xref: /freebsd/sys/contrib/openzfs/cmd/zpool_influxdb/zpool_influxdb.c (revision dd41de95a84d979615a2ef11df6850622bf6184e)
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_impl.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 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(char *s)
105 {
106 	char *c, *d;
107 	char *t = (char *)malloc(ZFS_MAX_DATASET_NAME_LEN * 2);
108 	if (t == NULL) {
109 		fprintf(stderr, "error: cannot allocate memory\n");
110 		exit(1);
111 	}
112 
113 	for (c = s, d = t; *c != '\0'; c++, d++) {
114 		switch (*c) {
115 		case ' ':
116 		case ',':
117 		case '=':
118 		case '\\':
119 			*d++ = '\\';
120 		default:
121 			*d = *c;
122 		}
123 	}
124 	*d = '\0';
125 	return (t);
126 }
127 
128 /*
129  * print key=value where value is a uint64_t
130  */
131 static void
132 print_kv(char *key, uint64_t value)
133 {
134 	printf("%s=%llu%c", key,
135 	    (u_longlong_t)value & metric_value_mask, metric_data_type);
136 }
137 
138 /*
139  * print_scan_status() prints the details as often seen in the "zpool status"
140  * output. However, unlike the zpool command, which is intended for humans,
141  * this output is suitable for long-term tracking in influxdb.
142  * TODO: update to include issued scan data
143  */
144 static int
145 print_scan_status(nvlist_t *nvroot, const char *pool_name)
146 {
147 	uint_t c;
148 	int64_t elapsed;
149 	uint64_t examined, pass_exam, paused_time, paused_ts, rate;
150 	uint64_t remaining_time;
151 	pool_scan_stat_t *ps = NULL;
152 	double pct_done;
153 	char *state[DSS_NUM_STATES] = {
154 	    "none", "scanning", "finished", "canceled"};
155 	char *func;
156 
157 	(void) nvlist_lookup_uint64_array(nvroot,
158 	    ZPOOL_CONFIG_SCAN_STATS,
159 	    (uint64_t **)&ps, &c);
160 
161 	/*
162 	 * ignore if there are no stats
163 	 */
164 	if (ps == NULL)
165 		return (0);
166 
167 	/*
168 	 * return error if state is bogus
169 	 */
170 	if (ps->pss_state >= DSS_NUM_STATES ||
171 	    ps->pss_func >= POOL_SCAN_FUNCS) {
172 		if (complained_about_sync % 1000 == 0) {
173 			fprintf(stderr, "error: cannot decode scan stats: "
174 			    "ZFS is out of sync with compiled zpool_influxdb");
175 			complained_about_sync++;
176 		}
177 		return (1);
178 	}
179 
180 	switch (ps->pss_func) {
181 	case POOL_SCAN_NONE:
182 		func = "none_requested";
183 		break;
184 	case POOL_SCAN_SCRUB:
185 		func = "scrub";
186 		break;
187 	case POOL_SCAN_RESILVER:
188 		func = "resilver";
189 		break;
190 #ifdef POOL_SCAN_REBUILD
191 	case POOL_SCAN_REBUILD:
192 		func = "rebuild";
193 		break;
194 #endif
195 	default:
196 		func = "scan";
197 	}
198 
199 	/* overall progress */
200 	examined = ps->pss_examined ? ps->pss_examined : 1;
201 	pct_done = 0.0;
202 	if (ps->pss_to_examine > 0)
203 		pct_done = 100.0 * examined / ps->pss_to_examine;
204 
205 #ifdef EZFS_SCRUB_PAUSED
206 	paused_ts = ps->pss_pass_scrub_pause;
207 	paused_time = ps->pss_pass_scrub_spent_paused;
208 #else
209 	paused_ts = 0;
210 	paused_time = 0;
211 #endif
212 
213 	/* calculations for this pass */
214 	if (ps->pss_state == DSS_SCANNING) {
215 		elapsed = (int64_t)time(NULL) - (int64_t)ps->pss_pass_start -
216 		    (int64_t)paused_time;
217 		elapsed = (elapsed > 0) ? elapsed : 1;
218 		pass_exam = ps->pss_pass_exam ? ps->pss_pass_exam : 1;
219 		rate = pass_exam / elapsed;
220 		rate = (rate > 0) ? rate : 1;
221 		remaining_time = ps->pss_to_examine - examined / rate;
222 	} else {
223 		elapsed =
224 		    (int64_t)ps->pss_end_time - (int64_t)ps->pss_pass_start -
225 		    (int64_t)paused_time;
226 		elapsed = (elapsed > 0) ? elapsed : 1;
227 		pass_exam = ps->pss_pass_exam ? ps->pss_pass_exam : 1;
228 		rate = pass_exam / elapsed;
229 		remaining_time = 0;
230 	}
231 	rate = rate ? rate : 1;
232 
233 	/* influxdb line protocol format: "tags metrics timestamp" */
234 	printf("%s%s,function=%s,name=%s,state=%s ",
235 	    SCAN_MEASUREMENT, tags, func, pool_name, state[ps->pss_state]);
236 	print_kv("end_ts", ps->pss_end_time);
237 	print_kv(",errors", ps->pss_errors);
238 	print_kv(",examined", examined);
239 	print_kv(",issued", ps->pss_issued);
240 	print_kv(",pass_examined", pass_exam);
241 	print_kv(",pass_issued", ps->pss_pass_issued);
242 	print_kv(",paused_ts", paused_ts);
243 	print_kv(",paused_t", paused_time);
244 	printf(",pct_done=%.2f", pct_done);
245 	print_kv(",processed", ps->pss_processed);
246 	print_kv(",rate", rate);
247 	print_kv(",remaining_t", remaining_time);
248 	print_kv(",start_ts", ps->pss_start_time);
249 	print_kv(",to_examine", ps->pss_to_examine);
250 	print_kv(",to_process", ps->pss_to_process);
251 	printf(" %llu\n", (u_longlong_t)timestamp);
252 	return (0);
253 }
254 
255 /*
256  * get a vdev name that corresponds to the top-level vdev names
257  * printed by `zpool status`
258  */
259 static char *
260 get_vdev_name(nvlist_t *nvroot, const char *parent_name)
261 {
262 	static char vdev_name[256];
263 	char *vdev_type = NULL;
264 	uint64_t vdev_id = 0;
265 
266 	if (nvlist_lookup_string(nvroot, ZPOOL_CONFIG_TYPE,
267 	    &vdev_type) != 0) {
268 		vdev_type = "unknown";
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 		    "%s/%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_type = NULL;
300 	uint64_t vdev_id = 0;
301 	char vdev_value[MAXPATHLEN];
302 	char *vdev_path = NULL;
303 	char *s, *t;
304 
305 	if (nvlist_lookup_string(nvroot, ZPOOL_CONFIG_TYPE, &vdev_type) != 0) {
306 		vdev_type = "unknown";
307 	}
308 	if (nvlist_lookup_uint64(nvroot, ZPOOL_CONFIG_ID, &vdev_id) != 0) {
309 		vdev_id = UINT64_MAX;
310 	}
311 	if (nvlist_lookup_string(
312 	    nvroot, ZPOOL_CONFIG_PATH, &vdev_path) != 0) {
313 		vdev_path = NULL;
314 	}
315 
316 	if (parent_name == NULL) {
317 		s = escape_string(vdev_type);
318 		(void) snprintf(vdev_value, sizeof (vdev_value), "vdev=%s", s);
319 		free(s);
320 	} else {
321 		s = escape_string((char *)parent_name);
322 		t = escape_string(vdev_type);
323 		(void) snprintf(vdev_value, sizeof (vdev_value),
324 		    "vdev=%s/%s-%llu", s, t, (u_longlong_t)vdev_id);
325 		free(s);
326 		free(t);
327 	}
328 	if (vdev_path == NULL) {
329 		(void) snprintf(vdev_desc, sizeof (vdev_desc), "%s",
330 		    vdev_value);
331 	} else {
332 		s = escape_string(vdev_path);
333 		(void) snprintf(vdev_desc, sizeof (vdev_desc), "path=%s,%s",
334 		    s, vdev_value);
335 		free(s);
336 	}
337 	return (vdev_desc);
338 }
339 
340 /*
341  * vdev summary stats are a combination of the data shown by
342  * `zpool status` and `zpool list -v`
343  */
344 static int
345 print_summary_stats(nvlist_t *nvroot, const char *pool_name,
346     const char *parent_name)
347 {
348 	uint_t c;
349 	vdev_stat_t *vs;
350 	char *vdev_desc = NULL;
351 	vdev_desc = get_vdev_desc(nvroot, parent_name);
352 	if (nvlist_lookup_uint64_array(nvroot, ZPOOL_CONFIG_VDEV_STATS,
353 	    (uint64_t **)&vs, &c) != 0) {
354 		return (1);
355 	}
356 	printf("%s%s,name=%s,state=%s,%s ", POOL_MEASUREMENT, tags,
357 	    pool_name, zpool_state_to_name((vdev_state_t)vs->vs_state,
358 	    (vdev_aux_t)vs->vs_aux), vdev_desc);
359 	print_kv("alloc", vs->vs_alloc);
360 	print_kv(",free", vs->vs_space - vs->vs_alloc);
361 	print_kv(",size", vs->vs_space);
362 	print_kv(",read_bytes", vs->vs_bytes[ZIO_TYPE_READ]);
363 	print_kv(",read_errors", vs->vs_read_errors);
364 	print_kv(",read_ops", vs->vs_ops[ZIO_TYPE_READ]);
365 	print_kv(",write_bytes", vs->vs_bytes[ZIO_TYPE_WRITE]);
366 	print_kv(",write_errors", vs->vs_write_errors);
367 	print_kv(",write_ops", vs->vs_ops[ZIO_TYPE_WRITE]);
368 	print_kv(",checksum_errors", vs->vs_checksum_errors);
369 	print_kv(",fragmentation", vs->vs_fragmentation);
370 	printf(" %llu\n", (u_longlong_t)timestamp);
371 	return (0);
372 }
373 
374 /*
375  * vdev latency stats are histograms stored as nvlist arrays of uint64.
376  * Latency stats include the ZIO scheduler classes plus lower-level
377  * vdev latencies.
378  *
379  * In many cases, the top-level "root" view obscures the underlying
380  * top-level vdev operations. For example, if a pool has a log, special,
381  * or cache device, then each can behave very differently. It is useful
382  * to see how each is responding.
383  */
384 static int
385 print_vdev_latency_stats(nvlist_t *nvroot, const char *pool_name,
386     const char *parent_name)
387 {
388 	uint_t c, end = 0;
389 	nvlist_t *nv_ex;
390 	char *vdev_desc = NULL;
391 
392 	/* short_names become part of the metric name and are influxdb-ready */
393 	struct lat_lookup {
394 	    char *name;
395 	    char *short_name;
396 	    uint64_t sum;
397 	    uint64_t *array;
398 	};
399 	struct lat_lookup lat_type[] = {
400 	    {ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO,   "total_read", 0},
401 	    {ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO,   "total_write", 0},
402 	    {ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO,  "disk_read", 0},
403 	    {ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO,  "disk_write", 0},
404 	    {ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO,  "sync_read", 0},
405 	    {ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO,  "sync_write", 0},
406 	    {ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO, "async_read", 0},
407 	    {ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO, "async_write", 0},
408 	    {ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO,   "scrub", 0},
409 #ifdef ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO
410 	    {ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO,    "trim", 0},
411 #endif
412 	    {NULL,	NULL}
413 	};
414 
415 	if (nvlist_lookup_nvlist(nvroot,
416 	    ZPOOL_CONFIG_VDEV_STATS_EX, &nv_ex) != 0) {
417 		return (6);
418 	}
419 
420 	vdev_desc = get_vdev_desc(nvroot, parent_name);
421 
422 	for (int i = 0; lat_type[i].name; i++) {
423 		if (nvlist_lookup_uint64_array(nv_ex,
424 		    lat_type[i].name, &lat_type[i].array, &c) != 0) {
425 			fprintf(stderr, "error: can't get %s\n",
426 			    lat_type[i].name);
427 			return (3);
428 		}
429 		/* end count count, all of the arrays are the same size */
430 		end = c - 1;
431 	}
432 
433 	for (int bucket = 0; bucket <= end; bucket++) {
434 		if (bucket < MIN_LAT_INDEX) {
435 			/* don't print, but collect the sum */
436 			for (int i = 0; lat_type[i].name; i++) {
437 				lat_type[i].sum += lat_type[i].array[bucket];
438 			}
439 			continue;
440 		}
441 		if (bucket < end) {
442 			printf("%s%s,le=%0.6f,name=%s,%s ",
443 			    POOL_LATENCY_MEASUREMENT, tags,
444 			    (float)(1ULL << bucket) * 1e-9,
445 			    pool_name, vdev_desc);
446 		} else {
447 			printf("%s%s,le=+Inf,name=%s,%s ",
448 			    POOL_LATENCY_MEASUREMENT, tags, pool_name,
449 			    vdev_desc);
450 		}
451 		for (int i = 0; lat_type[i].name; i++) {
452 			if (bucket <= MIN_LAT_INDEX || sum_histogram_buckets) {
453 				lat_type[i].sum += lat_type[i].array[bucket];
454 			} else {
455 				lat_type[i].sum = lat_type[i].array[bucket];
456 			}
457 			print_kv(lat_type[i].short_name, lat_type[i].sum);
458 			if (lat_type[i + 1].name != NULL) {
459 				printf(",");
460 			}
461 		}
462 		printf(" %llu\n", (u_longlong_t)timestamp);
463 	}
464 	return (0);
465 }
466 
467 /*
468  * vdev request size stats are histograms stored as nvlist arrays of uint64.
469  * Request size stats include the ZIO scheduler classes plus lower-level
470  * vdev sizes. Both independent (ind) and aggregated (agg) sizes are reported.
471  *
472  * In many cases, the top-level "root" view obscures the underlying
473  * top-level vdev operations. For example, if a pool has a log, special,
474  * or cache device, then each can behave very differently. It is useful
475  * to see how each is responding.
476  */
477 static int
478 print_vdev_size_stats(nvlist_t *nvroot, const char *pool_name,
479     const char *parent_name)
480 {
481 	uint_t c, end = 0;
482 	nvlist_t *nv_ex;
483 	char *vdev_desc = NULL;
484 
485 	/* short_names become the field name */
486 	struct size_lookup {
487 	    char *name;
488 	    char *short_name;
489 	    uint64_t sum;
490 	    uint64_t *array;
491 	};
492 	struct size_lookup size_type[] = {
493 	    {ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO,   "sync_read_ind"},
494 	    {ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO,   "sync_write_ind"},
495 	    {ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO,  "async_read_ind"},
496 	    {ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO,  "async_write_ind"},
497 	    {ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO,    "scrub_read_ind"},
498 	    {ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO,   "sync_read_agg"},
499 	    {ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO,   "sync_write_agg"},
500 	    {ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO,  "async_read_agg"},
501 	    {ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO,  "async_write_agg"},
502 	    {ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO,    "scrub_read_agg"},
503 #ifdef ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO
504 	    {ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO,    "trim_write_ind"},
505 	    {ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO,    "trim_write_agg"},
506 #endif
507 	    {NULL,	NULL}
508 	};
509 
510 	if (nvlist_lookup_nvlist(nvroot,
511 	    ZPOOL_CONFIG_VDEV_STATS_EX, &nv_ex) != 0) {
512 		return (6);
513 	}
514 
515 	vdev_desc = get_vdev_desc(nvroot, parent_name);
516 
517 	for (int i = 0; size_type[i].name; i++) {
518 		if (nvlist_lookup_uint64_array(nv_ex, size_type[i].name,
519 		    &size_type[i].array, &c) != 0) {
520 			fprintf(stderr, "error: can't get %s\n",
521 			    size_type[i].name);
522 			return (3);
523 		}
524 		/* end count count, all of the arrays are the same size */
525 		end = c - 1;
526 	}
527 
528 	for (int bucket = 0; bucket <= end; bucket++) {
529 		if (bucket < MIN_SIZE_INDEX) {
530 			/* don't print, but collect the sum */
531 			for (int i = 0; size_type[i].name; i++) {
532 				size_type[i].sum += size_type[i].array[bucket];
533 			}
534 			continue;
535 		}
536 
537 		if (bucket < end) {
538 			printf("%s%s,le=%llu,name=%s,%s ",
539 			    POOL_IO_SIZE_MEASUREMENT, tags, 1ULL << bucket,
540 			    pool_name, vdev_desc);
541 		} else {
542 			printf("%s%s,le=+Inf,name=%s,%s ",
543 			    POOL_IO_SIZE_MEASUREMENT, tags, pool_name,
544 			    vdev_desc);
545 		}
546 		for (int i = 0; size_type[i].name; i++) {
547 			if (bucket <= MIN_SIZE_INDEX || sum_histogram_buckets) {
548 				size_type[i].sum += size_type[i].array[bucket];
549 			} else {
550 				size_type[i].sum = size_type[i].array[bucket];
551 			}
552 			print_kv(size_type[i].short_name, size_type[i].sum);
553 			if (size_type[i + 1].name != NULL) {
554 				printf(",");
555 			}
556 		}
557 		printf(" %llu\n", (u_longlong_t)timestamp);
558 	}
559 	return (0);
560 }
561 
562 /*
563  * ZIO scheduler queue stats are stored as gauges. This is unfortunate
564  * because the values can change very rapidly and any point-in-time
565  * value will quickly be obsoleted. It is also not easy to downsample.
566  * Thus only the top-level queue stats might be beneficial... maybe.
567  */
568 static int
569 print_queue_stats(nvlist_t *nvroot, const char *pool_name,
570     const char *parent_name)
571 {
572 	nvlist_t *nv_ex;
573 	uint64_t value;
574 
575 	/* short_names are used for the field name */
576 	struct queue_lookup {
577 	    char *name;
578 	    char *short_name;
579 	};
580 	struct queue_lookup queue_type[] = {
581 	    {ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE,	"sync_r_active"},
582 	    {ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE,	"sync_w_active"},
583 	    {ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE,	"async_r_active"},
584 	    {ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE,	"async_w_active"},
585 	    {ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE,	"async_scrub_active"},
586 	    {ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE,	"sync_r_pend"},
587 	    {ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE,	"sync_w_pend"},
588 	    {ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE,	"async_r_pend"},
589 	    {ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE,	"async_w_pend"},
590 	    {ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE,	"async_scrub_pend"},
591 	    {NULL,	NULL}
592 	};
593 
594 	if (nvlist_lookup_nvlist(nvroot,
595 	    ZPOOL_CONFIG_VDEV_STATS_EX, &nv_ex) != 0) {
596 		return (6);
597 	}
598 
599 	printf("%s%s,name=%s,%s ", POOL_QUEUE_MEASUREMENT, tags, pool_name,
600 	    get_vdev_desc(nvroot, parent_name));
601 	for (int i = 0; queue_type[i].name; i++) {
602 		if (nvlist_lookup_uint64(nv_ex,
603 		    queue_type[i].name, &value) != 0) {
604 			fprintf(stderr, "error: can't get %s\n",
605 			    queue_type[i].name);
606 			return (3);
607 		}
608 		print_kv(queue_type[i].short_name, value);
609 		if (queue_type[i + 1].name != NULL) {
610 			printf(",");
611 		}
612 	}
613 	printf(" %llu\n", (u_longlong_t)timestamp);
614 	return (0);
615 }
616 
617 /*
618  * top-level vdev stats are at the pool level
619  */
620 static int
621 print_top_level_vdev_stats(nvlist_t *nvroot, const char *pool_name)
622 {
623 	nvlist_t *nv_ex;
624 	uint64_t value;
625 
626 	/* short_names become part of the metric name */
627 	struct queue_lookup {
628 	    char *name;
629 	    char *short_name;
630 	};
631 	struct queue_lookup queue_type[] = {
632 	    {ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE, "sync_r_active_queue"},
633 	    {ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE, "sync_w_active_queue"},
634 	    {ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE, "async_r_active_queue"},
635 	    {ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE, "async_w_active_queue"},
636 	    {ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE, "async_scrub_active_queue"},
637 	    {ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE, "sync_r_pend_queue"},
638 	    {ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE, "sync_w_pend_queue"},
639 	    {ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE, "async_r_pend_queue"},
640 	    {ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE, "async_w_pend_queue"},
641 	    {ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE, "async_scrub_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) strncpy(vdev_name, get_vdev_name(nvroot, parent_name),
687 		    sizeof (vdev_name));
688 		vdev_name[sizeof (vdev_name) - 1] = '\0';
689 
690 		for (c = 0; c < children; c++) {
691 			print_recursive_stats(func, child[c], pool_name,
692 			    vdev_name, descend);
693 		}
694 	}
695 	return (0);
696 }
697 
698 /*
699  * call-back to print the stats from the pool config
700  *
701  * Note: if the pool is broken, this can hang indefinitely and perhaps in an
702  * unkillable state.
703  */
704 static int
705 print_stats(zpool_handle_t *zhp, void *data)
706 {
707 	uint_t c;
708 	int err;
709 	boolean_t missing;
710 	nvlist_t *config, *nvroot;
711 	vdev_stat_t *vs;
712 	struct timespec tv;
713 	char *pool_name;
714 
715 	/* if not this pool return quickly */
716 	if (data &&
717 	    strncmp(data, zhp->zpool_name, ZFS_MAX_DATASET_NAME_LEN) != 0) {
718 		zpool_close(zhp);
719 		return (0);
720 	}
721 
722 	if (zpool_refresh_stats(zhp, &missing) != 0) {
723 		zpool_close(zhp);
724 		return (1);
725 	}
726 
727 	config = zpool_get_config(zhp, NULL);
728 	if (clock_gettime(CLOCK_REALTIME, &tv) != 0)
729 		timestamp = (uint64_t)time(NULL) * 1000000000;
730 	else
731 		timestamp =
732 		    ((uint64_t)tv.tv_sec * 1000000000) + (uint64_t)tv.tv_nsec;
733 
734 	if (nvlist_lookup_nvlist(
735 	    config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) != 0) {
736 	zpool_close(zhp);
737 		return (2);
738 	}
739 	if (nvlist_lookup_uint64_array(nvroot, ZPOOL_CONFIG_VDEV_STATS,
740 	    (uint64_t **)&vs, &c) != 0) {
741 	zpool_close(zhp);
742 		return (3);
743 	}
744 
745 	pool_name = escape_string(zhp->zpool_name);
746 	err = print_recursive_stats(print_summary_stats, nvroot,
747 	    pool_name, NULL, 1);
748 	/* if any of these return an error, skip the rest */
749 	if (err == 0)
750 	err = print_top_level_vdev_stats(nvroot, pool_name);
751 
752 	if (no_histograms == 0) {
753 	if (err == 0)
754 		err = print_recursive_stats(print_vdev_latency_stats, nvroot,
755 		    pool_name, NULL, 1);
756 	if (err == 0)
757 		err = print_recursive_stats(print_vdev_size_stats, nvroot,
758 		    pool_name, NULL, 1);
759 	if (err == 0)
760 		err = print_recursive_stats(print_queue_stats, nvroot,
761 		    pool_name, NULL, 0);
762 	}
763 	if (err == 0)
764 		err = print_scan_status(nvroot, pool_name);
765 
766 	free(pool_name);
767 	zpool_close(zhp);
768 	return (err);
769 }
770 
771 static void
772 usage(char *name)
773 {
774 	fprintf(stderr, "usage: %s [--execd][--no-histograms]"
775 	    "[--sum-histogram-buckets] [--signed-int] [poolname]\n", name);
776 	exit(EXIT_FAILURE);
777 }
778 
779 int
780 main(int argc, char *argv[])
781 {
782 	int opt;
783 	int ret = 8;
784 	char *line = NULL;
785 	size_t len, tagslen = 0;
786 	struct option long_options[] = {
787 	    {"execd", no_argument, NULL, 'e'},
788 	    {"help", no_argument, NULL, 'h'},
789 	    {"no-histograms", no_argument, NULL, 'n'},
790 	    {"signed-int", no_argument, NULL, 'i'},
791 	    {"sum-histogram-buckets", no_argument, NULL, 's'},
792 	    {"tags", required_argument, NULL, 't'},
793 	    {0, 0, 0, 0}
794 	};
795 	while ((opt = getopt_long(
796 	    argc, argv, "ehinst:", long_options, NULL)) != -1) {
797 		switch (opt) {
798 		case 'e':
799 			execd_mode = 1;
800 			break;
801 		case 'i':
802 			metric_data_type = 'i';
803 			metric_value_mask = INT64_MAX;
804 			break;
805 		case 'n':
806 			no_histograms = 1;
807 			break;
808 		case 's':
809 			sum_histogram_buckets = 1;
810 			break;
811 		case 't':
812 			tagslen = strlen(optarg) + 2;
813 			tags = calloc(tagslen, 1);
814 			if (tags == NULL) {
815 				fprintf(stderr,
816 				    "error: cannot allocate memory "
817 				    "for tags\n");
818 				exit(1);
819 			}
820 			(void) snprintf(tags, tagslen, ",%s", optarg);
821 			break;
822 		default:
823 			usage(argv[0]);
824 		}
825 	}
826 
827 	libzfs_handle_t *g_zfs;
828 	if ((g_zfs = libzfs_init()) == NULL) {
829 		fprintf(stderr,
830 		    "error: cannot initialize libzfs. "
831 		    "Is the zfs module loaded or zrepl running?\n");
832 		exit(EXIT_FAILURE);
833 	}
834 	if (execd_mode == 0) {
835 		ret = zpool_iter(g_zfs, print_stats, argv[optind]);
836 		return (ret);
837 	}
838 	while (getline(&line, &len, stdin) != -1) {
839 		ret = zpool_iter(g_zfs, print_stats, argv[optind]);
840 		fflush(stdout);
841 	}
842 	return (ret);
843 }
844