xref: /freebsd/sys/contrib/openzfs/cmd/zpool_influxdb/zpool_influxdb.c (revision 02e9120893770924227138ba49df1edb3896112a)
1 /*
2  * Gather top-level ZFS pool and resilver/scan statistics and print using
3  * influxdb line protocol
4  * usage: [options] [pool_name]
5  * where options are:
6  *   --execd, -e           run in telegraf execd input plugin mode, [CR] on
7  *                         stdin causes a sample to be printed and wait for
8  *                         the next [CR]
9  *   --no-histograms, -n   don't print histogram data (reduces cardinality
10  *                         if you don't care about histograms)
11  *   --sum-histogram-buckets, -s sum histogram bucket values
12  *
13  * To integrate into telegraf use one of:
14  * 1. the `inputs.execd` plugin with the `--execd` option
15  * 2. the `inputs.exec` plugin to simply run with no options
16  *
17  * NOTE: libzfs is an unstable interface. YMMV.
18  *
19  * The design goals of this software include:
20  * + be as lightweight as possible
21  * + reduce the number of external dependencies as far as possible, hence
22  *   there is no dependency on a client library for managing the metric
23  *   collection -- info is printed, KISS
24  * + broken pools or kernel bugs can cause this process to hang in an
25  *   unkillable state. For this reason, it is best to keep the damage limited
26  *   to a small process like zpool_influxdb rather than a larger collector.
27  *
28  * Copyright 2018-2020 Richard Elling
29  *
30  * This software is dual-licensed MIT and CDDL.
31  *
32  * The MIT License (MIT)
33  *
34  * Permission is hereby granted, free of charge, to any person obtaining a copy
35  * of this software and associated documentation files (the "Software"), to deal
36  * in the Software without restriction, including without limitation the rights
37  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
38  * copies of the Software, and to permit persons to whom the Software is
39  * furnished to do so, subject to the following conditions:
40  *
41  * The above copyright notice and this permission notice shall be included in
42  * all copies or substantial portions of the Software.
43  *
44  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
45  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
46  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
47  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
48  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
49  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
50  * SOFTWARE.
51  *
52  * CDDL HEADER START
53  *
54  * The contents of this file are subject to the terms of the
55  * Common Development and Distribution License (the "License").
56  * You may not use this file except in compliance with the License.
57  *
58  * The contents of this file are subject to the terms of the
59  * Common Development and Distribution License Version 1.0 (CDDL-1.0).
60  * You can obtain a copy of the license from the top-level file
61  * "OPENSOLARIS.LICENSE" or at <http://opensource.org/licenses/CDDL-1.0>.
62  * You may not use this file except in compliance with the license.
63  *
64  * See the License for the specific language governing permissions
65  * and limitations under the License.
66  *
67  * CDDL HEADER END
68  */
69 #include <string.h>
70 #include <getopt.h>
71 #include <stdio.h>
72 #include <stdint.h>
73 #include <inttypes.h>
74 #include <libzfs.h>
75 
76 #define	POOL_MEASUREMENT	"zpool_stats"
77 #define	SCAN_MEASUREMENT	"zpool_scan_stats"
78 #define	VDEV_MEASUREMENT	"zpool_vdev_stats"
79 #define	POOL_LATENCY_MEASUREMENT	"zpool_latency"
80 #define	POOL_QUEUE_MEASUREMENT	"zpool_vdev_queue"
81 #define	MIN_LAT_INDEX	10  /* minimum latency index 10 = 1024ns */
82 #define	POOL_IO_SIZE_MEASUREMENT	"zpool_io_size"
83 #define	MIN_SIZE_INDEX	9  /* minimum size index 9 = 512 bytes */
84 
85 /* global options */
86 int execd_mode = 0;
87 int no_histograms = 0;
88 int sum_histogram_buckets = 0;
89 char metric_data_type = 'u';
90 uint64_t metric_value_mask = UINT64_MAX;
91 uint64_t timestamp = 0;
92 int complained_about_sync = 0;
93 const char *tags = "";
94 
95 typedef int (*stat_printer_f)(nvlist_t *, const char *, const char *);
96 
97 /*
98  * influxdb line protocol rules for escaping are important because the
99  * zpool name can include characters that need to be escaped
100  *
101  * caller is responsible for freeing result
102  */
103 static char *
104 escape_string(const char *s)
105 {
106 	const char *c;
107 	char *d;
108 	char *t = (char *)malloc(ZFS_MAX_DATASET_NAME_LEN * 2);
109 	if (t == NULL) {
110 		fprintf(stderr, "error: cannot allocate memory\n");
111 		exit(1);
112 	}
113 
114 	for (c = s, d = t; *c != '\0'; c++, d++) {
115 		switch (*c) {
116 		case ' ':
117 		case ',':
118 		case '=':
119 		case '\\':
120 			*d++ = '\\';
121 			zfs_fallthrough;
122 		default:
123 			*d = *c;
124 		}
125 	}
126 	*d = '\0';
127 	return (t);
128 }
129 
130 /*
131  * print key=value where value is a uint64_t
132  */
133 static void
134 print_kv(const char *key, uint64_t value)
135 {
136 	printf("%s=%llu%c", key,
137 	    (u_longlong_t)value & metric_value_mask, metric_data_type);
138 }
139 
140 /*
141  * print_scan_status() prints the details as often seen in the "zpool status"
142  * output. However, unlike the zpool command, which is intended for humans,
143  * this output is suitable for long-term tracking in influxdb.
144  * TODO: update to include issued scan data
145  */
146 static int
147 print_scan_status(nvlist_t *nvroot, const char *pool_name)
148 {
149 	uint_t c;
150 	int64_t elapsed;
151 	uint64_t examined, pass_exam, paused_time, paused_ts, rate;
152 	uint64_t remaining_time;
153 	pool_scan_stat_t *ps = NULL;
154 	double pct_done;
155 	const char *const state[DSS_NUM_STATES] = {
156 	    "none", "scanning", "finished", "canceled"};
157 	const char *func;
158 
159 	(void) nvlist_lookup_uint64_array(nvroot,
160 	    ZPOOL_CONFIG_SCAN_STATS,
161 	    (uint64_t **)&ps, &c);
162 
163 	/*
164 	 * ignore if there are no stats
165 	 */
166 	if (ps == NULL)
167 		return (0);
168 
169 	/*
170 	 * return error if state is bogus
171 	 */
172 	if (ps->pss_state >= DSS_NUM_STATES ||
173 	    ps->pss_func >= POOL_SCAN_FUNCS) {
174 		if (complained_about_sync % 1000 == 0) {
175 			fprintf(stderr, "error: cannot decode scan stats: "
176 			    "ZFS is out of sync with compiled zpool_influxdb");
177 			complained_about_sync++;
178 		}
179 		return (1);
180 	}
181 
182 	switch (ps->pss_func) {
183 	case POOL_SCAN_NONE:
184 		func = "none_requested";
185 		break;
186 	case POOL_SCAN_SCRUB:
187 		func = "scrub";
188 		break;
189 	case POOL_SCAN_RESILVER:
190 		func = "resilver";
191 		break;
192 #ifdef POOL_SCAN_REBUILD
193 	case POOL_SCAN_REBUILD:
194 		func = "rebuild";
195 		break;
196 #endif
197 	default:
198 		func = "scan";
199 	}
200 
201 	/* overall progress */
202 	examined = ps->pss_examined ? ps->pss_examined : 1;
203 	pct_done = 0.0;
204 	if (ps->pss_to_examine > 0)
205 		pct_done = 100.0 * examined / ps->pss_to_examine;
206 
207 #ifdef EZFS_SCRUB_PAUSED
208 	paused_ts = ps->pss_pass_scrub_pause;
209 	paused_time = ps->pss_pass_scrub_spent_paused;
210 #else
211 	paused_ts = 0;
212 	paused_time = 0;
213 #endif
214 
215 	/* calculations for this pass */
216 	if (ps->pss_state == DSS_SCANNING) {
217 		elapsed = (int64_t)time(NULL) - (int64_t)ps->pss_pass_start -
218 		    (int64_t)paused_time;
219 		elapsed = (elapsed > 0) ? elapsed : 1;
220 		pass_exam = ps->pss_pass_exam ? ps->pss_pass_exam : 1;
221 		rate = pass_exam / elapsed;
222 		rate = (rate > 0) ? rate : 1;
223 		remaining_time = ps->pss_to_examine - examined / rate;
224 	} else {
225 		elapsed =
226 		    (int64_t)ps->pss_end_time - (int64_t)ps->pss_pass_start -
227 		    (int64_t)paused_time;
228 		elapsed = (elapsed > 0) ? elapsed : 1;
229 		pass_exam = ps->pss_pass_exam ? ps->pss_pass_exam : 1;
230 		rate = pass_exam / elapsed;
231 		remaining_time = 0;
232 	}
233 	rate = rate ? rate : 1;
234 
235 	/* influxdb line protocol format: "tags metrics timestamp" */
236 	printf("%s%s,function=%s,name=%s,state=%s ",
237 	    SCAN_MEASUREMENT, tags, func, pool_name, state[ps->pss_state]);
238 	print_kv("end_ts", ps->pss_end_time);
239 	print_kv(",errors", ps->pss_errors);
240 	print_kv(",examined", examined);
241 	print_kv(",skipped", ps->pss_skipped);
242 	print_kv(",issued", ps->pss_issued);
243 	print_kv(",pass_examined", pass_exam);
244 	print_kv(",pass_issued", ps->pss_pass_issued);
245 	print_kv(",paused_ts", paused_ts);
246 	print_kv(",paused_t", paused_time);
247 	printf(",pct_done=%.2f", pct_done);
248 	print_kv(",processed", ps->pss_processed);
249 	print_kv(",rate", rate);
250 	print_kv(",remaining_t", remaining_time);
251 	print_kv(",start_ts", ps->pss_start_time);
252 	print_kv(",to_examine", ps->pss_to_examine);
253 	printf(" %llu\n", (u_longlong_t)timestamp);
254 	return (0);
255 }
256 
257 /*
258  * get a vdev name that corresponds to the top-level vdev names
259  * printed by `zpool status`
260  */
261 static char *
262 get_vdev_name(nvlist_t *nvroot, const char *parent_name)
263 {
264 	static char vdev_name[256];
265 	uint64_t vdev_id = 0;
266 
267 	const char *vdev_type = "unknown";
268 	(void) nvlist_lookup_string(nvroot, ZPOOL_CONFIG_TYPE, &vdev_type);
269 
270 	if (nvlist_lookup_uint64(
271 	    nvroot, ZPOOL_CONFIG_ID, &vdev_id) != 0)
272 		vdev_id = UINT64_MAX;
273 
274 	if (parent_name == NULL) {
275 		(void) snprintf(vdev_name, sizeof (vdev_name), "%s",
276 		    vdev_type);
277 	} else {
278 		(void) snprintf(vdev_name, sizeof (vdev_name),
279 		    "%.220s/%s-%llu",
280 		    parent_name, vdev_type, (u_longlong_t)vdev_id);
281 	}
282 	return (vdev_name);
283 }
284 
285 /*
286  * get a string suitable for an influxdb tag that describes this vdev
287  *
288  * By default only the vdev hierarchical name is shown, separated by '/'
289  * If the vdev has an associated path, which is typical of leaf vdevs,
290  * then the path is added.
291  * It would be nice to have the devid instead of the path, but under
292  * Linux we cannot be sure a devid will exist and we'd rather have
293  * something than nothing, so we'll use path instead.
294  */
295 static char *
296 get_vdev_desc(nvlist_t *nvroot, const char *parent_name)
297 {
298 	static char vdev_desc[2 * MAXPATHLEN];
299 	char vdev_value[MAXPATHLEN];
300 	char *s, *t;
301 
302 	const char *vdev_type = "unknown";
303 	uint64_t vdev_id = UINT64_MAX;
304 	const char *vdev_path = NULL;
305 	(void) nvlist_lookup_string(nvroot, ZPOOL_CONFIG_TYPE, &vdev_type);
306 	(void) nvlist_lookup_uint64(nvroot, ZPOOL_CONFIG_ID, &vdev_id);
307 	(void) nvlist_lookup_string(nvroot, ZPOOL_CONFIG_PATH, &vdev_path);
308 
309 	if (parent_name == NULL) {
310 		s = escape_string(vdev_type);
311 		(void) snprintf(vdev_value, sizeof (vdev_value), "vdev=%s", s);
312 		free(s);
313 	} else {
314 		s = escape_string((char *)parent_name);
315 		t = escape_string(vdev_type);
316 		(void) snprintf(vdev_value, sizeof (vdev_value),
317 		    "vdev=%s/%s-%llu", s, t, (u_longlong_t)vdev_id);
318 		free(s);
319 		free(t);
320 	}
321 	if (vdev_path == NULL) {
322 		(void) snprintf(vdev_desc, sizeof (vdev_desc), "%s",
323 		    vdev_value);
324 	} else {
325 		s = escape_string(vdev_path);
326 		(void) snprintf(vdev_desc, sizeof (vdev_desc), "path=%s,%s",
327 		    s, vdev_value);
328 		free(s);
329 	}
330 	return (vdev_desc);
331 }
332 
333 /*
334  * vdev summary stats are a combination of the data shown by
335  * `zpool status` and `zpool list -v`
336  */
337 static int
338 print_summary_stats(nvlist_t *nvroot, const char *pool_name,
339     const char *parent_name)
340 {
341 	uint_t c;
342 	vdev_stat_t *vs;
343 	char *vdev_desc = NULL;
344 	vdev_desc = get_vdev_desc(nvroot, parent_name);
345 	if (nvlist_lookup_uint64_array(nvroot, ZPOOL_CONFIG_VDEV_STATS,
346 	    (uint64_t **)&vs, &c) != 0) {
347 		return (1);
348 	}
349 	printf("%s%s,name=%s,state=%s,%s ", POOL_MEASUREMENT, tags,
350 	    pool_name, zpool_state_to_name((vdev_state_t)vs->vs_state,
351 	    (vdev_aux_t)vs->vs_aux), vdev_desc);
352 	print_kv("alloc", vs->vs_alloc);
353 	print_kv(",free", vs->vs_space - vs->vs_alloc);
354 	print_kv(",size", vs->vs_space);
355 	print_kv(",read_bytes", vs->vs_bytes[ZIO_TYPE_READ]);
356 	print_kv(",read_errors", vs->vs_read_errors);
357 	print_kv(",read_ops", vs->vs_ops[ZIO_TYPE_READ]);
358 	print_kv(",write_bytes", vs->vs_bytes[ZIO_TYPE_WRITE]);
359 	print_kv(",write_errors", vs->vs_write_errors);
360 	print_kv(",write_ops", vs->vs_ops[ZIO_TYPE_WRITE]);
361 	print_kv(",checksum_errors", vs->vs_checksum_errors);
362 	print_kv(",fragmentation", vs->vs_fragmentation);
363 	printf(" %llu\n", (u_longlong_t)timestamp);
364 	return (0);
365 }
366 
367 /*
368  * vdev latency stats are histograms stored as nvlist arrays of uint64.
369  * Latency stats include the ZIO scheduler classes plus lower-level
370  * vdev latencies.
371  *
372  * In many cases, the top-level "root" view obscures the underlying
373  * top-level vdev operations. For example, if a pool has a log, special,
374  * or cache device, then each can behave very differently. It is useful
375  * to see how each is responding.
376  */
377 static int
378 print_vdev_latency_stats(nvlist_t *nvroot, const char *pool_name,
379     const char *parent_name)
380 {
381 	uint_t c, end = 0;
382 	nvlist_t *nv_ex;
383 	char *vdev_desc = NULL;
384 
385 	/* short_names become part of the metric name and are influxdb-ready */
386 	struct lat_lookup {
387 	    const char *name;
388 	    const char *short_name;
389 	    uint64_t sum;
390 	    uint64_t *array;
391 	};
392 	struct lat_lookup lat_type[] = {
393 	    {ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO,   "total_read", 0},
394 	    {ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO,   "total_write", 0},
395 	    {ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO,  "disk_read", 0},
396 	    {ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO,  "disk_write", 0},
397 	    {ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO,  "sync_read", 0},
398 	    {ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO,  "sync_write", 0},
399 	    {ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO, "async_read", 0},
400 	    {ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO, "async_write", 0},
401 	    {ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO,   "scrub", 0},
402 #ifdef ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO
403 	    {ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO,    "trim", 0},
404 #endif
405 	    {ZPOOL_CONFIG_VDEV_REBUILD_LAT_HISTO,    "rebuild", 0},
406 	    {NULL,	NULL}
407 	};
408 
409 	if (nvlist_lookup_nvlist(nvroot,
410 	    ZPOOL_CONFIG_VDEV_STATS_EX, &nv_ex) != 0) {
411 		return (6);
412 	}
413 
414 	vdev_desc = get_vdev_desc(nvroot, parent_name);
415 
416 	for (int i = 0; lat_type[i].name; i++) {
417 		if (nvlist_lookup_uint64_array(nv_ex,
418 		    lat_type[i].name, &lat_type[i].array, &c) != 0) {
419 			fprintf(stderr, "error: can't get %s\n",
420 			    lat_type[i].name);
421 			return (3);
422 		}
423 		/* end count count, all of the arrays are the same size */
424 		end = c - 1;
425 	}
426 
427 	for (int bucket = 0; bucket <= end; bucket++) {
428 		if (bucket < MIN_LAT_INDEX) {
429 			/* don't print, but collect the sum */
430 			for (int i = 0; lat_type[i].name; i++) {
431 				lat_type[i].sum += lat_type[i].array[bucket];
432 			}
433 			continue;
434 		}
435 		if (bucket < end) {
436 			printf("%s%s,le=%0.6f,name=%s,%s ",
437 			    POOL_LATENCY_MEASUREMENT, tags,
438 			    (float)(1ULL << bucket) * 1e-9,
439 			    pool_name, vdev_desc);
440 		} else {
441 			printf("%s%s,le=+Inf,name=%s,%s ",
442 			    POOL_LATENCY_MEASUREMENT, tags, pool_name,
443 			    vdev_desc);
444 		}
445 		for (int i = 0; lat_type[i].name; i++) {
446 			if (bucket <= MIN_LAT_INDEX || sum_histogram_buckets) {
447 				lat_type[i].sum += lat_type[i].array[bucket];
448 			} else {
449 				lat_type[i].sum = lat_type[i].array[bucket];
450 			}
451 			print_kv(lat_type[i].short_name, lat_type[i].sum);
452 			if (lat_type[i + 1].name != NULL) {
453 				printf(",");
454 			}
455 		}
456 		printf(" %llu\n", (u_longlong_t)timestamp);
457 	}
458 	return (0);
459 }
460 
461 /*
462  * vdev request size stats are histograms stored as nvlist arrays of uint64.
463  * Request size stats include the ZIO scheduler classes plus lower-level
464  * vdev sizes. Both independent (ind) and aggregated (agg) sizes are reported.
465  *
466  * In many cases, the top-level "root" view obscures the underlying
467  * top-level vdev operations. For example, if a pool has a log, special,
468  * or cache device, then each can behave very differently. It is useful
469  * to see how each is responding.
470  */
471 static int
472 print_vdev_size_stats(nvlist_t *nvroot, const char *pool_name,
473     const char *parent_name)
474 {
475 	uint_t c, end = 0;
476 	nvlist_t *nv_ex;
477 	char *vdev_desc = NULL;
478 
479 	/* short_names become the field name */
480 	struct size_lookup {
481 	    const char *name;
482 	    const char *short_name;
483 	    uint64_t sum;
484 	    uint64_t *array;
485 	};
486 	struct size_lookup size_type[] = {
487 	    {ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO,   "sync_read_ind"},
488 	    {ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO,   "sync_write_ind"},
489 	    {ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO,  "async_read_ind"},
490 	    {ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO,  "async_write_ind"},
491 	    {ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO,    "scrub_read_ind"},
492 	    {ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO,   "sync_read_agg"},
493 	    {ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO,   "sync_write_agg"},
494 	    {ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO,  "async_read_agg"},
495 	    {ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO,  "async_write_agg"},
496 	    {ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO,    "scrub_read_agg"},
497 #ifdef ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO
498 	    {ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO,    "trim_write_ind"},
499 	    {ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO,    "trim_write_agg"},
500 #endif
501 	    {ZPOOL_CONFIG_VDEV_IND_REBUILD_HISTO,    "rebuild_write_ind"},
502 	    {ZPOOL_CONFIG_VDEV_AGG_REBUILD_HISTO,    "rebuild_write_agg"},
503 	    {NULL,	NULL}
504 	};
505 
506 	if (nvlist_lookup_nvlist(nvroot,
507 	    ZPOOL_CONFIG_VDEV_STATS_EX, &nv_ex) != 0) {
508 		return (6);
509 	}
510 
511 	vdev_desc = get_vdev_desc(nvroot, parent_name);
512 
513 	for (int i = 0; size_type[i].name; i++) {
514 		if (nvlist_lookup_uint64_array(nv_ex, size_type[i].name,
515 		    &size_type[i].array, &c) != 0) {
516 			fprintf(stderr, "error: can't get %s\n",
517 			    size_type[i].name);
518 			return (3);
519 		}
520 		/* end count count, all of the arrays are the same size */
521 		end = c - 1;
522 	}
523 
524 	for (int bucket = 0; bucket <= end; bucket++) {
525 		if (bucket < MIN_SIZE_INDEX) {
526 			/* don't print, but collect the sum */
527 			for (int i = 0; size_type[i].name; i++) {
528 				size_type[i].sum += size_type[i].array[bucket];
529 			}
530 			continue;
531 		}
532 
533 		if (bucket < end) {
534 			printf("%s%s,le=%llu,name=%s,%s ",
535 			    POOL_IO_SIZE_MEASUREMENT, tags, 1ULL << bucket,
536 			    pool_name, vdev_desc);
537 		} else {
538 			printf("%s%s,le=+Inf,name=%s,%s ",
539 			    POOL_IO_SIZE_MEASUREMENT, tags, pool_name,
540 			    vdev_desc);
541 		}
542 		for (int i = 0; size_type[i].name; i++) {
543 			if (bucket <= MIN_SIZE_INDEX || sum_histogram_buckets) {
544 				size_type[i].sum += size_type[i].array[bucket];
545 			} else {
546 				size_type[i].sum = size_type[i].array[bucket];
547 			}
548 			print_kv(size_type[i].short_name, size_type[i].sum);
549 			if (size_type[i + 1].name != NULL) {
550 				printf(",");
551 			}
552 		}
553 		printf(" %llu\n", (u_longlong_t)timestamp);
554 	}
555 	return (0);
556 }
557 
558 /*
559  * ZIO scheduler queue stats are stored as gauges. This is unfortunate
560  * because the values can change very rapidly and any point-in-time
561  * value will quickly be obsoleted. It is also not easy to downsample.
562  * Thus only the top-level queue stats might be beneficial... maybe.
563  */
564 static int
565 print_queue_stats(nvlist_t *nvroot, const char *pool_name,
566     const char *parent_name)
567 {
568 	nvlist_t *nv_ex;
569 	uint64_t value;
570 
571 	/* short_names are used for the field name */
572 	struct queue_lookup {
573 	    const char *name;
574 	    const char *short_name;
575 	};
576 	struct queue_lookup queue_type[] = {
577 	    {ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE,	"sync_r_active"},
578 	    {ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE,	"sync_w_active"},
579 	    {ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE,	"async_r_active"},
580 	    {ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE,	"async_w_active"},
581 	    {ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE,	"async_scrub_active"},
582 	    {ZPOOL_CONFIG_VDEV_REBUILD_ACTIVE_QUEUE,	"rebuild_active"},
583 	    {ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE,	"sync_r_pend"},
584 	    {ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE,	"sync_w_pend"},
585 	    {ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE,	"async_r_pend"},
586 	    {ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE,	"async_w_pend"},
587 	    {ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE,	"async_scrub_pend"},
588 	    {ZPOOL_CONFIG_VDEV_REBUILD_PEND_QUEUE,	"rebuild_pend"},
589 	    {NULL,	NULL}
590 	};
591 
592 	if (nvlist_lookup_nvlist(nvroot,
593 	    ZPOOL_CONFIG_VDEV_STATS_EX, &nv_ex) != 0) {
594 		return (6);
595 	}
596 
597 	printf("%s%s,name=%s,%s ", POOL_QUEUE_MEASUREMENT, tags, pool_name,
598 	    get_vdev_desc(nvroot, parent_name));
599 	for (int i = 0; queue_type[i].name; i++) {
600 		if (nvlist_lookup_uint64(nv_ex,
601 		    queue_type[i].name, &value) != 0) {
602 			fprintf(stderr, "error: can't get %s\n",
603 			    queue_type[i].name);
604 			return (3);
605 		}
606 		print_kv(queue_type[i].short_name, value);
607 		if (queue_type[i + 1].name != NULL) {
608 			printf(",");
609 		}
610 	}
611 	printf(" %llu\n", (u_longlong_t)timestamp);
612 	return (0);
613 }
614 
615 /*
616  * top-level vdev stats are at the pool level
617  */
618 static int
619 print_top_level_vdev_stats(nvlist_t *nvroot, const char *pool_name)
620 {
621 	nvlist_t *nv_ex;
622 	uint64_t value;
623 
624 	/* short_names become part of the metric name */
625 	struct queue_lookup {
626 	    const char *name;
627 	    const char *short_name;
628 	};
629 	struct queue_lookup queue_type[] = {
630 	    {ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE, "sync_r_active_queue"},
631 	    {ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE, "sync_w_active_queue"},
632 	    {ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE, "async_r_active_queue"},
633 	    {ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE, "async_w_active_queue"},
634 	    {ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE, "async_scrub_active_queue"},
635 	    {ZPOOL_CONFIG_VDEV_REBUILD_ACTIVE_QUEUE, "rebuild_active_queue"},
636 	    {ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE, "sync_r_pend_queue"},
637 	    {ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE, "sync_w_pend_queue"},
638 	    {ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE, "async_r_pend_queue"},
639 	    {ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE, "async_w_pend_queue"},
640 	    {ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE, "async_scrub_pend_queue"},
641 	    {ZPOOL_CONFIG_VDEV_REBUILD_PEND_QUEUE, "rebuild_pend_queue"},
642 	    {NULL, NULL}
643 	};
644 
645 	if (nvlist_lookup_nvlist(nvroot,
646 	    ZPOOL_CONFIG_VDEV_STATS_EX, &nv_ex) != 0) {
647 		return (6);
648 	}
649 
650 	printf("%s%s,name=%s,vdev=root ", VDEV_MEASUREMENT, tags,
651 	    pool_name);
652 	for (int i = 0; queue_type[i].name; i++) {
653 		if (nvlist_lookup_uint64(nv_ex,
654 		    queue_type[i].name, &value) != 0) {
655 			fprintf(stderr, "error: can't get %s\n",
656 			    queue_type[i].name);
657 			return (3);
658 		}
659 		if (i > 0)
660 			printf(",");
661 		print_kv(queue_type[i].short_name, value);
662 	}
663 
664 	printf(" %llu\n", (u_longlong_t)timestamp);
665 	return (0);
666 }
667 
668 /*
669  * recursive stats printer
670  */
671 static int
672 print_recursive_stats(stat_printer_f func, nvlist_t *nvroot,
673     const char *pool_name, const char *parent_name, int descend)
674 {
675 	uint_t c, children;
676 	nvlist_t **child;
677 	char vdev_name[256];
678 	int err;
679 
680 	err = func(nvroot, pool_name, parent_name);
681 	if (err)
682 		return (err);
683 
684 	if (descend && nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
685 	    &child, &children) == 0) {
686 		(void) strlcpy(vdev_name, get_vdev_name(nvroot, parent_name),
687 		    sizeof (vdev_name));
688 
689 		for (c = 0; c < children; c++) {
690 			err = print_recursive_stats(func, child[c], pool_name,
691 			    vdev_name, descend);
692 			if (err)
693 				return (err);
694 		}
695 	}
696 	return (0);
697 }
698 
699 /*
700  * call-back to print the stats from the pool config
701  *
702  * Note: if the pool is broken, this can hang indefinitely and perhaps in an
703  * unkillable state.
704  */
705 static int
706 print_stats(zpool_handle_t *zhp, void *data)
707 {
708 	uint_t c;
709 	int err;
710 	boolean_t missing;
711 	nvlist_t *config, *nvroot;
712 	vdev_stat_t *vs;
713 	struct timespec tv;
714 	char *pool_name;
715 
716 	/* if not this pool return quickly */
717 	if (data &&
718 	    strncmp(data, zpool_get_name(zhp), ZFS_MAX_DATASET_NAME_LEN) != 0) {
719 		zpool_close(zhp);
720 		return (0);
721 	}
722 
723 	if (zpool_refresh_stats(zhp, &missing) != 0) {
724 		zpool_close(zhp);
725 		return (1);
726 	}
727 
728 	config = zpool_get_config(zhp, NULL);
729 	if (clock_gettime(CLOCK_REALTIME, &tv) != 0)
730 		timestamp = (uint64_t)time(NULL) * 1000000000;
731 	else
732 		timestamp =
733 		    ((uint64_t)tv.tv_sec * 1000000000) + (uint64_t)tv.tv_nsec;
734 
735 	if (nvlist_lookup_nvlist(
736 	    config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) != 0) {
737 	zpool_close(zhp);
738 		return (2);
739 	}
740 	if (nvlist_lookup_uint64_array(nvroot, ZPOOL_CONFIG_VDEV_STATS,
741 	    (uint64_t **)&vs, &c) != 0) {
742 	zpool_close(zhp);
743 		return (3);
744 	}
745 
746 	pool_name = escape_string(zpool_get_name(zhp));
747 	err = print_recursive_stats(print_summary_stats, nvroot,
748 	    pool_name, NULL, 1);
749 	/* if any of these return an error, skip the rest */
750 	if (err == 0)
751 	err = print_top_level_vdev_stats(nvroot, pool_name);
752 
753 	if (no_histograms == 0) {
754 	if (err == 0)
755 		err = print_recursive_stats(print_vdev_latency_stats, nvroot,
756 		    pool_name, NULL, 1);
757 	if (err == 0)
758 		err = print_recursive_stats(print_vdev_size_stats, nvroot,
759 		    pool_name, NULL, 1);
760 	if (err == 0)
761 		err = print_recursive_stats(print_queue_stats, nvroot,
762 		    pool_name, NULL, 0);
763 	}
764 	if (err == 0)
765 		err = print_scan_status(nvroot, pool_name);
766 
767 	free(pool_name);
768 	zpool_close(zhp);
769 	return (err);
770 }
771 
772 static void
773 usage(char *name)
774 {
775 	fprintf(stderr, "usage: %s [--execd][--no-histograms]"
776 	    "[--sum-histogram-buckets] [--signed-int] [poolname]\n", name);
777 	exit(EXIT_FAILURE);
778 }
779 
780 int
781 main(int argc, char *argv[])
782 {
783 	int opt;
784 	int ret = 8;
785 	char *line = NULL, *ttags = NULL;
786 	size_t len, tagslen = 0;
787 	struct option long_options[] = {
788 	    {"execd", no_argument, NULL, 'e'},
789 	    {"help", no_argument, NULL, 'h'},
790 	    {"no-histograms", no_argument, NULL, 'n'},
791 	    {"signed-int", no_argument, NULL, 'i'},
792 	    {"sum-histogram-buckets", no_argument, NULL, 's'},
793 	    {"tags", required_argument, NULL, 't'},
794 	    {0, 0, 0, 0}
795 	};
796 	while ((opt = getopt_long(
797 	    argc, argv, "ehinst:", long_options, NULL)) != -1) {
798 		switch (opt) {
799 		case 'e':
800 			execd_mode = 1;
801 			break;
802 		case 'i':
803 			metric_data_type = 'i';
804 			metric_value_mask = INT64_MAX;
805 			break;
806 		case 'n':
807 			no_histograms = 1;
808 			break;
809 		case 's':
810 			sum_histogram_buckets = 1;
811 			break;
812 		case 't':
813 			free(ttags);
814 			tagslen = strlen(optarg) + 2;
815 			ttags = calloc(1, tagslen);
816 			if (ttags == NULL) {
817 				fprintf(stderr,
818 				    "error: cannot allocate memory "
819 				    "for tags\n");
820 				exit(1);
821 			}
822 			(void) snprintf(ttags, tagslen, ",%s", optarg);
823 			tags = ttags;
824 			break;
825 		default:
826 			usage(argv[0]);
827 		}
828 	}
829 
830 	libzfs_handle_t *g_zfs;
831 	if ((g_zfs = libzfs_init()) == NULL) {
832 		fprintf(stderr,
833 		    "error: cannot initialize libzfs. "
834 		    "Is the zfs module loaded or zrepl running?\n");
835 		exit(EXIT_FAILURE);
836 	}
837 	if (execd_mode == 0) {
838 		ret = zpool_iter(g_zfs, print_stats, argv[optind]);
839 		return (ret);
840 	}
841 	while (getline(&line, &len, stdin) != -1) {
842 		ret = zpool_iter(g_zfs, print_stats, argv[optind]);
843 		fflush(stdout);
844 	}
845 	return (ret);
846 }
847