TD_FSYNCING:
	100	c = 'F';
	101	break;
	102	case TD_FINISHING:
	103	c = 'f';
	104	break;
	105	case TD_CREATED:
	106	c = 'C';
	107	break;
	108	case TD_INITIALIZED:
	109	case TD_SETTING_UP:
	110	c = 'I';
	111	break;
	112	case TD_NOT_CREATED:
	113	c = 'P';
	114	break;
	115	default:
	116	log_err("state %d\n", td->runstate);
	117	}
	118
	119	__run_str[td->thread_number - 1] = c;
	120	update_condensed_str(__run_str, run_str);
	121	}
	122
	123	/*
	124	* Convert seconds to a printable string.
	125	*/
	126	void eta_to_str(char *str, unsigned long eta_sec)
	127	{
	128	unsigned int d, h, m, s;
	129	int disp_hour = 0;
	130
	131	if (eta_sec == -1) {
	132	sprintf(str, "--");
	133	return;
	134	}
	135
	136	s = eta_sec % 60;
	137	eta_sec /= 60;
	138	m = eta_sec % 60;
	139	eta_sec /= 60;
	140	h = eta_sec % 24;
	141	eta_sec /= 24;
	142	d = eta_sec;
	143
	144	if (d) {
	145	disp_hour = 1;
	146	str += sprintf(str, "%02ud:", d);
	147	}
	148
	149	if (h \|\| disp_hour)
	150	str += sprintf(str, "%02uh:", h);
	151
	152	str += sprintf(str, "%02um:", m);
	153	str += sprintf(str, "%02us", s);
	154	}
	155
	156	/*
	157	* Best effort calculation of the estimated pending runtime of a job.
	158	*/
	159	static unsigned long thread_eta(struct thread_data *td)
	160	{
	161	unsigned long long bytes_total, bytes_done;
	162	unsigned long eta_sec = 0;
	163	unsigned long elapsed;
	164	uint64_t timeout;
	165
	166	elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
	167	timeout = td->o.timeout / 1000000UL;
	168
	169	bytes_total = td->total_io_size;
	170
	171	if (td->flags & TD_F_NO_PROGRESS)
	172	return -1;
	173
	174	if (td->o.fill_device && td->o.size == -1ULL) {
	175	if (!td->fill_device_size \|\| td->fill_device_size == -1ULL)
	176	return 0;
	177
	178	bytes_total = td->fill_device_size;
	179	}
	180
	181	if (td->o.zone_size && td->o.zone_skip && bytes_total) {
	182	unsigned int nr_zones;
	183	uint64_t zone_bytes;
	184
	185	zone_bytes = bytes_total + td->o.zone_size + td->o.zone_skip;
	186	nr_zones = (zone_bytes - 1) / (td->o.zone_size + td->o.zone_skip);
	187	bytes_total -= nr_zones * td->o.zone_skip;
	188	}
	189
	190	/*
	191	* if writing and verifying afterwards, bytes_total will be twice the
	192	* size. In a mixed workload, verify phase will be the size of the
	193	* first stage writes.
	194	*/
	195	if (td->o.do_verify && td->o.verify && td_write(td)) {
	196	if (td_rw(td)) {
	197	unsigned int perc = 50;
	198
	199	if (td->o.rwmix[DDIR_WRITE])
	200	perc = td->o.rwmix[DDIR_WRITE];
	201
	202	bytes_total += (bytes_total * perc) / 100;
	203	} else
	204	bytes_total <<= 1;
	205	}
	206
	207	if (td->runstate == TD_RUNNING \|\| td->runstate == TD_VERIFYING) {
	208	double perc, perc_t;
	209
	210	bytes_done = ddir_rw_sum(td->io_bytes);
	211
	212	if (bytes_total) {
	213	perc = (double) bytes_done / (double) bytes_total;
	214	if (perc > 1.0)
	215	perc = 1.0;
	216	} else
	217	perc = 0.0;
	218
	219	if (td->o.time_based) {
	220	if (timeout) {
	221	perc_t = (double) elapsed / (double) timeout;
	222	if (perc_t < perc)
	223	perc = perc_t;
	224	} else {
	225	/*
	226	* Will never hit, we can't have time_based
	227	* without a timeout set.
	228	*/
	229	perc = 0.0;
	230	}
	231	}
	232
	233	if (perc == 0.0) {
	234	eta_sec = timeout;
	235	} else {
	236	eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
	237	}
	238
	239	if (td->o.timeout &&
	240	eta_sec > (timeout + done_secs - elapsed))
	241	eta_sec = timeout + done_secs - elapsed;
	242	} else if (td->runstate == TD_NOT_CREATED \|\| td->runstate == TD_CREATED
	243	\|\| td->runstate == TD_INITIALIZED
	244	\|\| td->runstate == TD_SETTING_UP
	245	\|\| td->runstate == TD_RAMP
	246	\|\| td->runstate == TD_PRE_READING) {
	247	int64_t t_eta = 0, r_eta = 0;
	248	unsigned long long rate_bytes;
	249
	250	/*
	251	* We can only guess - assume it'll run the full timeout
	252	* if given, otherwise assume it'll run at the specified rate.
	253	*/
	254	if (td->o.timeout) {
	255	uint64_t __timeout = td->o.timeout;
	256	uint64_t start_delay = td->o.start_delay;
	257	uint64_t ramp_time = td->o.ramp_time;
	258
	259	t_eta = __timeout + start_delay;
	260	if (!td->ramp_time_over) {
	261	t_eta += ramp_time;
	262	}
	263	t_eta /= 1000000ULL;
	264
	265	if ((td->runstate == TD_RAMP) && in_ramp_time(td)) {
	266	unsigned long ramp_left;
	267
	268	ramp_left = mtime_since_now(&td->epoch);
	269	ramp_left = (ramp_left + 999) / 1000;
	270	if (ramp_left <= t_eta)
	271	t_eta -= ramp_left;
	272	}
	273	}
	274	rate_bytes = 0;
	275	if (td_read(td))
	276	rate_bytes = td->o.rate[DDIR_READ];
	277	if (td_write(td))
	278	rate_bytes += td->o.rate[DDIR_WRITE];
	279	if (td_trim(td))
	280	rate_bytes += td->o.rate[DDIR_TRIM];
	281
	282	if (rate_bytes) {
	283	r_eta = bytes_total / rate_bytes;
	284	r_eta += (td->o.start_delay / 1000000ULL);
	285	}
	286
	287	if (r_eta && t_eta)
	288	eta_sec = min(r_eta, t_eta);
	289	else if (r_eta)
	290	eta_sec = r_eta;
	291	else if (t_eta)
	292	eta_sec = t_eta;
	293	else
	294	eta_sec = 0;
	295	} else {
	296	/*
	297	* thread is already done or waiting for fsync
	298	*/
	299	eta_sec = 0;
	300	}
	301
	302	return eta_sec;
	303	}
	304
	305	static void calc_rate(int unified_rw_rep, unsigned long mtime,
	306	unsigned long long *io_bytes,
	307	unsigned long long prev_io_bytes, uint64_t rate)
	308	{
	309	int i;
	310
	311	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	312	unsigned long long diff, this_rate;
	313
	314	diff = io_bytes[i] - prev_io_bytes[i];
	315	if (mtime)
	316	this_rate = ((1000 * diff) / mtime) / 1024; /* KiB/s */
	317	else
	318	this_rate = 0;
	319
	320	if (unified_rw_rep) {
	321	rate[i] = 0;
	322	rate[0] += this_rate;
	323	} else
	324	rate[i] = this_rate;
	325
	326	prev_io_bytes[i] = io_bytes[i];
	327	}
	328	}
	329
	330	static void calc_iops(int unified_rw_rep, unsigned long mtime,
	331	unsigned long long *io_iops,
	332	unsigned long long prev_io_iops, unsigned int iops)
	333	{
	334	int i;
	335
	336	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	337	unsigned long long diff, this_iops;
	338
	339	diff = io_iops[i] - prev_io_iops[i];
	340	if (mtime)
	341	this_iops = (diff * 1000) / mtime;
	342	else
	343	this_iops = 0;
	344
	345	if (unified_rw_rep) {
	346	iops[i] = 0;
	347	iops[0] += this_iops;
	348	} else
	349	iops[i] = this_iops;
	350
	351	prev_io_iops[i] = io_iops[i];
	352	}
	353	}
	354
	355	/*
	356	* Allow a little slack - if we're within 95% of the time, allow ETA.
	357	*/
	358	bool eta_time_within_slack(unsigned int time)
	359	{
	360	return time > ((eta_interval_msec * 95) / 100);
	361	}
	362
	363	/*
	364	* Print status of the jobs we know about. This includes rate estimates,
	365	* ETA, thread state, etc.
	366	*/
	367	bool calc_thread_status(struct jobs_eta *je, int force)
	368	{
	369	struct thread_data *td;
	370	int i, unified_rw_rep;
	371	uint64_t rate_time, disp_time, bw_avg_time, *eta_secs;
	372	unsigned long long io_bytes[DDIR_RWDIR_CNT];
	373	unsigned long long io_iops[DDIR_RWDIR_CNT];
	374	struct timespec now;
	375
	376	static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
	377	static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
	378	static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
	379	static struct timespec rate_prev_time, disp_prev_time;
	380
	381	if (!force) {
	382	if (!(output_format & FIO_OUTPUT_NORMAL) &&
	383	f_out == stdout)
	384	return false;
	385	if (temp_stall_ts \|\| eta_print == FIO_ETA_NEVER)
	386	return false;
	387
	388	if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
	389	return false;
	390	}
	391
	392	if (!ddir_rw_sum(rate_io_bytes))
	393	fill_start_time(&rate_prev_time);
	394	if (!ddir_rw_sum(disp_io_bytes))
	395	fill_start_time(&disp_prev_time);
	396
	397	eta_secs = malloc(thread_number * sizeof(uint64_t));
	398	memset(eta_secs, 0, thread_number * sizeof(uint64_t));
	399
	400	je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
	401
	402	io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
	403	io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
	404	bw_avg_time = ULONG_MAX;
	405	unified_rw_rep = 0;
	406	for_each_td(td, i) {
	407	unified_rw_rep += td->o.unified_rw_rep;
	408	if (is_power_of_2(td->o.kb_base))
	409	je->is_pow2 = 1;
	410	je->unit_base = td->o.unit_base;
	411	if (td->o.bw_avg_time < bw_avg_time)
	412	bw_avg_time = td->o.bw_avg_time;
	413	if (td->runstate == TD_RUNNING \|\| td->runstate == TD_VERIFYING
	414	\|\| td->runstate == TD_FSYNCING
	415	\|\| td->runstate == TD_PRE_READING
	416	\|\| td->runstate == TD_FINISHING) {
	417	je->nr_running++;
	418	if (td_read(td)) {
	419	je->t_rate[0] += td->o.rate[DDIR_READ];
	420	je->t_iops[0] += td->o.rate_iops[DDIR_READ];
	421	je->m_rate[0] += td->o.ratemin[DDIR_READ];
	422	je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
	423	}
	424	if (td_write(td)) {
	425	je->t_rate[1] += td->o.rate[DDIR_WRITE];
	426	je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
	427	je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
	428	je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
	429	}
	430	if (td_trim(td)) {
	431	je->t_rate[2] += td->o.rate[DDIR_TRIM];
	432	je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
	433	je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
	434	je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
	435	}
	436
	437	je->files_open += td->nr_open_files;
	438	} else if (td->runstate == TD_RAMP) {
	439	je->nr_running++;
	440	je->nr_ramp++;
	441	} else if (td->runstate == TD_SETTING_UP)
	442	je->nr_setting_up++;
	443	else if (td->runstate < TD_RUNNING)
	444	je->nr_pending++;
	445
	446	if (je->elapsed_sec >= 3)
	447	eta_secs[i] = thread_eta(td);
	448	else
	449	eta_secs[i] = INT_MAX;
	450
	451	check_str_update(td);
	452
	453	if (td->runstate > TD_SETTING_UP) {
	454	int ddir;
	455
	456	for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
	457	if (unified_rw_rep) {
	458	io_bytes[0] += td->io_bytes[ddir];
	459	io_iops[0] += td->io_blocks[ddir];
	460	} else {
	461	io_bytes[ddir] += td->io_bytes[ddir];
	462	io_iops[ddir] += td->io_blocks[ddir];
	463	}
	464	}
	465	}
	466	}
	467
	468	if (exitall_on_terminate) {
	469	je->eta_sec = INT_MAX;
	470	for_each_td(td, i) {
	471	if (eta_secs[i] < je->eta_sec)
	472	je->eta_sec = eta_secs[i];
	473	}
	474	} else {
	475	unsigned long eta_stone = 0;
	476
	477	je->eta_sec = 0;
	478	for_each_td(td, i) {
	479	if ((td->runstate == TD_NOT_CREATED) && td->o.stonewall)
	480	eta_stone += eta_secs[i];
	481	else {
	482	if (eta_secs[i] > je->eta_sec)
	483	je->eta_sec = eta_secs[i];
	484	}
	485	}
	486	je->eta_sec += eta_stone;
	487	}
	488
	489	free(eta_secs);
	490
	491	fio_gettime(&now, NULL);
	492	rate_time = mtime_since(&rate_prev_time, &now);
	493
	494	if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
	495	calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
	496	je->rate);
	497	memcpy(&rate_prev_time, &now, sizeof(now));
	498	add_agg_sample(sample_val(je->rate[DDIR_READ]), DDIR_READ, 0);
	499	add_agg_sample(sample_val(je->rate[DDIR_WRITE]), DDIR_WRITE, 0);
	500	add_agg_sample(sample_val(je->rate[DDIR_TRIM]), DDIR_TRIM, 0);
	501	}
	502
	503	disp_time = mtime_since(&disp_prev_time, &now);
	504
	505	if (!force && !eta_time_within_slack(disp_time))
	506	return false;
	507
	508	calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
	509	calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);
	510
	511	memcpy(&disp_prev_time, &now, sizeof(now));
	512
	513	if (!force && !je->nr_running && !je->nr_pending)
	514	return false;
	515
	516	je->nr_threads = thread_number;
	517	update_condensed_str(__run_str, run_str);
	518	memcpy(je->run_str, run_str, strlen(run_str));
	519	return true;
	520	}
	521
	522	void display_thread_status(struct jobs_eta *je)
	523	{
	524	static struct timespec disp_eta_new_line;
	525	static int eta_new_line_init, eta_new_line_pending;
	526	static int linelen_last;
	527	static int eta_good;
	528	char output[__THREAD_RUNSTR_SZ(REAL_MAX_JOBS) + 512], *p = output;
	529	char eta_str[128];
	530	double perc = 0.0;
	531
	532	if (je->eta_sec != INT_MAX && je->elapsed_sec) {
	533	perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
	534	eta_to_str(eta_str, je->eta_sec);
	535	}
	536
	537	if (eta_new_line_pending) {
	538	eta_new_line_pending = 0;
	539	linelen_last = 0;
	540	p += sprintf(p, "\n");
	541	}
	542
	543	p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
	544
	545	/* rate limits, if any */
	546	if (je->m_rate[0] \|\| je->m_rate[1] \|\| je->m_rate[2] \|\|
	547	je->t_rate[0] \|\| je->t_rate[1] \|\| je->t_rate[2]) {
	548	char tr, mr;
	549
	550	mr = num2str(je->m_rate[0] + je->m_rate[1] + je->m_rate[2],
	551	je->sig_figs, 0, je->is_pow2, N2S_BYTEPERSEC);
	552	tr = num2str(je->t_rate[0] + je->t_rate[1] + je->t_rate[2],
	553	je->sig_figs, 0, je->is_pow2, N2S_BYTEPERSEC);
	554
	555	p += sprintf(p, ", %s-%s", mr, tr);
	556	free(tr);
	557	free(mr);
	558	} else if (je->m_iops[0] \|\| je->m_iops[1] \|\| je->m_iops[2] \|\|
	559	je->t_iops[0] \|\| je->t_iops[1] \|\| je->t_iops[2]) {
	560	p += sprintf(p, ", %d-%d IOPS",
	561	je->m_iops[0] + je->m_iops[1] + je->m_iops[2],
	562	je->t_iops[0] + je->t_iops[1] + je->t_iops[2]);
	563	}
	564
	565	/* current run string, % done, bandwidth, iops, eta */
	566	if (je->eta_sec != INT_MAX && je->nr_running) {
	567	char perc_str[32];
	568	char *iops_str[DDIR_RWDIR_CNT];
	569	char *rate_str[DDIR_RWDIR_CNT];
	570	size_t left;
	571	int l;
	572	int ddir;
	573	int linelen;
	574
	575	if ((!je->eta_sec && !eta_good) \|\| je->nr_ramp == je->nr_running \|\|
	576	je->eta_sec == -1)
	577	strcpy(perc_str, "-.-%");
	578	else {
	579	double mult = 100.0;
	580
	581	if (je->nr_setting_up && je->nr_running)
	582	mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);
	583
	584	eta_good = 1;
	585	perc *= mult;
	586	sprintf(perc_str, "%3.1f%%", perc);
	587	}
	588
	589	for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
	590	rate_str[ddir] = num2str(je->rate[ddir], 4,
	591	1024, je->is_pow2, je->unit_base);
	592	iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, N2S_NONE);
	593	}
	594
	595	left = sizeof(output) - (p - output) - 1;
	596
	597	if (je->rate[DDIR_TRIM] \|\| je->iops[DDIR_TRIM])
	598	l = snprintf(p, left,
	599	": [%s][%s][r=%s,w=%s,t=%s][r=%s,w=%s,t=%s IOPS][eta %s]",
	600	je->run_str, perc_str, rate_str[DDIR_READ],
	601	rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
	602	iops_str[DDIR_READ], iops_str[DDIR_WRITE],
	603	iops_str[DDIR_TRIM], eta_str);
	604	else
	605	l = snprintf(p, left,
	606	": [%s][%s][r=%s,w=%s][r=%s,w=%s IOPS][eta %s]",
	607	je->run_str, perc_str,
	608	rate_str[DDIR_READ], rate_str[DDIR_WRITE],
	609	iops_str[DDIR_READ], iops_str[DDIR_WRITE],
	610	eta_str);
	611	/* If truncation occurred adjust l so p is on the null */
	612	if (l >= left)
	613	l = left - 1;
	614	p += l;
	615	linelen = p - output;
	616	if (l >= 0 && linelen < linelen_last)
	617	p += sprintf(p, "%*s", linelen_last - linelen, "");
	618	linelen_last = linelen;
	619
	620	for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
	621	free(rate_str[ddir]);
	622	free(iops_str[ddir]);
	623	}
	624	}
	625	p += sprintf(p, "\r");
	626
	627	printf("%s", output);
	628
	629	if (!eta_new_line_init) {
	630	fio_gettime(&disp_eta_new_line, NULL);
	631	eta_new_line_init = 1;
	632	} else if (eta_new_line && mtime_since_now(&disp_eta_new_line) > eta_new_line) {
	633	fio_gettime(&disp_eta_new_line, NULL);
	634	eta_new_line_pending = 1;
	635	}
	636
	637	fflush(stdout);
	638	}
	639
	640	struct jobs_eta get_jobs_eta(bool force, size_t size)
	641	{
	642	struct jobs_eta *je;
	643
	644	if (!thread_number)
	645	return NULL;
	646
	647	size = sizeof(je) + THREAD_RUNSTR_SZ + 8;
	648	je = malloc(*size);
	649	if (!je)
	650	return NULL;
	651	memset(je, 0, *size);
	652
	653	if (!calc_thread_status(je, force)) {
	654	free(je);
	655	return NULL;
	656	}
	657
	658	size = sizeof(je) + strlen((char *) je->run_str) + 1;
	659	return je;
	660	}
	661
	662	void print_thread_status(void)
	663	{
	664	struct jobs_eta *je;
	665	size_t size;
	666
	667	je = get_jobs_eta(false, &size);
	668	if (je)
	669	display_thread_status(je);
	670
	671	free(je);
	672	}
	673
	674	void print_status_init(int thr_number)
	675	{
	676	DRD_IGNORE_VAR(__run_str);
	677	__run_str[thr_number] = 'P';
	678	update_condensed_str(__run_str, run_str);
	679	}