[fio.git] / eta.c

/*
 * Status and ETA code
 */
#include <unistd.h>
#include <string.h>
#include <stdlib.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 io_size is set, bytes_total is an exact value that does not need
	 * adjustment.
	 */
	if (td->o.zone_size && td->o.zone_skip && bytes_total &&
	    !fio_option_is_set(&td->o, io_size)) {
		unsigned int nr_zones;
		uint64_t zone_bytes;

		/*
		 * Calculate the upper bound of the number of zones that will
		 * be processed, including skipped bytes between zones. If this
		 * is larger than total_io_size (e.g. when --io_size or --size
		 * specify a small value), use the lower bound to avoid
		 * adjustments to a negative value that would result in a very
		 * large bytes_total and an incorrect eta.
		 */
		zone_bytes = td->o.zone_size + td->o.zone_skip;
		nr_zones = (bytes_total + zone_bytes - 1) / zone_bytes;
		if (bytes_total < nr_zones * td->o.zone_skip)
			nr_zones = bytes_total / zone_bytes;
		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 == UNIFIED_MIXED) {
			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 == UNIFIED_MIXED) {
			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)
{
	int unified_rw_rep;
	bool any_td_in_ramp;
	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;

	bw_avg_time = ULONG_MAX;
	unified_rw_rep = 0;
	for_each_td(td) {
		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;
		je->sig_figs = td->o.sig_figs;
		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[__td_index] = thread_eta(td);
		else
			eta_secs[__td_index] = 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];
				}
			}
		}
	} end_for_each();

	if (exitall_on_terminate) {
		je->eta_sec = INT_MAX;
		for_each_td_index() {
			if (eta_secs[__td_index] < je->eta_sec)
				je->eta_sec = eta_secs[__td_index];
		} end_for_each();
	} else {
		unsigned long eta_stone = 0;

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

	free(eta_secs);

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

	any_td_in_ramp = false;
	for_each_td(td) {
		any_td_in_ramp |= in_ramp_time(td);
	} end_for_each();
	if (write_bw_log && rate_time > bw_avg_time && !any_td_in_ramp) {
		calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
				je->rate);
		memcpy(&rate_prev_time, &now, sizeof(now));
		regrow_agg_logs();
		for_each_rw_ddir(ddir) {
			add_agg_sample(sample_val(je->rate[ddir]), ddir, 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;
}

static int gen_eta_str(struct jobs_eta *je, char *p, size_t left,
		       char **rate_str, char **iops_str)
{
	static const char c[DDIR_RWDIR_CNT] = {'r', 'w', 't'};
	bool has[DDIR_RWDIR_CNT];
	bool has_any = false;
	const char *sep;
	int l = 0;

	for_each_rw_ddir(ddir) {
		has[ddir] = (je->rate[ddir] || je->iops[ddir]);
		has_any |= has[ddir];
	}
	if (!has_any)
		return 0;

	l += snprintf(p + l, left - l, "[");
	sep = "";
	for_each_rw_ddir(ddir) {
		if (has[ddir]) {
			l += snprintf(p + l, left - l, "%s%c=%s",
					sep, c[ddir], rate_str[ddir]);
			sep = ",";
		}
	}
	l += snprintf(p + l, left - l, "][");
	sep = "";
	for_each_rw_ddir(ddir) {
		if (has[ddir]) {
			l += snprintf(p + l, left - l, "%s%c=%s",
					sep, c[ddir], iops_str[ddir]);
			sep = ",";
		}
	}
	l += snprintf(p + l, left - l, " IOPS]");

	return l;
}

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, 1, je->is_pow2, N2S_BYTEPERSEC);
		tr = num2str(je->t_rate[0] + je->t_rate[1] + je->t_rate[2],
				je->sig_figs, 1, 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;
		l = snprintf(p, left, ": [%s][%s]", je->run_str, perc_str);
		l += gen_eta_str(je, p + l, left - l, rate_str, iops_str);
		l += snprintf(p + l, left - l, "[eta %s]", 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)
{
	struct jobs_eta_packed jep;

	compiletime_assert(sizeof(struct jobs_eta) == sizeof(jep), "jobs_eta");

	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	#include <stdlib.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	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	/*
	182	* If io_size is set, bytes_total is an exact value that does not need
	183	* adjustment.
	184	*/
	185	if (td->o.zone_size && td->o.zone_skip && bytes_total &&
	186	!fio_option_is_set(&td->o, io_size)) {
	187	unsigned int nr_zones;
	188	uint64_t zone_bytes;
	189
	190	/*
	191	* Calculate the upper bound of the number of zones that will
	192	* be processed, including skipped bytes between zones. If this
	193	* is larger than total_io_size (e.g. when --io_size or --size
	194	* specify a small value), use the lower bound to avoid
	195	* adjustments to a negative value that would result in a very
	196	* large bytes_total and an incorrect eta.
	197	*/
	198	zone_bytes = td->o.zone_size + td->o.zone_skip;
	199	nr_zones = (bytes_total + zone_bytes - 1) / zone_bytes;
	200	if (bytes_total < nr_zones * td->o.zone_skip)
	201	nr_zones = bytes_total / zone_bytes;
	202	bytes_total -= nr_zones * td->o.zone_skip;
	203	}
	204
	205	/*
	206	* if writing and verifying afterwards, bytes_total will be twice the
	207	* size. In a mixed workload, verify phase will be the size of the
	208	* first stage writes.
	209	*/
	210	if (td->o.do_verify && td->o.verify && td_write(td)) {
	211	if (td_rw(td)) {
	212	unsigned int perc = 50;
	213
	214	if (td->o.rwmix[DDIR_WRITE])
	215	perc = td->o.rwmix[DDIR_WRITE];
	216
	217	bytes_total += (bytes_total * perc) / 100;
	218	} else
	219	bytes_total <<= 1;
	220	}
	221
	222	if (td->runstate == TD_RUNNING \|\| td->runstate == TD_VERIFYING) {
	223	double perc, perc_t;
	224
	225	bytes_done = ddir_rw_sum(td->io_bytes);
	226
	227	if (bytes_total) {
	228	perc = (double) bytes_done / (double) bytes_total;
	229	if (perc > 1.0)
	230	perc = 1.0;
	231	} else
	232	perc = 0.0;
	233
	234	if (td->o.time_based) {
	235	if (timeout) {
	236	perc_t = (double) elapsed / (double) timeout;
	237	if (perc_t < perc)
	238	perc = perc_t;
	239	} else {
	240	/*
	241	* Will never hit, we can't have time_based
	242	* without a timeout set.
	243	*/
	244	perc = 0.0;
	245	}
	246	}
	247
	248	if (perc == 0.0) {
	249	eta_sec = timeout;
	250	} else {
	251	eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
	252	}
	253
	254	if (td->o.timeout &&
	255	eta_sec > (timeout + done_secs - elapsed))
	256	eta_sec = timeout + done_secs - elapsed;
	257	} else if (td->runstate == TD_NOT_CREATED \|\| td->runstate == TD_CREATED
	258	\|\| td->runstate == TD_INITIALIZED
	259	\|\| td->runstate == TD_SETTING_UP
	260	\|\| td->runstate == TD_RAMP
	261	\|\| td->runstate == TD_PRE_READING) {
	262	int64_t t_eta = 0, r_eta = 0;
	263	unsigned long long rate_bytes;
	264
	265	/*
	266	* We can only guess - assume it'll run the full timeout
	267	* if given, otherwise assume it'll run at the specified rate.
	268	*/
	269	if (td->o.timeout) {
	270	uint64_t __timeout = td->o.timeout;
	271	uint64_t start_delay = td->o.start_delay;
	272	uint64_t ramp_time = td->o.ramp_time;
	273
	274	t_eta = __timeout + start_delay;
	275	if (!td->ramp_time_over) {
	276	t_eta += ramp_time;
	277	}
	278	t_eta /= 1000000ULL;
	279
	280	if ((td->runstate == TD_RAMP) && in_ramp_time(td)) {
	281	unsigned long ramp_left;
	282
	283	ramp_left = mtime_since_now(&td->epoch);
	284	ramp_left = (ramp_left + 999) / 1000;
	285	if (ramp_left <= t_eta)
	286	t_eta -= ramp_left;
	287	}
	288	}
	289	rate_bytes = 0;
	290	if (td_read(td))
	291	rate_bytes = td->o.rate[DDIR_READ];
	292	if (td_write(td))
	293	rate_bytes += td->o.rate[DDIR_WRITE];
	294	if (td_trim(td))
	295	rate_bytes += td->o.rate[DDIR_TRIM];
	296
	297	if (rate_bytes) {
	298	r_eta = bytes_total / rate_bytes;
	299	r_eta += (td->o.start_delay / 1000000ULL);
	300	}
	301
	302	if (r_eta && t_eta)
	303	eta_sec = min(r_eta, t_eta);
	304	else if (r_eta)
	305	eta_sec = r_eta;
	306	else if (t_eta)
	307	eta_sec = t_eta;
	308	else
	309	eta_sec = 0;
	310	} else {
	311	/*
	312	* thread is already done or waiting for fsync
	313	*/
	314	eta_sec = 0;
	315	}
	316
	317	return eta_sec;
	318	}
	319
	320	static void calc_rate(int unified_rw_rep, unsigned long mtime,
	321	unsigned long long *io_bytes,
	322	unsigned long long prev_io_bytes, uint64_t rate)
	323	{
	324	int i;
	325
	326	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	327	unsigned long long diff, this_rate;
	328
	329	diff = io_bytes[i] - prev_io_bytes[i];
	330	if (mtime)
	331	this_rate = ((1000 * diff) / mtime) / 1024; /* KiB/s */
	332	else
	333	this_rate = 0;
	334
	335	if (unified_rw_rep == UNIFIED_MIXED) {
	336	rate[i] = 0;
	337	rate[0] += this_rate;
	338	} else
	339	rate[i] = this_rate;
	340
	341	prev_io_bytes[i] = io_bytes[i];
	342	}
	343	}
	344
	345	static void calc_iops(int unified_rw_rep, unsigned long mtime,
	346	unsigned long long *io_iops,
	347	unsigned long long prev_io_iops, unsigned int iops)
	348	{
	349	int i;
	350
	351	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	352	unsigned long long diff, this_iops;
	353
	354	diff = io_iops[i] - prev_io_iops[i];
	355	if (mtime)
	356	this_iops = (diff * 1000) / mtime;
	357	else
	358	this_iops = 0;
	359
	360	if (unified_rw_rep == UNIFIED_MIXED) {
	361	iops[i] = 0;
	362	iops[0] += this_iops;
	363	} else
	364	iops[i] = this_iops;
	365
	366	prev_io_iops[i] = io_iops[i];
	367	}
	368	}
	369
	370	/*
	371	* Allow a little slack - if we're within 95% of the time, allow ETA.
	372	*/
	373	bool eta_time_within_slack(unsigned int time)
	374	{
	375	return time > ((eta_interval_msec * 95) / 100);
	376	}
	377
	378	/*
	379	* Print status of the jobs we know about. This includes rate estimates,
	380	* ETA, thread state, etc.
	381	*/
	382	bool calc_thread_status(struct jobs_eta *je, int force)
	383	{
	384	int unified_rw_rep;
	385	bool any_td_in_ramp;
	386	uint64_t rate_time, disp_time, bw_avg_time, *eta_secs;
	387	unsigned long long io_bytes[DDIR_RWDIR_CNT] = {};
	388	unsigned long long io_iops[DDIR_RWDIR_CNT] = {};
	389	struct timespec now;
	390
	391	static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
	392	static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
	393	static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
	394	static struct timespec rate_prev_time, disp_prev_time;
	395
	396	if (!force) {
	397	if (!(output_format & FIO_OUTPUT_NORMAL) &&
	398	f_out == stdout)
	399	return false;
	400	if (temp_stall_ts \|\| eta_print == FIO_ETA_NEVER)
	401	return false;
	402
	403	if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
	404	return false;
	405	}
	406
	407	if (!ddir_rw_sum(rate_io_bytes))
	408	fill_start_time(&rate_prev_time);
	409	if (!ddir_rw_sum(disp_io_bytes))
	410	fill_start_time(&disp_prev_time);
	411
	412	eta_secs = malloc(thread_number * sizeof(uint64_t));
	413	memset(eta_secs, 0, thread_number * sizeof(uint64_t));
	414
	415	je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
	416
	417	bw_avg_time = ULONG_MAX;
	418	unified_rw_rep = 0;
	419	for_each_td(td) {
	420	unified_rw_rep += td->o.unified_rw_rep;
	421	if (is_power_of_2(td->o.kb_base))
	422	je->is_pow2 = 1;
	423	je->unit_base = td->o.unit_base;
	424	je->sig_figs = td->o.sig_figs;
	425	if (td->o.bw_avg_time < bw_avg_time)
	426	bw_avg_time = td->o.bw_avg_time;
	427	if (td->runstate == TD_RUNNING \|\| td->runstate == TD_VERIFYING
	428	\|\| td->runstate == TD_FSYNCING
	429	\|\| td->runstate == TD_PRE_READING
	430	\|\| td->runstate == TD_FINISHING) {
	431	je->nr_running++;
	432	if (td_read(td)) {
	433	je->t_rate[0] += td->o.rate[DDIR_READ];
	434	je->t_iops[0] += td->o.rate_iops[DDIR_READ];
	435	je->m_rate[0] += td->o.ratemin[DDIR_READ];
	436	je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
	437	}
	438	if (td_write(td)) {
	439	je->t_rate[1] += td->o.rate[DDIR_WRITE];
	440	je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
	441	je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
	442	je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
	443	}
	444	if (td_trim(td)) {
	445	je->t_rate[2] += td->o.rate[DDIR_TRIM];
	446	je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
	447	je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
	448	je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
	449	}
	450
	451	je->files_open += td->nr_open_files;
	452	} else if (td->runstate == TD_RAMP) {
	453	je->nr_running++;
	454	je->nr_ramp++;
	455	} else if (td->runstate == TD_SETTING_UP)
	456	je->nr_setting_up++;
	457	else if (td->runstate < TD_RUNNING)
	458	je->nr_pending++;
	459
	460	if (je->elapsed_sec >= 3)
	461	eta_secs[__td_index] = thread_eta(td);
	462	else
	463	eta_secs[__td_index] = INT_MAX;
	464
	465	check_str_update(td);
	466
	467	if (td->runstate > TD_SETTING_UP) {
	468	int ddir;
	469
	470	for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
	471	if (unified_rw_rep) {
	472	io_bytes[0] += td->io_bytes[ddir];
	473	io_iops[0] += td->io_blocks[ddir];
	474	} else {
	475	io_bytes[ddir] += td->io_bytes[ddir];
	476	io_iops[ddir] += td->io_blocks[ddir];
	477	}
	478	}
	479	}
	480	} end_for_each();
	481
	482	if (exitall_on_terminate) {
	483	je->eta_sec = INT_MAX;
	484	for_each_td_index() {
	485	if (eta_secs[__td_index] < je->eta_sec)
	486	je->eta_sec = eta_secs[__td_index];
	487	} end_for_each();
	488	} else {
	489	unsigned long eta_stone = 0;
	490
	491	je->eta_sec = 0;
	492	for_each_td(td) {
	493	if ((td->runstate == TD_NOT_CREATED) && td->o.stonewall)
	494	eta_stone += eta_secs[__td_index];
	495	else {
	496	if (eta_secs[__td_index] > je->eta_sec)
	497	je->eta_sec = eta_secs[__td_index];
	498	}
	499	} end_for_each();
	500	je->eta_sec += eta_stone;
	501	}
	502
	503	free(eta_secs);
	504
	505	fio_gettime(&now, NULL);
	506	rate_time = mtime_since(&rate_prev_time, &now);
	507
	508	any_td_in_ramp = false;
	509	for_each_td(td) {
	510	any_td_in_ramp \|= in_ramp_time(td);
	511	} end_for_each();
	512	if (write_bw_log && rate_time > bw_avg_time && !any_td_in_ramp) {
	513	calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
	514	je->rate);
	515	memcpy(&rate_prev_time, &now, sizeof(now));
	516	regrow_agg_logs();
	517	for_each_rw_ddir(ddir) {
	518	add_agg_sample(sample_val(je->rate[ddir]), ddir, 0);
	519	}
	520	}
	521
	522	disp_time = mtime_since(&disp_prev_time, &now);
	523
	524	if (!force && !eta_time_within_slack(disp_time))
	525	return false;
	526
	527	calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
	528	calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);
	529
	530	memcpy(&disp_prev_time, &now, sizeof(now));
	531
	532	if (!force && !je->nr_running && !je->nr_pending)
	533	return false;
	534
	535	je->nr_threads = thread_number;
	536	update_condensed_str(__run_str, run_str);
	537	memcpy(je->run_str, run_str, strlen(run_str));
	538	return true;
	539	}
	540
	541	static int gen_eta_str(struct jobs_eta je, char p, size_t left,
	542	char rate_str, char iops_str)
	543	{
	544	static const char c[DDIR_RWDIR_CNT] = {'r', 'w', 't'};
	545	bool has[DDIR_RWDIR_CNT];
	546	bool has_any = false;
	547	const char *sep;
	548	int l = 0;
	549
	550	for_each_rw_ddir(ddir) {
	551	has[ddir] = (je->rate[ddir] \|\| je->iops[ddir]);
	552	has_any \|= has[ddir];
	553	}
	554	if (!has_any)
	555	return 0;
	556
	557	l += snprintf(p + l, left - l, "[");
	558	sep = "";
	559	for_each_rw_ddir(ddir) {
	560	if (has[ddir]) {
	561	l += snprintf(p + l, left - l, "%s%c=%s",
	562	sep, c[ddir], rate_str[ddir]);
	563	sep = ",";
	564	}
	565	}
	566	l += snprintf(p + l, left - l, "][");
	567	sep = "";
	568	for_each_rw_ddir(ddir) {
	569	if (has[ddir]) {
	570	l += snprintf(p + l, left - l, "%s%c=%s",
	571	sep, c[ddir], iops_str[ddir]);
	572	sep = ",";
	573	}
	574	}
	575	l += snprintf(p + l, left - l, " IOPS]");
	576
	577	return l;
	578	}
	579
	580	void display_thread_status(struct jobs_eta *je)
	581	{
	582	static struct timespec disp_eta_new_line;
	583	static int eta_new_line_init, eta_new_line_pending;
	584	static int linelen_last;
	585	static int eta_good;
	586	char output[__THREAD_RUNSTR_SZ(REAL_MAX_JOBS) + 512], *p = output;
	587	char eta_str[128];
	588	double perc = 0.0;
	589
	590	if (je->eta_sec != INT_MAX && je->elapsed_sec) {
	591	perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
	592	eta_to_str(eta_str, je->eta_sec);
	593	}
	594
	595	if (eta_new_line_pending) {
	596	eta_new_line_pending = 0;
	597	linelen_last = 0;
	598	p += sprintf(p, "\n");
	599	}
	600
	601	p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
	602
	603	/* rate limits, if any */
	604	if (je->m_rate[0] \|\| je->m_rate[1] \|\| je->m_rate[2] \|\|
	605	je->t_rate[0] \|\| je->t_rate[1] \|\| je->t_rate[2]) {
	606	char tr, mr;
	607
	608	mr = num2str(je->m_rate[0] + je->m_rate[1] + je->m_rate[2],
	609	je->sig_figs, 1, je->is_pow2, N2S_BYTEPERSEC);
	610	tr = num2str(je->t_rate[0] + je->t_rate[1] + je->t_rate[2],
	611	je->sig_figs, 1, je->is_pow2, N2S_BYTEPERSEC);
	612
	613	p += sprintf(p, ", %s-%s", mr, tr);
	614	free(tr);
	615	free(mr);
	616	} else if (je->m_iops[0] \|\| je->m_iops[1] \|\| je->m_iops[2] \|\|
	617	je->t_iops[0] \|\| je->t_iops[1] \|\| je->t_iops[2]) {
	618	p += sprintf(p, ", %d-%d IOPS",
	619	je->m_iops[0] + je->m_iops[1] + je->m_iops[2],
	620	je->t_iops[0] + je->t_iops[1] + je->t_iops[2]);
	621	}
	622
	623	/* current run string, % done, bandwidth, iops, eta */
	624	if (je->eta_sec != INT_MAX && je->nr_running) {
	625	char perc_str[32];
	626	char *iops_str[DDIR_RWDIR_CNT];
	627	char *rate_str[DDIR_RWDIR_CNT];
	628	size_t left;
	629	int l;
	630	int ddir;
	631	int linelen;
	632
	633	if ((!je->eta_sec && !eta_good) \|\| je->nr_ramp == je->nr_running \|\|
	634	je->eta_sec == -1)
	635	strcpy(perc_str, "-.-%");
	636	else {
	637	double mult = 100.0;
	638
	639	if (je->nr_setting_up && je->nr_running)
	640	mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);
	641
	642	eta_good = 1;
	643	perc *= mult;
	644	sprintf(perc_str, "%3.1f%%", perc);
	645	}
	646
	647	for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
	648	rate_str[ddir] = num2str(je->rate[ddir], 4,
	649	1024, je->is_pow2, je->unit_base);
	650	iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, N2S_NONE);
	651	}
	652
	653	left = sizeof(output) - (p - output) - 1;
	654	l = snprintf(p, left, ": [%s][%s]", je->run_str, perc_str);
	655	l += gen_eta_str(je, p + l, left - l, rate_str, iops_str);
	656	l += snprintf(p + l, left - l, "[eta %s]", eta_str);
	657
	658	/* If truncation occurred adjust l so p is on the null */
	659	if (l >= left)
	660	l = left - 1;
	661	p += l;
	662	linelen = p - output;
	663	if (l >= 0 && linelen < linelen_last)
	664	p += sprintf(p, "%*s", linelen_last - linelen, "");
	665	linelen_last = linelen;
	666
	667	for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
	668	free(rate_str[ddir]);
	669	free(iops_str[ddir]);
	670	}
	671	}
	672	sprintf(p, "\r");
	673
	674	printf("%s", output);
	675
	676	if (!eta_new_line_init) {
	677	fio_gettime(&disp_eta_new_line, NULL);
	678	eta_new_line_init = 1;
	679	} else if (eta_new_line && mtime_since_now(&disp_eta_new_line) > eta_new_line) {
	680	fio_gettime(&disp_eta_new_line, NULL);
	681	eta_new_line_pending = 1;
	682	}
	683
	684	fflush(stdout);
	685	}
	686
	687	struct jobs_eta get_jobs_eta(bool force, size_t size)
	688	{
	689	struct jobs_eta *je;
	690
	691	if (!thread_number)
	692	return NULL;
	693
	694	size = sizeof(je) + THREAD_RUNSTR_SZ + 8;
	695	je = malloc(*size);
	696	if (!je)
	697	return NULL;
	698	memset(je, 0, *size);
	699
	700	if (!calc_thread_status(je, force)) {
	701	free(je);
	702	return NULL;
	703	}
	704
	705	size = sizeof(je) + strlen((char *) je->run_str) + 1;
	706	return je;
	707	}
	708
	709	void print_thread_status(void)
	710	{
	711	struct jobs_eta *je;
	712	size_t size;
	713
	714	je = get_jobs_eta(false, &size);
	715	if (je) {
	716	display_thread_status(je);
	717	free(je);
	718	}
	719	}
	720
	721	void print_status_init(int thr_number)
	722	{
	723	struct jobs_eta_packed jep;
	724
	725	compiletime_assert(sizeof(struct jobs_eta) == sizeof(jep), "jobs_eta");
	726
	727	DRD_IGNORE_VAR(__run_str);
	728	__run_str[thr_number] = 'P';
	729	update_condensed_str(__run_str, run_str);
	730	}