[fio.git] / eta.c

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