1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
2 // vim: ts=8 sw=2 smarttab
4 #include <boost/algorithm/string.hpp>
8 #define dout_subsys ceph_subsys_mon
9 #include "common/debug.h"
10 #include "common/Formatter.h"
11 #include "include/ceph_features.h"
12 #include "include/stringify.h"
14 #include "osd/osd_types.h"
15 #include "osd/OSDMap.h"
17 #define dout_context g_ceph_context
19 MEMPOOL_DEFINE_OBJECT_FACTORY(PGMapDigest
, pgmap_digest
, pgmap
);
20 MEMPOOL_DEFINE_OBJECT_FACTORY(PGMap
, pgmap
, pgmap
);
21 MEMPOOL_DEFINE_OBJECT_FACTORY(PGMap::Incremental
, pgmap_inc
, pgmap
);
24 // ---------------------
27 void PGMapDigest::encode(bufferlist
& bl
, uint64_t features
) const
29 // NOTE: see PGMap::encode_digest
30 ENCODE_START(1, 1, bl
);
32 ::encode(num_pg_active
, bl
);
33 ::encode(num_pg_unknown
, bl
);
34 ::encode(num_osd
, bl
);
35 ::encode(pg_pool_sum
, bl
, features
);
36 ::encode(pg_sum
, bl
, features
);
37 ::encode(osd_sum
, bl
);
38 ::encode(num_pg_by_state
, bl
);
39 ::encode(num_pg_by_osd
, bl
);
40 ::encode(num_pg_by_pool
, bl
);
41 ::encode(osd_last_seq
, bl
);
42 ::encode(per_pool_sum_delta
, bl
, features
);
43 ::encode(per_pool_sum_deltas_stamps
, bl
);
44 ::encode(pg_sum_delta
, bl
, features
);
45 ::encode(stamp_delta
, bl
);
46 ::encode(avail_space_by_rule
, bl
);
50 void PGMapDigest::decode(bufferlist::iterator
& p
)
54 ::decode(num_pg_active
, p
);
55 ::decode(num_pg_unknown
, p
);
57 ::decode(pg_pool_sum
, p
);
60 ::decode(num_pg_by_state
, p
);
61 ::decode(num_pg_by_osd
, p
);
62 ::decode(num_pg_by_pool
, p
);
63 ::decode(osd_last_seq
, p
);
64 ::decode(per_pool_sum_delta
, p
);
65 ::decode(per_pool_sum_deltas_stamps
, p
);
66 ::decode(pg_sum_delta
, p
);
67 ::decode(stamp_delta
, p
);
68 ::decode(avail_space_by_rule
, p
);
72 void PGMapDigest::dump(Formatter
*f
) const
74 f
->dump_unsigned("num_pg", num_pg
);
75 f
->dump_unsigned("num_pg_active", num_pg_active
);
76 f
->dump_unsigned("num_pg_unknown", num_pg_unknown
);
77 f
->dump_unsigned("num_osd", num_osd
);
78 f
->dump_object("pool_sum", pg_sum
);
79 f
->dump_object("osd_sum", osd_sum
);
80 f
->open_array_section("pool_stats");
81 for (auto& p
: pg_pool_sum
) {
82 f
->open_object_section("pool_stat");
83 f
->dump_int("poolid", p
.first
);
84 auto q
= num_pg_by_pool
.find(p
.first
);
85 if (q
!= num_pg_by_pool
.end())
86 f
->dump_unsigned("num_pg", q
->second
);
91 f
->open_array_section("osd_stats");
93 // TODO: this isn't really correct since we can dump non-existent OSDs
94 // I dunno what osd_last_seq is set to in that case...
95 for (auto& p
: osd_last_seq
) {
96 f
->open_object_section("osd_stat");
97 f
->dump_int("osd", i
);
98 f
->dump_unsigned("seq", p
);
103 f
->open_array_section("num_pg_by_state");
104 for (auto& p
: num_pg_by_state
) {
105 f
->open_object_section("count");
106 f
->dump_string("state", pg_state_string(p
.first
));
107 f
->dump_unsigned("num", p
.second
);
111 f
->open_array_section("num_pg_by_osd");
112 for (auto& p
: num_pg_by_osd
) {
113 f
->open_object_section("count");
114 f
->dump_unsigned("osd", p
.first
);
115 f
->dump_unsigned("num_primary_pg", p
.second
.primary
);
116 f
->dump_unsigned("num_acting_pg", p
.second
.acting
);
117 f
->dump_unsigned("num_up_pg", p
.second
.up
);
123 void PGMapDigest::generate_test_instances(list
<PGMapDigest
*>& ls
)
125 ls
.push_back(new PGMapDigest
);
128 inline std::string
percentify(const float& a
) {
129 std::stringstream ss
;
133 ss
<< std::fixed
<< std::setprecision(2) << a
;
137 void PGMapDigest::print_summary(Formatter
*f
, ostream
*out
) const
140 f
->open_array_section("pgs_by_state");
142 // list is descending numeric order (by count)
143 multimap
<int,int> state_by_count
; // count -> state
144 for (auto p
= num_pg_by_state
.begin();
145 p
!= num_pg_by_state
.end();
147 state_by_count
.insert(make_pair(p
->second
, p
->first
));
150 for (auto p
= state_by_count
.rbegin();
151 p
!= state_by_count
.rend();
154 f
->open_object_section("pgs_by_state_element");
155 f
->dump_string("state_name", pg_state_string(p
->second
));
156 f
->dump_unsigned("count", p
->first
);
164 f
->dump_unsigned("num_pgs", num_pg
);
165 f
->dump_unsigned("num_pools", pg_pool_sum
.size());
166 f
->dump_unsigned("num_objects", pg_sum
.stats
.sum
.num_objects
);
167 f
->dump_unsigned("data_bytes", pg_sum
.stats
.sum
.num_bytes
);
168 f
->dump_unsigned("bytes_used", osd_sum
.kb_used
* 1024ull);
169 f
->dump_unsigned("bytes_avail", osd_sum
.kb_avail
* 1024ull);
170 f
->dump_unsigned("bytes_total", osd_sum
.kb
* 1024ull);
172 *out
<< " pools: " << pg_pool_sum
.size() << " pools, "
173 << num_pg
<< " pgs\n";
174 *out
<< " objects: " << si_t(pg_sum
.stats
.sum
.num_objects
) << " objects, "
175 << prettybyte_t(pg_sum
.stats
.sum
.num_bytes
) << "\n";
177 << kb_t(osd_sum
.kb_used
) << " used, "
178 << kb_t(osd_sum
.kb_avail
) << " / "
179 << kb_t(osd_sum
.kb
) << " avail\n";
185 if (num_pg_unknown
> 0) {
186 float p
= (float)num_pg_unknown
/ (float)num_pg
;
188 f
->dump_float("unknown_pgs_ratio", p
);
191 snprintf(b
, sizeof(b
), "%.3lf", p
* 100.0);
192 *out
<< b
<< "% pgs unknown\n";
197 int num_pg_inactive
= num_pg
- num_pg_active
- num_pg_unknown
;
198 if (num_pg_inactive
> 0) {
199 float p
= (float)num_pg_inactive
/ (float)num_pg
;
201 f
->dump_float("inactive_pgs_ratio", p
);
207 snprintf(b
, sizeof(b
), "%.3f", p
* 100.0);
208 *out
<< b
<< "% pgs not active\n";
214 overall_recovery_summary(f
, &sl
);
215 if (!f
&& !sl
.empty()) {
216 for (auto p
= sl
.begin(); p
!= sl
.end(); ++p
) {
227 unsigned max_width
= 1;
228 for (multimap
<int,int>::reverse_iterator p
= state_by_count
.rbegin();
229 p
!= state_by_count
.rend();
232 std::stringstream ss
;
234 max_width
= MAX(ss
.str().size(), max_width
);
237 for (multimap
<int,int>::reverse_iterator p
= state_by_count
.rbegin();
238 p
!= state_by_count
.rend();
245 out
->setf(std::ios::left
);
246 *out
<< std::setw(max_width
) << p
->first
247 << " " << pg_state_string(p
->second
) << "\n";
248 out
->unsetf(std::ios::left
);
252 ostringstream ss_rec_io
;
253 overall_recovery_rate_summary(f
, &ss_rec_io
);
254 ostringstream ss_client_io
;
255 overall_client_io_rate_summary(f
, &ss_client_io
);
256 ostringstream ss_cache_io
;
257 overall_cache_io_rate_summary(f
, &ss_cache_io
);
259 if (!f
&& (ss_client_io
.str().length() || ss_rec_io
.str().length()
260 || ss_cache_io
.str().length())) {
265 if (!f
&& ss_client_io
.str().length())
266 *out
<< " client: " << ss_client_io
.str() << "\n";
267 if (!f
&& ss_rec_io
.str().length())
268 *out
<< " recovery: " << ss_rec_io
.str() << "\n";
269 if (!f
&& ss_cache_io
.str().length())
270 *out
<< " cache: " << ss_cache_io
.str() << "\n";
273 void PGMapDigest::print_oneline_summary(Formatter
*f
, ostream
*out
) const
275 std::stringstream ss
;
278 f
->open_array_section("num_pg_by_state");
279 for (auto p
= num_pg_by_state
.begin();
280 p
!= num_pg_by_state
.end();
283 f
->open_object_section("state");
284 f
->dump_string("name", pg_state_string(p
->first
));
285 f
->dump_unsigned("num", p
->second
);
288 if (p
!= num_pg_by_state
.begin())
290 ss
<< p
->second
<< " " << pg_state_string(p
->first
);
295 string states
= ss
.str();
297 *out
<< num_pg
<< " pgs: "
299 << prettybyte_t(pg_sum
.stats
.sum
.num_bytes
) << " data, "
300 << kb_t(osd_sum
.kb_used
) << " used, "
301 << kb_t(osd_sum
.kb_avail
) << " / "
302 << kb_t(osd_sum
.kb
) << " avail";
304 f
->dump_unsigned("num_pgs", num_pg
);
305 f
->dump_unsigned("num_bytes", pg_sum
.stats
.sum
.num_bytes
);
306 f
->dump_unsigned("raw_bytes_used", osd_sum
.kb_used
<< 10);
307 f
->dump_unsigned("raw_bytes_avail", osd_sum
.kb_avail
<< 10);
308 f
->dump_unsigned("raw_bytes", osd_sum
.kb
<< 10);
311 // make non-negative; we can get negative values if osds send
312 // uncommitted stats and then "go backward" or if they are just
314 pool_stat_t pos_delta
= pg_sum_delta
;
316 if (pos_delta
.stats
.sum
.num_rd
||
317 pos_delta
.stats
.sum
.num_wr
) {
320 if (pos_delta
.stats
.sum
.num_rd
) {
321 int64_t rd
= (pos_delta
.stats
.sum
.num_rd_kb
<< 10) / (double)stamp_delta
;
323 *out
<< pretty_si_t(rd
) << "B/s rd, ";
325 f
->dump_unsigned("read_bytes_sec", rd
);
327 if (pos_delta
.stats
.sum
.num_wr
) {
328 int64_t wr
= (pos_delta
.stats
.sum
.num_wr_kb
<< 10) / (double)stamp_delta
;
330 *out
<< pretty_si_t(wr
) << "B/s wr, ";
332 f
->dump_unsigned("write_bytes_sec", wr
);
334 int64_t iops
= (pos_delta
.stats
.sum
.num_rd
+ pos_delta
.stats
.sum
.num_wr
) / (double)stamp_delta
;
336 *out
<< pretty_si_t(iops
) << "op/s";
338 f
->dump_unsigned("io_sec", iops
);
342 overall_recovery_summary(f
, &sl
);
344 for (auto p
= sl
.begin(); p
!= sl
.end(); ++p
)
346 std::stringstream ssr
;
347 overall_recovery_rate_summary(f
, &ssr
);
348 if (out
&& ssr
.str().length())
349 *out
<< "; " << ssr
.str() << " recovering";
352 void PGMapDigest::recovery_summary(Formatter
*f
, list
<string
> *psl
,
353 const pool_stat_t
& delta_sum
) const
355 if (delta_sum
.stats
.sum
.num_objects_degraded
&& delta_sum
.stats
.sum
.num_object_copies
> 0) {
356 double pc
= (double)delta_sum
.stats
.sum
.num_objects_degraded
/
357 (double)delta_sum
.stats
.sum
.num_object_copies
* (double)100.0;
359 snprintf(b
, sizeof(b
), "%.3lf", pc
);
361 f
->dump_unsigned("degraded_objects", delta_sum
.stats
.sum
.num_objects_degraded
);
362 f
->dump_unsigned("degraded_total", delta_sum
.stats
.sum
.num_object_copies
);
363 f
->dump_float("degraded_ratio", pc
/ 100.0);
366 ss
<< delta_sum
.stats
.sum
.num_objects_degraded
367 << "/" << delta_sum
.stats
.sum
.num_object_copies
<< " objects degraded (" << b
<< "%)";
368 psl
->push_back(ss
.str());
371 if (delta_sum
.stats
.sum
.num_objects_misplaced
&& delta_sum
.stats
.sum
.num_object_copies
> 0) {
372 double pc
= (double)delta_sum
.stats
.sum
.num_objects_misplaced
/
373 (double)delta_sum
.stats
.sum
.num_object_copies
* (double)100.0;
375 snprintf(b
, sizeof(b
), "%.3lf", pc
);
377 f
->dump_unsigned("misplaced_objects", delta_sum
.stats
.sum
.num_objects_misplaced
);
378 f
->dump_unsigned("misplaced_total", delta_sum
.stats
.sum
.num_object_copies
);
379 f
->dump_float("misplaced_ratio", pc
/ 100.0);
382 ss
<< delta_sum
.stats
.sum
.num_objects_misplaced
383 << "/" << delta_sum
.stats
.sum
.num_object_copies
<< " objects misplaced (" << b
<< "%)";
384 psl
->push_back(ss
.str());
387 if (delta_sum
.stats
.sum
.num_objects_unfound
&& delta_sum
.stats
.sum
.num_objects
) {
388 double pc
= (double)delta_sum
.stats
.sum
.num_objects_unfound
/
389 (double)delta_sum
.stats
.sum
.num_objects
* (double)100.0;
391 snprintf(b
, sizeof(b
), "%.3lf", pc
);
393 f
->dump_unsigned("unfound_objects", delta_sum
.stats
.sum
.num_objects_unfound
);
394 f
->dump_unsigned("unfound_total", delta_sum
.stats
.sum
.num_objects
);
395 f
->dump_float("unfound_ratio", pc
/ 100.0);
398 ss
<< delta_sum
.stats
.sum
.num_objects_unfound
399 << "/" << delta_sum
.stats
.sum
.num_objects
<< " objects unfound (" << b
<< "%)";
400 psl
->push_back(ss
.str());
405 void PGMapDigest::recovery_rate_summary(Formatter
*f
, ostream
*out
,
406 const pool_stat_t
& delta_sum
,
407 utime_t delta_stamp
) const
409 // make non-negative; we can get negative values if osds send
410 // uncommitted stats and then "go backward" or if they are just
412 pool_stat_t pos_delta
= delta_sum
;
414 if (pos_delta
.stats
.sum
.num_objects_recovered
||
415 pos_delta
.stats
.sum
.num_bytes_recovered
||
416 pos_delta
.stats
.sum
.num_keys_recovered
) {
417 int64_t objps
= pos_delta
.stats
.sum
.num_objects_recovered
/ (double)delta_stamp
;
418 int64_t bps
= pos_delta
.stats
.sum
.num_bytes_recovered
/ (double)delta_stamp
;
419 int64_t kps
= pos_delta
.stats
.sum
.num_keys_recovered
/ (double)delta_stamp
;
421 f
->dump_int("recovering_objects_per_sec", objps
);
422 f
->dump_int("recovering_bytes_per_sec", bps
);
423 f
->dump_int("recovering_keys_per_sec", kps
);
424 f
->dump_int("num_objects_recovered", pos_delta
.stats
.sum
.num_objects_recovered
);
425 f
->dump_int("num_bytes_recovered", pos_delta
.stats
.sum
.num_bytes_recovered
);
426 f
->dump_int("num_keys_recovered", pos_delta
.stats
.sum
.num_keys_recovered
);
428 *out
<< pretty_si_t(bps
) << "B/s";
429 if (pos_delta
.stats
.sum
.num_keys_recovered
)
430 *out
<< ", " << pretty_si_t(kps
) << "keys/s";
431 *out
<< ", " << pretty_si_t(objps
) << "objects/s";
436 void PGMapDigest::overall_recovery_rate_summary(Formatter
*f
, ostream
*out
) const
438 recovery_rate_summary(f
, out
, pg_sum_delta
, stamp_delta
);
441 void PGMapDigest::overall_recovery_summary(Formatter
*f
, list
<string
> *psl
) const
443 recovery_summary(f
, psl
, pg_sum
);
446 void PGMapDigest::pool_recovery_rate_summary(Formatter
*f
, ostream
*out
,
447 uint64_t poolid
) const
449 auto p
= per_pool_sum_delta
.find(poolid
);
450 if (p
== per_pool_sum_delta
.end())
453 auto ts
= per_pool_sum_deltas_stamps
.find(p
->first
);
454 assert(ts
!= per_pool_sum_deltas_stamps
.end());
455 recovery_rate_summary(f
, out
, p
->second
.first
, ts
->second
);
458 void PGMapDigest::pool_recovery_summary(Formatter
*f
, list
<string
> *psl
,
459 uint64_t poolid
) const
461 auto p
= per_pool_sum_delta
.find(poolid
);
462 if (p
== per_pool_sum_delta
.end())
465 recovery_summary(f
, psl
, p
->second
.first
);
468 void PGMapDigest::client_io_rate_summary(Formatter
*f
, ostream
*out
,
469 const pool_stat_t
& delta_sum
,
470 utime_t delta_stamp
) const
472 pool_stat_t pos_delta
= delta_sum
;
474 if (pos_delta
.stats
.sum
.num_rd
||
475 pos_delta
.stats
.sum
.num_wr
) {
476 if (pos_delta
.stats
.sum
.num_rd
) {
477 int64_t rd
= (pos_delta
.stats
.sum
.num_rd_kb
<< 10) / (double)delta_stamp
;
479 f
->dump_int("read_bytes_sec", rd
);
481 *out
<< pretty_si_t(rd
) << "B/s rd, ";
484 if (pos_delta
.stats
.sum
.num_wr
) {
485 int64_t wr
= (pos_delta
.stats
.sum
.num_wr_kb
<< 10) / (double)delta_stamp
;
487 f
->dump_int("write_bytes_sec", wr
);
489 *out
<< pretty_si_t(wr
) << "B/s wr, ";
492 int64_t iops_rd
= pos_delta
.stats
.sum
.num_rd
/ (double)delta_stamp
;
493 int64_t iops_wr
= pos_delta
.stats
.sum
.num_wr
/ (double)delta_stamp
;
495 f
->dump_int("read_op_per_sec", iops_rd
);
496 f
->dump_int("write_op_per_sec", iops_wr
);
498 *out
<< pretty_si_t(iops_rd
) << "op/s rd, " << pretty_si_t(iops_wr
) << "op/s wr";
503 void PGMapDigest::overall_client_io_rate_summary(Formatter
*f
, ostream
*out
) const
505 client_io_rate_summary(f
, out
, pg_sum_delta
, stamp_delta
);
508 void PGMapDigest::pool_client_io_rate_summary(Formatter
*f
, ostream
*out
,
509 uint64_t poolid
) const
511 auto p
= per_pool_sum_delta
.find(poolid
);
512 if (p
== per_pool_sum_delta
.end())
515 auto ts
= per_pool_sum_deltas_stamps
.find(p
->first
);
516 assert(ts
!= per_pool_sum_deltas_stamps
.end());
517 client_io_rate_summary(f
, out
, p
->second
.first
, ts
->second
);
520 void PGMapDigest::cache_io_rate_summary(Formatter
*f
, ostream
*out
,
521 const pool_stat_t
& delta_sum
,
522 utime_t delta_stamp
) const
524 pool_stat_t pos_delta
= delta_sum
;
526 bool have_output
= false;
528 if (pos_delta
.stats
.sum
.num_flush
) {
529 int64_t flush
= (pos_delta
.stats
.sum
.num_flush_kb
<< 10) / (double)delta_stamp
;
531 f
->dump_int("flush_bytes_sec", flush
);
533 *out
<< pretty_si_t(flush
) << "B/s flush";
537 if (pos_delta
.stats
.sum
.num_evict
) {
538 int64_t evict
= (pos_delta
.stats
.sum
.num_evict_kb
<< 10) / (double)delta_stamp
;
540 f
->dump_int("evict_bytes_sec", evict
);
544 *out
<< pretty_si_t(evict
) << "B/s evict";
548 if (pos_delta
.stats
.sum
.num_promote
) {
549 int64_t promote
= pos_delta
.stats
.sum
.num_promote
/ (double)delta_stamp
;
551 f
->dump_int("promote_op_per_sec", promote
);
555 *out
<< pretty_si_t(promote
) << "op/s promote";
559 if (pos_delta
.stats
.sum
.num_flush_mode_low
) {
561 f
->dump_int("num_flush_mode_low", pos_delta
.stats
.sum
.num_flush_mode_low
);
565 *out
<< pretty_si_t(pos_delta
.stats
.sum
.num_flush_mode_low
) << "PG(s) flushing";
569 if (pos_delta
.stats
.sum
.num_flush_mode_high
) {
571 f
->dump_int("num_flush_mode_high", pos_delta
.stats
.sum
.num_flush_mode_high
);
575 *out
<< pretty_si_t(pos_delta
.stats
.sum
.num_flush_mode_high
) << "PG(s) flushing (high)";
579 if (pos_delta
.stats
.sum
.num_evict_mode_some
) {
581 f
->dump_int("num_evict_mode_some", pos_delta
.stats
.sum
.num_evict_mode_some
);
585 *out
<< pretty_si_t(pos_delta
.stats
.sum
.num_evict_mode_some
) << "PG(s) evicting";
589 if (pos_delta
.stats
.sum
.num_evict_mode_full
) {
591 f
->dump_int("num_evict_mode_full", pos_delta
.stats
.sum
.num_evict_mode_full
);
595 *out
<< pretty_si_t(pos_delta
.stats
.sum
.num_evict_mode_full
) << "PG(s) evicting (full)";
600 void PGMapDigest::overall_cache_io_rate_summary(Formatter
*f
, ostream
*out
) const
602 cache_io_rate_summary(f
, out
, pg_sum_delta
, stamp_delta
);
605 void PGMapDigest::pool_cache_io_rate_summary(Formatter
*f
, ostream
*out
,
606 uint64_t poolid
) const
608 auto p
= per_pool_sum_delta
.find(poolid
);
609 if (p
== per_pool_sum_delta
.end())
612 auto ts
= per_pool_sum_deltas_stamps
.find(p
->first
);
613 assert(ts
!= per_pool_sum_deltas_stamps
.end());
614 cache_io_rate_summary(f
, out
, p
->second
.first
, ts
->second
);
617 static float pool_raw_used_rate(const OSDMap
&osd_map
, int64_t poolid
)
619 const pg_pool_t
*pool
= osd_map
.get_pg_pool(poolid
);
621 switch (pool
->get_type()) {
622 case pg_pool_t::TYPE_REPLICATED
:
623 return pool
->get_size();
625 case pg_pool_t::TYPE_ERASURE
:
628 osd_map
.get_erasure_code_profile(pool
->erasure_code_profile
);
629 auto pm
= ecp
.find("m");
630 auto pk
= ecp
.find("k");
631 if (pm
!= ecp
.end() && pk
!= ecp
.end()) {
632 int k
= atoi(pk
->second
.c_str());
633 int m
= atoi(pm
->second
.c_str());
637 return (float)mk
/ k
;
644 assert(0 == "unrecognized pool type");
648 ceph_statfs
PGMapDigest::get_statfs(OSDMap
&osdmap
,
649 boost::optional
<int64_t> data_pool
) const
653 object_stat_sum_t sum
;
656 auto i
= pg_pool_sum
.find(*data_pool
);
657 if (i
!= pg_pool_sum
.end()) {
658 sum
= i
->second
.stats
.sum
;
664 statfs
.kb_used
= (sum
.num_bytes
>> 10);
665 statfs
.kb_avail
= get_pool_free_space(osdmap
, *data_pool
) >> 10;
666 statfs
.num_objects
= sum
.num_objects
;
667 statfs
.kb
= statfs
.kb_used
+ statfs
.kb_avail
;
670 statfs
.kb
= osd_sum
.kb
;
671 statfs
.kb_used
= osd_sum
.kb_used
;
672 statfs
.kb_avail
= osd_sum
.kb_avail
;
673 statfs
.num_objects
= pg_sum
.stats
.sum
.num_objects
;
679 void PGMapDigest::dump_pool_stats_full(
680 const OSDMap
&osd_map
,
688 f
->open_array_section("pools");
690 tbl
.define_column("NAME", TextTable::LEFT
, TextTable::LEFT
);
691 tbl
.define_column("ID", TextTable::LEFT
, TextTable::LEFT
);
693 tbl
.define_column("QUOTA OBJECTS", TextTable::LEFT
, TextTable::LEFT
);
694 tbl
.define_column("QUOTA BYTES", TextTable::LEFT
, TextTable::LEFT
);
697 tbl
.define_column("USED", TextTable::LEFT
, TextTable::RIGHT
);
698 tbl
.define_column("%USED", TextTable::LEFT
, TextTable::RIGHT
);
699 tbl
.define_column("MAX AVAIL", TextTable::LEFT
, TextTable::RIGHT
);
700 tbl
.define_column("OBJECTS", TextTable::LEFT
, TextTable::RIGHT
);
702 tbl
.define_column("DIRTY", TextTable::LEFT
, TextTable::RIGHT
);
703 tbl
.define_column("READ", TextTable::LEFT
, TextTable::RIGHT
);
704 tbl
.define_column("WRITE", TextTable::LEFT
, TextTable::RIGHT
);
705 tbl
.define_column("RAW USED", TextTable::LEFT
, TextTable::RIGHT
);
709 map
<int,uint64_t> avail_by_rule
;
710 for (auto p
= osd_map
.get_pools().begin();
711 p
!= osd_map
.get_pools().end(); ++p
) {
712 int64_t pool_id
= p
->first
;
713 if ((pool_id
< 0) || (pg_pool_sum
.count(pool_id
) == 0))
715 const string
& pool_name
= osd_map
.get_pool_name(pool_id
);
716 const pool_stat_t
&stat
= pg_pool_sum
.at(pool_id
);
718 const pg_pool_t
*pool
= osd_map
.get_pg_pool(pool_id
);
719 int ruleno
= osd_map
.crush
->find_rule(pool
->get_crush_rule(),
724 if (avail_by_rule
.count(ruleno
) == 0) {
725 // FIXME: we don't guarantee avail_space_by_rule is up-to-date before this function is invoked
726 avail
= get_rule_avail(ruleno
);
729 avail_by_rule
[ruleno
] = avail
;
731 avail
= avail_by_rule
[ruleno
];
734 raw_used_rate
= ::pool_raw_used_rate(osd_map
, pool_id
);
737 f
->open_object_section("pool");
738 f
->dump_string("name", pool_name
);
739 f
->dump_int("id", pool_id
);
740 f
->open_object_section("stats");
745 if (pool
->quota_max_objects
== 0)
748 tbl
<< si_t(pool
->quota_max_objects
);
750 if (pool
->quota_max_bytes
== 0)
753 tbl
<< si_t(pool
->quota_max_bytes
);
757 dump_object_stat_sum(tbl
, f
, stat
.stats
.sum
, avail
, raw_used_rate
, verbose
, pool
);
759 f
->close_section(); // stats
761 tbl
<< TextTable::endrow
;
764 f
->close_section(); // pool
769 assert(ss
!= nullptr);
776 void PGMapDigest::dump_fs_stats(stringstream
*ss
, Formatter
*f
, bool verbose
) const
779 f
->open_object_section("stats");
780 f
->dump_int("total_bytes", osd_sum
.kb
* 1024ull);
781 f
->dump_int("total_used_bytes", osd_sum
.kb_used
* 1024ull);
782 f
->dump_int("total_avail_bytes", osd_sum
.kb_avail
* 1024ull);
784 f
->dump_int("total_objects", pg_sum
.stats
.sum
.num_objects
);
788 assert(ss
!= nullptr);
790 tbl
.define_column("SIZE", TextTable::LEFT
, TextTable::RIGHT
);
791 tbl
.define_column("AVAIL", TextTable::LEFT
, TextTable::RIGHT
);
792 tbl
.define_column("RAW USED", TextTable::LEFT
, TextTable::RIGHT
);
793 tbl
.define_column("%RAW USED", TextTable::LEFT
, TextTable::RIGHT
);
795 tbl
.define_column("OBJECTS", TextTable::LEFT
, TextTable::RIGHT
);
797 tbl
<< stringify(si_t(osd_sum
.kb
*1024))
798 << stringify(si_t(osd_sum
.kb_avail
*1024))
799 << stringify(si_t(osd_sum
.kb_used
*1024));
801 if (osd_sum
.kb
> 0) {
802 used
= ((float)osd_sum
.kb_used
/ osd_sum
.kb
);
804 tbl
<< percentify(used
*100);
806 tbl
<< stringify(si_t(pg_sum
.stats
.sum
.num_objects
));
808 tbl
<< TextTable::endrow
;
816 void PGMapDigest::dump_object_stat_sum(
817 TextTable
&tbl
, Formatter
*f
,
818 const object_stat_sum_t
&sum
, uint64_t avail
,
819 float raw_used_rate
, bool verbose
,
820 const pg_pool_t
*pool
)
822 float curr_object_copies_rate
= 0.0;
823 if (sum
.num_object_copies
> 0)
824 curr_object_copies_rate
= (float)(sum
.num_object_copies
- sum
.num_objects_degraded
) / sum
.num_object_copies
;
828 used
= sum
.num_bytes
* curr_object_copies_rate
;
829 used
/= used
+ avail
;
830 } else if (sum
.num_bytes
) {
835 f
->dump_int("kb_used", SHIFT_ROUND_UP(sum
.num_bytes
, 10));
836 f
->dump_int("bytes_used", sum
.num_bytes
);
837 f
->dump_format_unquoted("percent_used", "%.2f", (used
*100));
838 f
->dump_unsigned("max_avail", avail
);
839 f
->dump_int("objects", sum
.num_objects
);
841 f
->dump_int("quota_objects", pool
->quota_max_objects
);
842 f
->dump_int("quota_bytes", pool
->quota_max_bytes
);
843 f
->dump_int("dirty", sum
.num_objects_dirty
);
844 f
->dump_int("rd", sum
.num_rd
);
845 f
->dump_int("rd_bytes", sum
.num_rd_kb
* 1024ull);
846 f
->dump_int("wr", sum
.num_wr
);
847 f
->dump_int("wr_bytes", sum
.num_wr_kb
* 1024ull);
848 f
->dump_int("raw_bytes_used", sum
.num_bytes
* raw_used_rate
* curr_object_copies_rate
);
851 tbl
<< stringify(si_t(sum
.num_bytes
));
852 tbl
<< percentify(used
*100);
854 tbl
<< sum
.num_objects
;
856 tbl
<< stringify(si_t(sum
.num_objects_dirty
))
857 << stringify(si_t(sum
.num_rd
))
858 << stringify(si_t(sum
.num_wr
))
859 << stringify(si_t(sum
.num_bytes
* raw_used_rate
* curr_object_copies_rate
));
864 int64_t PGMapDigest::get_pool_free_space(const OSDMap
&osd_map
,
865 int64_t poolid
) const
867 const pg_pool_t
*pool
= osd_map
.get_pg_pool(poolid
);
868 int ruleno
= osd_map
.crush
->find_rule(pool
->get_crush_rule(),
872 avail
= get_rule_avail(ruleno
);
876 return avail
/ ::pool_raw_used_rate(osd_map
, poolid
);
879 int64_t PGMap::get_rule_avail(const OSDMap
& osdmap
, int ruleno
) const
882 int r
= osdmap
.crush
->get_rule_weight_osd_map(ruleno
, &wm
);
891 if (osdmap
.require_osd_release
>= CEPH_RELEASE_LUMINOUS
&&
892 osdmap
.get_full_ratio() > 0) {
893 fratio
= osdmap
.get_full_ratio();
895 fratio
= get_fallback_full_ratio();
899 for (auto p
= wm
.begin(); p
!= wm
.end(); ++p
) {
900 auto osd_info
= osd_stat
.find(p
->first
);
901 if (osd_info
!= osd_stat
.end()) {
902 if (osd_info
->second
.kb
== 0 || p
->second
== 0) {
903 // osd must be out, hence its stats have been zeroed
904 // (unless we somehow managed to have a disk with size 0...)
906 // (p->second == 0), if osd weight is 0, no need to
907 // calculate proj below.
910 double unusable
= (double)osd_info
->second
.kb
*
912 double avail
= MAX(0.0, (double)osd_info
->second
.kb_avail
- unusable
);
914 int64_t proj
= (int64_t)(avail
/ (double)p
->second
);
915 if (min
< 0 || proj
< min
) {
919 dout(0) << "Cannot get stat of OSD " << p
->first
<< dendl
;
925 void PGMap::get_rules_avail(const OSDMap
& osdmap
,
926 std::map
<int,int64_t> *avail_map
) const
929 for (auto p
: osdmap
.get_pools()) {
930 int64_t pool_id
= p
.first
;
931 if ((pool_id
< 0) || (pg_pool_sum
.count(pool_id
) == 0))
933 const pg_pool_t
*pool
= osdmap
.get_pg_pool(pool_id
);
934 int ruleno
= osdmap
.crush
->find_rule(pool
->get_crush_rule(),
937 if (avail_map
->count(ruleno
) == 0)
938 (*avail_map
)[ruleno
] = get_rule_avail(osdmap
, ruleno
);
942 // ---------------------
945 void PGMap::Incremental::encode(bufferlist
&bl
, uint64_t features
) const
947 if ((features
& CEPH_FEATURE_MONENC
) == 0) {
950 ::encode(version
, bl
);
951 ::encode(pg_stat_updates
, bl
);
952 ::encode(osd_stat_updates
, bl
);
953 ::encode(osd_stat_rm
, bl
);
954 ::encode(osdmap_epoch
, bl
);
955 ::encode(pg_scan
, bl
);
956 ::encode(full_ratio
, bl
);
957 ::encode(nearfull_ratio
, bl
);
958 ::encode(pg_remove
, bl
);
962 ENCODE_START(7, 5, bl
);
963 ::encode(version
, bl
);
964 ::encode(pg_stat_updates
, bl
);
965 ::encode(osd_stat_updates
, bl
);
966 ::encode(osd_stat_rm
, bl
);
967 ::encode(osdmap_epoch
, bl
);
968 ::encode(pg_scan
, bl
);
969 ::encode(full_ratio
, bl
);
970 ::encode(nearfull_ratio
, bl
);
971 ::encode(pg_remove
, bl
);
973 ::encode(osd_epochs
, bl
);
977 void PGMap::Incremental::decode(bufferlist::iterator
&bl
)
979 DECODE_START_LEGACY_COMPAT_LEN(7, 5, 5, bl
);
980 ::decode(version
, bl
);
982 pg_stat_updates
.clear();
989 ::decode(pg_stat_updates
[pgid
], bl
);
992 ::decode(pg_stat_updates
, bl
);
994 ::decode(osd_stat_updates
, bl
);
995 ::decode(osd_stat_rm
, bl
);
996 ::decode(osdmap_epoch
, bl
);
997 ::decode(pg_scan
, bl
);
999 ::decode(full_ratio
, bl
);
1000 ::decode(nearfull_ratio
, bl
);
1008 ::decode(opgid
, bl
);
1009 pg_remove
.insert(pg_t(opgid
));
1012 ::decode(pg_remove
, bl
);
1014 if (struct_v
< 4 && full_ratio
== 0) {
1017 if (struct_v
< 4 && nearfull_ratio
== 0) {
1018 nearfull_ratio
= -1;
1021 ::decode(stamp
, bl
);
1022 if (struct_v
>= 7) {
1023 ::decode(osd_epochs
, bl
);
1025 for (auto i
= osd_stat_updates
.begin();
1026 i
!= osd_stat_updates
.end();
1028 // This isn't accurate, but will cause trimming to behave like
1030 osd_epochs
.insert(make_pair(i
->first
, osdmap_epoch
));
1036 void PGMap::Incremental::dump(Formatter
*f
) const
1038 f
->dump_unsigned("version", version
);
1039 f
->dump_stream("stamp") << stamp
;
1040 f
->dump_unsigned("osdmap_epoch", osdmap_epoch
);
1041 f
->dump_unsigned("pg_scan_epoch", pg_scan
);
1042 f
->dump_float("full_ratio", full_ratio
);
1043 f
->dump_float("nearfull_ratio", nearfull_ratio
);
1045 f
->open_array_section("pg_stat_updates");
1046 for (auto p
= pg_stat_updates
.begin(); p
!= pg_stat_updates
.end(); ++p
) {
1047 f
->open_object_section("pg_stat");
1048 f
->dump_stream("pgid") << p
->first
;
1054 f
->open_array_section("osd_stat_updates");
1055 for (auto p
= osd_stat_updates
.begin(); p
!= osd_stat_updates
.end(); ++p
) {
1056 f
->open_object_section("osd_stat");
1057 f
->dump_int("osd", p
->first
);
1063 f
->open_array_section("osd_stat_removals");
1064 for (auto p
= osd_stat_rm
.begin(); p
!= osd_stat_rm
.end(); ++p
)
1065 f
->dump_int("osd", *p
);
1068 f
->open_array_section("pg_removals");
1069 for (auto p
= pg_remove
.begin(); p
!= pg_remove
.end(); ++p
)
1070 f
->dump_stream("pgid") << *p
;
1074 void PGMap::Incremental::generate_test_instances(list
<PGMap::Incremental
*>& o
)
1076 o
.push_back(new Incremental
);
1077 o
.push_back(new Incremental
);
1078 o
.back()->version
= 1;
1079 o
.back()->stamp
= utime_t(123,345);
1080 o
.push_back(new Incremental
);
1081 o
.back()->version
= 2;
1082 o
.back()->pg_stat_updates
[pg_t(1,2,3)] = pg_stat_t();
1083 o
.back()->osd_stat_updates
[5] = osd_stat_t();
1084 o
.back()->osd_epochs
[5] = 12;
1085 o
.push_back(new Incremental
);
1086 o
.back()->version
= 3;
1087 o
.back()->osdmap_epoch
= 1;
1088 o
.back()->pg_scan
= 2;
1089 o
.back()->full_ratio
= .2;
1090 o
.back()->nearfull_ratio
= .3;
1091 o
.back()->pg_stat_updates
[pg_t(4,5,6)] = pg_stat_t();
1092 o
.back()->osd_stat_updates
[6] = osd_stat_t();
1093 o
.back()->osd_epochs
[6] = 12;
1094 o
.back()->pg_remove
.insert(pg_t(1,2,3));
1095 o
.back()->osd_stat_rm
.insert(5);
1101 void PGMap::apply_incremental(CephContext
*cct
, const Incremental
& inc
)
1103 assert(inc
.version
== version
+1);
1107 delta_t
= inc
.stamp
;
1111 pool_stat_t pg_sum_old
= pg_sum
;
1112 mempool::pgmap::unordered_map
<uint64_t, pool_stat_t
> pg_pool_sum_old
;
1114 bool ratios_changed
= false;
1115 if (inc
.full_ratio
!= full_ratio
&& inc
.full_ratio
!= -1) {
1116 full_ratio
= inc
.full_ratio
;
1117 ratios_changed
= true;
1119 if (inc
.nearfull_ratio
!= nearfull_ratio
&& inc
.nearfull_ratio
!= -1) {
1120 nearfull_ratio
= inc
.nearfull_ratio
;
1121 ratios_changed
= true;
1126 for (auto p
= inc
.pg_stat_updates
.begin();
1127 p
!= inc
.pg_stat_updates
.end();
1129 const pg_t
&update_pg(p
->first
);
1130 const pg_stat_t
&update_stat(p
->second
);
1132 if (pg_pool_sum_old
.count(update_pg
.pool()) == 0)
1133 pg_pool_sum_old
[update_pg
.pool()] = pg_pool_sum
[update_pg
.pool()];
1135 auto t
= pg_stat
.find(update_pg
);
1136 if (t
== pg_stat
.end()) {
1137 pg_stat
.insert(make_pair(update_pg
, update_stat
));
1139 stat_pg_sub(update_pg
, t
->second
);
1140 t
->second
= update_stat
;
1142 stat_pg_add(update_pg
, update_stat
);
1144 assert(osd_stat
.size() == osd_epochs
.size());
1145 for (auto p
= inc
.get_osd_stat_updates().begin();
1146 p
!= inc
.get_osd_stat_updates().end();
1149 const osd_stat_t
&new_stats(p
->second
);
1151 auto t
= osd_stat
.find(osd
);
1152 if (t
== osd_stat
.end()) {
1153 osd_stat
.insert(make_pair(osd
, new_stats
));
1155 stat_osd_sub(t
->first
, t
->second
);
1156 t
->second
= new_stats
;
1158 auto i
= osd_epochs
.find(osd
);
1159 auto j
= inc
.get_osd_epochs().find(osd
);
1160 assert(j
!= inc
.get_osd_epochs().end());
1162 if (i
== osd_epochs
.end())
1163 osd_epochs
.insert(*j
);
1165 i
->second
= j
->second
;
1167 stat_osd_add(osd
, new_stats
);
1169 // adjust [near]full status
1170 register_nearfull_status(osd
, new_stats
);
1172 set
<int64_t> deleted_pools
;
1173 for (auto p
= inc
.pg_remove
.begin();
1174 p
!= inc
.pg_remove
.end();
1176 const pg_t
&removed_pg(*p
);
1177 auto s
= pg_stat
.find(removed_pg
);
1178 if (s
!= pg_stat
.end()) {
1179 stat_pg_sub(removed_pg
, s
->second
);
1182 deleted_pools
.insert(removed_pg
.pool());
1185 for (auto p
= inc
.get_osd_stat_rm().begin();
1186 p
!= inc
.get_osd_stat_rm().end();
1188 auto t
= osd_stat
.find(*p
);
1189 if (t
!= osd_stat
.end()) {
1190 stat_osd_sub(t
->first
, t
->second
);
1192 osd_epochs
.erase(*p
);
1195 // remove these old osds from full/nearfull set(s), too
1196 nearfull_osds
.erase(*p
);
1197 full_osds
.erase(*p
);
1200 // calculate a delta, and average over the last 2 deltas.
1201 pool_stat_t d
= pg_sum
;
1202 d
.stats
.sub(pg_sum_old
.stats
);
1203 pg_sum_deltas
.push_back(make_pair(d
, delta_t
));
1204 stamp_delta
+= delta_t
;
1206 pg_sum_delta
.stats
.add(d
.stats
);
1207 if (pg_sum_deltas
.size() > (unsigned)MAX(1, cct
? cct
->_conf
->mon_stat_smooth_intervals
: 1)) {
1208 pg_sum_delta
.stats
.sub(pg_sum_deltas
.front().first
.stats
);
1209 stamp_delta
-= pg_sum_deltas
.front().second
;
1210 pg_sum_deltas
.pop_front();
1213 update_pool_deltas(cct
, inc
.stamp
, pg_pool_sum_old
);
1215 for (auto p
: deleted_pools
) {
1217 dout(20) << " deleted pool " << p
<< dendl
;
1221 if (inc
.osdmap_epoch
)
1222 last_osdmap_epoch
= inc
.osdmap_epoch
;
1224 last_pg_scan
= inc
.pg_scan
;
1226 min_last_epoch_clean
= 0; // invalidate
1229 void PGMap::redo_full_sets()
1232 nearfull_osds
.clear();
1233 for (auto i
= osd_stat
.begin();
1234 i
!= osd_stat
.end();
1236 register_nearfull_status(i
->first
, i
->second
);
1240 void PGMap::register_nearfull_status(int osd
, const osd_stat_t
& s
)
1242 float ratio
= ((float)s
.kb_used
) / ((float)s
.kb
);
1244 if (full_ratio
> 0 && ratio
> full_ratio
) {
1246 full_osds
.insert(osd
);
1247 nearfull_osds
.erase(osd
);
1248 } else if (nearfull_ratio
> 0 && ratio
> nearfull_ratio
) {
1250 full_osds
.erase(osd
);
1251 nearfull_osds
.insert(osd
);
1254 full_osds
.erase(osd
);
1255 nearfull_osds
.erase(osd
);
1259 void PGMap::calc_stats()
1265 pg_pool_sum
.clear();
1266 num_pg_by_pool
.clear();
1268 pg_sum
= pool_stat_t();
1269 osd_sum
= osd_stat_t();
1270 num_pg_by_state
.clear();
1271 num_pg_by_osd
.clear();
1273 for (auto p
= pg_stat
.begin();
1276 stat_pg_add(p
->first
, p
->second
);
1278 for (auto p
= osd_stat
.begin();
1279 p
!= osd_stat
.end();
1281 stat_osd_add(p
->first
, p
->second
);
1285 min_last_epoch_clean
= calc_min_last_epoch_clean();
1288 void PGMap::update_pg(pg_t pgid
, bufferlist
& bl
)
1290 bufferlist::iterator p
= bl
.begin();
1291 auto s
= pg_stat
.find(pgid
);
1292 epoch_t old_lec
= 0, lec
;
1293 if (s
!= pg_stat
.end()) {
1294 old_lec
= s
->second
.get_effective_last_epoch_clean();
1295 stat_pg_update(pgid
, s
->second
, p
);
1296 lec
= s
->second
.get_effective_last_epoch_clean();
1298 pg_stat_t
& r
= pg_stat
[pgid
];
1300 stat_pg_add(pgid
, r
);
1301 lec
= r
.get_effective_last_epoch_clean();
1304 if (min_last_epoch_clean
&&
1305 (lec
< min_last_epoch_clean
|| // we did
1306 (lec
> min_last_epoch_clean
&& // we might
1307 old_lec
== min_last_epoch_clean
)
1309 min_last_epoch_clean
= 0;
1312 void PGMap::remove_pg(pg_t pgid
)
1314 auto s
= pg_stat
.find(pgid
);
1315 if (s
!= pg_stat
.end()) {
1316 if (min_last_epoch_clean
&&
1317 s
->second
.get_effective_last_epoch_clean() == min_last_epoch_clean
)
1318 min_last_epoch_clean
= 0;
1319 stat_pg_sub(pgid
, s
->second
);
1324 void PGMap::update_osd(int osd
, bufferlist
& bl
)
1326 bufferlist::iterator p
= bl
.begin();
1327 auto o
= osd_stat
.find(osd
);
1328 epoch_t old_lec
= 0;
1329 if (o
!= osd_stat
.end()) {
1330 auto i
= osd_epochs
.find(osd
);
1331 if (i
!= osd_epochs
.end())
1332 old_lec
= i
->second
;
1333 stat_osd_sub(osd
, o
->second
);
1335 osd_stat_t
& r
= osd_stat
[osd
];
1337 stat_osd_add(osd
, r
);
1339 // adjust [near]full status
1340 register_nearfull_status(osd
, r
);
1347 if (e
< min_last_epoch_clean
||
1348 (e
> min_last_epoch_clean
&&
1349 old_lec
== min_last_epoch_clean
))
1350 min_last_epoch_clean
= 0;
1352 // WARNING: we are not refreshing min_last_epoch_clean! must be old store
1353 // or old mon running.
1357 void PGMap::remove_osd(int osd
)
1359 auto o
= osd_stat
.find(osd
);
1360 if (o
!= osd_stat
.end()) {
1361 stat_osd_sub(osd
, o
->second
);
1364 // remove these old osds from full/nearfull set(s), too
1365 nearfull_osds
.erase(osd
);
1366 full_osds
.erase(osd
);
1370 void PGMap::stat_pg_add(const pg_t
&pgid
, const pg_stat_t
&s
,
1373 pg_pool_sum
[pgid
.pool()].add(s
);
1377 num_pg_by_state
[s
.state
]++;
1378 num_pg_by_pool
[pgid
.pool()]++;
1380 if ((s
.state
& PG_STATE_CREATING
) &&
1381 s
.parent_split_bits
== 0) {
1382 creating_pgs
.insert(pgid
);
1383 if (s
.acting_primary
>= 0) {
1384 creating_pgs_by_osd_epoch
[s
.acting_primary
][s
.mapping_epoch
].insert(pgid
);
1388 if (s
.state
& PG_STATE_ACTIVE
) {
1398 for (auto p
= s
.blocked_by
.begin();
1399 p
!= s
.blocked_by
.end();
1401 ++blocked_by_sum
[*p
];
1404 for (auto p
= s
.acting
.begin(); p
!= s
.acting
.end(); ++p
) {
1405 pg_by_osd
[*p
].insert(pgid
);
1406 num_pg_by_osd
[*p
].acting
++;
1408 for (auto p
= s
.up
.begin(); p
!= s
.up
.end(); ++p
) {
1409 pg_by_osd
[*p
].insert(pgid
);
1410 num_pg_by_osd
[*p
].up
++;
1413 if (s
.up_primary
>= 0) {
1414 num_pg_by_osd
[s
.up_primary
].primary
++;
1418 void PGMap::stat_pg_sub(const pg_t
&pgid
, const pg_stat_t
&s
,
1421 pool_stat_t
& ps
= pg_pool_sum
[pgid
.pool()];
1426 int end
= --num_pg_by_state
[s
.state
];
1429 num_pg_by_state
.erase(s
.state
);
1430 end
= --num_pg_by_pool
[pgid
.pool()];
1432 num_pg_by_pool
.erase(pgid
.pool());
1433 pg_pool_sum
.erase(pgid
.pool());
1436 if ((s
.state
& PG_STATE_CREATING
) &&
1437 s
.parent_split_bits
== 0) {
1438 creating_pgs
.erase(pgid
);
1439 if (s
.acting_primary
>= 0) {
1440 map
<epoch_t
,set
<pg_t
> >& r
= creating_pgs_by_osd_epoch
[s
.acting_primary
];
1441 r
[s
.mapping_epoch
].erase(pgid
);
1442 if (r
[s
.mapping_epoch
].empty())
1443 r
.erase(s
.mapping_epoch
);
1445 creating_pgs_by_osd_epoch
.erase(s
.acting_primary
);
1449 if (s
.state
& PG_STATE_ACTIVE
) {
1459 for (auto p
= s
.blocked_by
.begin();
1460 p
!= s
.blocked_by
.end();
1462 auto q
= blocked_by_sum
.find(*p
);
1463 assert(q
!= blocked_by_sum
.end());
1466 blocked_by_sum
.erase(q
);
1469 for (auto p
= s
.acting
.begin(); p
!= s
.acting
.end(); ++p
) {
1470 auto& oset
= pg_by_osd
[*p
];
1473 pg_by_osd
.erase(*p
);
1474 auto it
= num_pg_by_osd
.find(*p
);
1475 if (it
!= num_pg_by_osd
.end() && it
->second
.acting
> 0)
1476 it
->second
.acting
--;
1478 for (auto p
= s
.up
.begin(); p
!= s
.up
.end(); ++p
) {
1479 auto& oset
= pg_by_osd
[*p
];
1482 pg_by_osd
.erase(*p
);
1483 auto it
= num_pg_by_osd
.find(*p
);
1484 if (it
!= num_pg_by_osd
.end() && it
->second
.up
> 0)
1488 if (s
.up_primary
>= 0) {
1489 auto it
= num_pg_by_osd
.find(s
.up_primary
);
1490 if (it
!= num_pg_by_osd
.end() && it
->second
.primary
> 0)
1491 it
->second
.primary
--;
1495 void PGMap::stat_pg_update(const pg_t pgid
, pg_stat_t
& s
,
1496 bufferlist::iterator
& blp
)
1502 s
.acting
== n
.acting
&&
1504 s
.blocked_by
== n
.blocked_by
;
1506 stat_pg_sub(pgid
, s
, sameosds
);
1508 // if acting_primary has shift to an just restored osd, and pg yet to finish
1509 // peering, many attributes in current stats remain stale. others seem don't
1510 // mater much while faulty last_active will make "pg stuck in" check unhappy.
1511 if (!(n
.state
& (PG_STATE_ACTIVE
| PG_STATE_PEERED
)) &&
1512 n
.last_active
< s
.last_active
)
1513 n
.last_active
= s
.last_active
;
1515 stat_pg_add(pgid
, n
, sameosds
);
1518 void PGMap::stat_osd_add(int osd
, const osd_stat_t
&s
)
1522 if (osd
>= (int)osd_last_seq
.size()) {
1523 osd_last_seq
.resize(osd
+ 1);
1525 osd_last_seq
[osd
] = s
.seq
;
1528 void PGMap::stat_osd_sub(int osd
, const osd_stat_t
&s
)
1532 assert(osd
< (int)osd_last_seq
.size());
1533 osd_last_seq
[osd
] = 0;
1536 epoch_t
PGMap::calc_min_last_epoch_clean() const
1538 if (pg_stat
.empty())
1541 auto p
= pg_stat
.begin();
1542 epoch_t min
= p
->second
.get_effective_last_epoch_clean();
1543 for (++p
; p
!= pg_stat
.end(); ++p
) {
1544 epoch_t lec
= p
->second
.get_effective_last_epoch_clean();
1548 // also scan osd epochs
1549 // don't trim past the oldest reported osd epoch
1550 for (auto i
= osd_epochs
.begin();
1551 i
!= osd_epochs
.end();
1553 if (i
->second
< min
)
1559 void PGMap::encode_digest(const OSDMap
& osdmap
,
1560 bufferlist
& bl
, uint64_t features
) const
1562 get_rules_avail(osdmap
, &avail_space_by_rule
);
1563 PGMapDigest::encode(bl
, features
);
1566 void PGMap::encode(bufferlist
&bl
, uint64_t features
) const
1568 if ((features
& CEPH_FEATURE_MONENC
) == 0) {
1571 ::encode(version
, bl
);
1572 ::encode(pg_stat
, bl
);
1573 ::encode(osd_stat
, bl
);
1574 ::encode(last_osdmap_epoch
, bl
);
1575 ::encode(last_pg_scan
, bl
);
1576 ::encode(full_ratio
, bl
);
1577 ::encode(nearfull_ratio
, bl
);
1581 ENCODE_START(6, 4, bl
);
1582 ::encode(version
, bl
);
1583 ::encode(pg_stat
, bl
);
1584 ::encode(osd_stat
, bl
);
1585 ::encode(last_osdmap_epoch
, bl
);
1586 ::encode(last_pg_scan
, bl
);
1587 ::encode(full_ratio
, bl
);
1588 ::encode(nearfull_ratio
, bl
);
1589 ::encode(stamp
, bl
);
1590 ::encode(osd_epochs
, bl
);
1594 void PGMap::decode(bufferlist::iterator
&bl
)
1596 DECODE_START_LEGACY_COMPAT_LEN(6, 4, 4, bl
);
1597 ::decode(version
, bl
);
1604 ::decode(opgid
, bl
);
1606 ::decode(pg_stat
[pgid
], bl
);
1609 ::decode(pg_stat
, bl
);
1611 ::decode(osd_stat
, bl
);
1612 ::decode(last_osdmap_epoch
, bl
);
1613 ::decode(last_pg_scan
, bl
);
1614 if (struct_v
>= 2) {
1615 ::decode(full_ratio
, bl
);
1616 ::decode(nearfull_ratio
, bl
);
1619 ::decode(stamp
, bl
);
1620 if (struct_v
>= 6) {
1621 ::decode(osd_epochs
, bl
);
1623 for (auto i
= osd_stat
.begin();
1624 i
!= osd_stat
.end();
1626 // This isn't accurate, but will cause trimming to behave like
1628 osd_epochs
.insert(make_pair(i
->first
, last_osdmap_epoch
));
1636 void PGMap::dirty_all(Incremental
& inc
)
1638 inc
.osdmap_epoch
= last_osdmap_epoch
;
1639 inc
.pg_scan
= last_pg_scan
;
1640 inc
.full_ratio
= full_ratio
;
1641 inc
.nearfull_ratio
= nearfull_ratio
;
1643 for (auto p
= pg_stat
.begin(); p
!= pg_stat
.end(); ++p
) {
1644 inc
.pg_stat_updates
[p
->first
] = p
->second
;
1646 for (auto p
= osd_stat
.begin(); p
!= osd_stat
.end(); ++p
) {
1647 assert(osd_epochs
.count(p
->first
));
1648 inc
.update_stat(p
->first
,
1649 inc
.get_osd_epochs().find(p
->first
)->second
,
1654 void PGMap::dump(Formatter
*f
) const
1657 dump_pg_stats(f
, false);
1662 void PGMap::dump_basic(Formatter
*f
) const
1664 f
->dump_unsigned("version", version
);
1665 f
->dump_stream("stamp") << stamp
;
1666 f
->dump_unsigned("last_osdmap_epoch", last_osdmap_epoch
);
1667 f
->dump_unsigned("last_pg_scan", last_pg_scan
);
1668 f
->dump_unsigned("min_last_epoch_clean", min_last_epoch_clean
);
1669 f
->dump_float("full_ratio", full_ratio
);
1670 f
->dump_float("near_full_ratio", nearfull_ratio
);
1672 f
->open_object_section("pg_stats_sum");
1676 f
->open_object_section("osd_stats_sum");
1680 f
->open_array_section("osd_epochs");
1681 for (auto p
= osd_epochs
.begin(); p
!= osd_epochs
.end(); ++p
) {
1682 f
->open_object_section("osd");
1683 f
->dump_unsigned("osd", p
->first
);
1684 f
->dump_unsigned("epoch", p
->second
);
1692 void PGMap::dump_delta(Formatter
*f
) const
1694 f
->open_object_section("pg_stats_delta");
1695 pg_sum_delta
.dump(f
);
1699 void PGMap::dump_pg_stats(Formatter
*f
, bool brief
) const
1701 f
->open_array_section("pg_stats");
1702 for (auto i
= pg_stat
.begin();
1705 f
->open_object_section("pg_stat");
1706 f
->dump_stream("pgid") << i
->first
;
1708 i
->second
.dump_brief(f
);
1716 void PGMap::dump_pool_stats(Formatter
*f
) const
1718 f
->open_array_section("pool_stats");
1719 for (auto p
= pg_pool_sum
.begin();
1720 p
!= pg_pool_sum
.end();
1722 f
->open_object_section("pool_stat");
1723 f
->dump_int("poolid", p
->first
);
1724 auto q
= num_pg_by_pool
.find(p
->first
);
1725 if (q
!= num_pg_by_pool
.end())
1726 f
->dump_unsigned("num_pg", q
->second
);
1733 void PGMap::dump_osd_stats(Formatter
*f
) const
1735 f
->open_array_section("osd_stats");
1736 for (auto q
= osd_stat
.begin();
1737 q
!= osd_stat
.end();
1739 f
->open_object_section("osd_stat");
1740 f
->dump_int("osd", q
->first
);
1747 void PGMap::dump_pg_stats_plain(
1749 const mempool::pgmap::unordered_map
<pg_t
, pg_stat_t
>& pg_stats
,
1755 tab
.define_column("PG_STAT", TextTable::LEFT
, TextTable::LEFT
);
1756 tab
.define_column("STATE", TextTable::LEFT
, TextTable::RIGHT
);
1757 tab
.define_column("UP", TextTable::LEFT
, TextTable::RIGHT
);
1758 tab
.define_column("UP_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1759 tab
.define_column("ACTING", TextTable::LEFT
, TextTable::RIGHT
);
1760 tab
.define_column("ACTING_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1763 tab
.define_column("PG_STAT", TextTable::LEFT
, TextTable::LEFT
);
1764 tab
.define_column("OBJECTS", TextTable::LEFT
, TextTable::RIGHT
);
1765 tab
.define_column("MISSING_ON_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1766 tab
.define_column("DEGRADED", TextTable::LEFT
, TextTable::RIGHT
);
1767 tab
.define_column("MISPLACED", TextTable::LEFT
, TextTable::RIGHT
);
1768 tab
.define_column("UNFOUND", TextTable::LEFT
, TextTable::RIGHT
);
1769 tab
.define_column("BYTES", TextTable::LEFT
, TextTable::RIGHT
);
1770 tab
.define_column("LOG", TextTable::LEFT
, TextTable::RIGHT
);
1771 tab
.define_column("DISK_LOG", TextTable::LEFT
, TextTable::RIGHT
);
1772 tab
.define_column("STATE", TextTable::LEFT
, TextTable::RIGHT
);
1773 tab
.define_column("STATE_STAMP", TextTable::LEFT
, TextTable::RIGHT
);
1774 tab
.define_column("VERSION", TextTable::LEFT
, TextTable::RIGHT
);
1775 tab
.define_column("REPORTED", TextTable::LEFT
, TextTable::RIGHT
);
1776 tab
.define_column("UP", TextTable::LEFT
, TextTable::RIGHT
);
1777 tab
.define_column("UP_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1778 tab
.define_column("ACTING", TextTable::LEFT
, TextTable::RIGHT
);
1779 tab
.define_column("ACTING_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1780 tab
.define_column("LAST_SCRUB", TextTable::LEFT
, TextTable::RIGHT
);
1781 tab
.define_column("SCRUB_STAMP", TextTable::LEFT
, TextTable::RIGHT
);
1782 tab
.define_column("LAST_DEEP_SCRUB", TextTable::LEFT
, TextTable::RIGHT
);
1783 tab
.define_column("DEEP_SCRUB_STAMP", TextTable::LEFT
, TextTable::RIGHT
);
1786 for (auto i
= pg_stats
.begin();
1787 i
!= pg_stats
.end(); ++i
) {
1788 const pg_stat_t
&st(i
->second
);
1791 << pg_state_string(st
.state
)
1795 << st
.acting_primary
1796 << TextTable::endrow
;
1798 ostringstream reported
;
1799 reported
<< st
.reported_epoch
<< ":" << st
.reported_seq
;
1802 << st
.stats
.sum
.num_objects
1803 << st
.stats
.sum
.num_objects_missing_on_primary
1804 << st
.stats
.sum
.num_objects_degraded
1805 << st
.stats
.sum
.num_objects_misplaced
1806 << st
.stats
.sum
.num_objects_unfound
1807 << st
.stats
.sum
.num_bytes
1809 << st
.ondisk_log_size
1810 << pg_state_string(st
.state
)
1814 << pg_vector_string(st
.up
)
1816 << pg_vector_string(st
.acting
)
1817 << st
.acting_primary
1819 << st
.last_scrub_stamp
1820 << st
.last_deep_scrub
1821 << st
.last_deep_scrub_stamp
1822 << TextTable::endrow
;
1829 void PGMap::dump(ostream
& ss
) const
1832 dump_pg_stats(ss
, false);
1833 dump_pool_stats(ss
, false);
1834 dump_pg_sum_stats(ss
, false);
1838 void PGMap::dump_basic(ostream
& ss
) const
1840 ss
<< "version " << version
<< std::endl
;
1841 ss
<< "stamp " << stamp
<< std::endl
;
1842 ss
<< "last_osdmap_epoch " << last_osdmap_epoch
<< std::endl
;
1843 ss
<< "last_pg_scan " << last_pg_scan
<< std::endl
;
1844 ss
<< "full_ratio " << full_ratio
<< std::endl
;
1845 ss
<< "nearfull_ratio " << nearfull_ratio
<< std::endl
;
1848 void PGMap::dump_pg_stats(ostream
& ss
, bool brief
) const
1850 dump_pg_stats_plain(ss
, pg_stat
, brief
);
1853 void PGMap::dump_pool_stats(ostream
& ss
, bool header
) const
1858 tab
.define_column("POOLID", TextTable::LEFT
, TextTable::LEFT
);
1859 tab
.define_column("OBJECTS", TextTable::LEFT
, TextTable::RIGHT
);
1860 tab
.define_column("MISSING_ON_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1861 tab
.define_column("DEGRADED", TextTable::LEFT
, TextTable::RIGHT
);
1862 tab
.define_column("MISPLACED", TextTable::LEFT
, TextTable::RIGHT
);
1863 tab
.define_column("UNFOUND", TextTable::LEFT
, TextTable::RIGHT
);
1864 tab
.define_column("BYTES", TextTable::LEFT
, TextTable::RIGHT
);
1865 tab
.define_column("LOG", TextTable::LEFT
, TextTable::RIGHT
);
1866 tab
.define_column("DISK_LOG", TextTable::LEFT
, TextTable::RIGHT
);
1868 tab
.define_column("", TextTable::LEFT
, TextTable::LEFT
);
1869 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1870 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1871 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1872 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1873 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1874 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1875 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1876 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1879 for (auto p
= pg_pool_sum
.begin();
1880 p
!= pg_pool_sum
.end();
1883 << p
->second
.stats
.sum
.num_objects
1884 << p
->second
.stats
.sum
.num_objects_missing_on_primary
1885 << p
->second
.stats
.sum
.num_objects_degraded
1886 << p
->second
.stats
.sum
.num_objects_misplaced
1887 << p
->second
.stats
.sum
.num_objects_unfound
1888 << p
->second
.stats
.sum
.num_bytes
1889 << p
->second
.log_size
1890 << p
->second
.ondisk_log_size
1891 << TextTable::endrow
;
1897 void PGMap::dump_pg_sum_stats(ostream
& ss
, bool header
) const
1902 tab
.define_column("PG_STAT", TextTable::LEFT
, TextTable::LEFT
);
1903 tab
.define_column("OBJECTS", TextTable::LEFT
, TextTable::RIGHT
);
1904 tab
.define_column("MISSING_ON_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1905 tab
.define_column("DEGRADED", TextTable::LEFT
, TextTable::RIGHT
);
1906 tab
.define_column("MISPLACED", TextTable::LEFT
, TextTable::RIGHT
);
1907 tab
.define_column("UNFOUND", TextTable::LEFT
, TextTable::RIGHT
);
1908 tab
.define_column("BYTES", TextTable::LEFT
, TextTable::RIGHT
);
1909 tab
.define_column("LOG", TextTable::LEFT
, TextTable::RIGHT
);
1910 tab
.define_column("DISK_LOG", TextTable::LEFT
, TextTable::RIGHT
);
1912 tab
.define_column("", TextTable::LEFT
, TextTable::LEFT
);
1913 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1914 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1915 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1916 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1917 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1918 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1919 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1920 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1924 << pg_sum
.stats
.sum
.num_objects
1925 << pg_sum
.stats
.sum
.num_objects_missing_on_primary
1926 << pg_sum
.stats
.sum
.num_objects_degraded
1927 << pg_sum
.stats
.sum
.num_objects_misplaced
1928 << pg_sum
.stats
.sum
.num_objects_unfound
1929 << pg_sum
.stats
.sum
.num_bytes
1931 << pg_sum
.ondisk_log_size
1932 << TextTable::endrow
;
1937 void PGMap::dump_osd_stats(ostream
& ss
) const
1941 tab
.define_column("OSD_STAT", TextTable::LEFT
, TextTable::LEFT
);
1942 tab
.define_column("USED", TextTable::LEFT
, TextTable::RIGHT
);
1943 tab
.define_column("AVAIL", TextTable::LEFT
, TextTable::RIGHT
);
1944 tab
.define_column("TOTAL", TextTable::LEFT
, TextTable::RIGHT
);
1945 tab
.define_column("HB_PEERS", TextTable::LEFT
, TextTable::RIGHT
);
1946 tab
.define_column("PG_SUM", TextTable::LEFT
, TextTable::RIGHT
);
1947 tab
.define_column("PRIMARY_PG_SUM", TextTable::LEFT
, TextTable::RIGHT
);
1949 for (auto p
= osd_stat
.begin();
1950 p
!= osd_stat
.end();
1953 << si_t(p
->second
.kb_used
<< 10)
1954 << si_t(p
->second
.kb_avail
<< 10)
1955 << si_t(p
->second
.kb
<< 10)
1956 << p
->second
.hb_peers
1957 << get_num_pg_by_osd(p
->first
)
1958 << get_num_primary_pg_by_osd(p
->first
)
1959 << TextTable::endrow
;
1963 << si_t(osd_sum
.kb_used
<< 10)
1964 << si_t(osd_sum
.kb_avail
<< 10)
1965 << si_t(osd_sum
.kb
<< 10)
1966 << TextTable::endrow
;
1971 void PGMap::dump_osd_sum_stats(ostream
& ss
) const
1975 tab
.define_column("OSD_STAT", TextTable::LEFT
, TextTable::LEFT
);
1976 tab
.define_column("USED", TextTable::LEFT
, TextTable::RIGHT
);
1977 tab
.define_column("AVAIL", TextTable::LEFT
, TextTable::RIGHT
);
1978 tab
.define_column("TOTAL", TextTable::LEFT
, TextTable::RIGHT
);
1981 << si_t(osd_sum
.kb_used
<< 10)
1982 << si_t(osd_sum
.kb_avail
<< 10)
1983 << si_t(osd_sum
.kb
<< 10)
1984 << TextTable::endrow
;
1989 void PGMap::get_stuck_stats(
1990 int types
, const utime_t cutoff
,
1991 mempool::pgmap::unordered_map
<pg_t
, pg_stat_t
>& stuck_pgs
) const
1994 for (auto i
= pg_stat
.begin();
1997 utime_t val
= cutoff
; // don't care about >= cutoff so that is infinity
1999 if ((types
& STUCK_INACTIVE
) && !(i
->second
.state
& PG_STATE_ACTIVE
)) {
2000 if (i
->second
.last_active
< val
)
2001 val
= i
->second
.last_active
;
2004 if ((types
& STUCK_UNCLEAN
) && !(i
->second
.state
& PG_STATE_CLEAN
)) {
2005 if (i
->second
.last_clean
< val
)
2006 val
= i
->second
.last_clean
;
2009 if ((types
& STUCK_DEGRADED
) && (i
->second
.state
& PG_STATE_DEGRADED
)) {
2010 if (i
->second
.last_undegraded
< val
)
2011 val
= i
->second
.last_undegraded
;
2014 if ((types
& STUCK_UNDERSIZED
) && (i
->second
.state
& PG_STATE_UNDERSIZED
)) {
2015 if (i
->second
.last_fullsized
< val
)
2016 val
= i
->second
.last_fullsized
;
2019 if ((types
& STUCK_STALE
) && (i
->second
.state
& PG_STATE_STALE
)) {
2020 if (i
->second
.last_unstale
< val
)
2021 val
= i
->second
.last_unstale
;
2024 // val is now the earliest any of the requested stuck states began
2026 stuck_pgs
[i
->first
] = i
->second
;
2031 bool PGMap::get_stuck_counts(const utime_t cutoff
, map
<string
, int>& note
) const
2039 for (auto i
= pg_stat
.begin();
2042 if (! (i
->second
.state
& PG_STATE_ACTIVE
)) {
2043 if (i
->second
.last_active
< cutoff
)
2046 if (! (i
->second
.state
& PG_STATE_CLEAN
)) {
2047 if (i
->second
.last_clean
< cutoff
)
2050 if (i
->second
.state
& PG_STATE_DEGRADED
) {
2051 if (i
->second
.last_undegraded
< cutoff
)
2054 if (i
->second
.state
& PG_STATE_UNDERSIZED
) {
2055 if (i
->second
.last_fullsized
< cutoff
)
2058 if (i
->second
.state
& PG_STATE_STALE
) {
2059 if (i
->second
.last_unstale
< cutoff
)
2065 note
["stuck inactive"] = inactive
;
2068 note
["stuck unclean"] = unclean
;
2071 note
["stuck undersized"] = undersized
;
2074 note
["stuck degraded"] = degraded
;
2077 note
["stuck stale"] = stale
;
2079 return inactive
|| unclean
|| undersized
|| degraded
|| stale
;
2082 void PGMap::dump_stuck(Formatter
*f
, int types
, utime_t cutoff
) const
2084 mempool::pgmap::unordered_map
<pg_t
, pg_stat_t
> stuck_pg_stats
;
2085 get_stuck_stats(types
, cutoff
, stuck_pg_stats
);
2086 f
->open_array_section("stuck_pg_stats");
2087 for (auto i
= stuck_pg_stats
.begin();
2088 i
!= stuck_pg_stats
.end();
2090 f
->open_object_section("pg_stat");
2091 f
->dump_stream("pgid") << i
->first
;
2098 void PGMap::dump_stuck_plain(ostream
& ss
, int types
, utime_t cutoff
) const
2100 mempool::pgmap::unordered_map
<pg_t
, pg_stat_t
> stuck_pg_stats
;
2101 get_stuck_stats(types
, cutoff
, stuck_pg_stats
);
2102 if (!stuck_pg_stats
.empty())
2103 dump_pg_stats_plain(ss
, stuck_pg_stats
, true);
2106 int PGMap::dump_stuck_pg_stats(
2110 vector
<string
>& args
) const
2112 int stuck_types
= 0;
2114 for (auto i
= args
.begin(); i
!= args
.end(); ++i
) {
2115 if (*i
== "inactive")
2116 stuck_types
|= PGMap::STUCK_INACTIVE
;
2117 else if (*i
== "unclean")
2118 stuck_types
|= PGMap::STUCK_UNCLEAN
;
2119 else if (*i
== "undersized")
2120 stuck_types
|= PGMap::STUCK_UNDERSIZED
;
2121 else if (*i
== "degraded")
2122 stuck_types
|= PGMap::STUCK_DEGRADED
;
2123 else if (*i
== "stale")
2124 stuck_types
|= PGMap::STUCK_STALE
;
2126 ds
<< "Unknown type: " << *i
<< std::endl
;
2131 utime_t
now(ceph_clock_now());
2132 utime_t cutoff
= now
- utime_t(threshold
, 0);
2135 dump_stuck_plain(ds
, stuck_types
, cutoff
);
2137 dump_stuck(f
, stuck_types
, cutoff
);
2144 void PGMap::dump_osd_perf_stats(Formatter
*f
) const
2146 f
->open_array_section("osd_perf_infos");
2147 for (auto i
= osd_stat
.begin();
2148 i
!= osd_stat
.end();
2150 f
->open_object_section("osd");
2151 f
->dump_int("id", i
->first
);
2153 f
->open_object_section("perf_stats");
2154 i
->second
.os_perf_stat
.dump(f
);
2161 void PGMap::print_osd_perf_stats(std::ostream
*ss
) const
2164 tab
.define_column("osd", TextTable::LEFT
, TextTable::RIGHT
);
2165 tab
.define_column("commit_latency(ms)", TextTable::LEFT
, TextTable::RIGHT
);
2166 tab
.define_column("apply_latency(ms)", TextTable::LEFT
, TextTable::RIGHT
);
2167 for (auto i
= osd_stat
.begin();
2168 i
!= osd_stat
.end();
2171 tab
<< i
->second
.os_perf_stat
.os_commit_latency
;
2172 tab
<< i
->second
.os_perf_stat
.os_apply_latency
;
2173 tab
<< TextTable::endrow
;
2178 void PGMap::dump_osd_blocked_by_stats(Formatter
*f
) const
2180 f
->open_array_section("osd_blocked_by_infos");
2181 for (auto i
= blocked_by_sum
.begin();
2182 i
!= blocked_by_sum
.end();
2184 f
->open_object_section("osd");
2185 f
->dump_int("id", i
->first
);
2186 f
->dump_int("num_blocked", i
->second
);
2191 void PGMap::print_osd_blocked_by_stats(std::ostream
*ss
) const
2194 tab
.define_column("osd", TextTable::LEFT
, TextTable::RIGHT
);
2195 tab
.define_column("num_blocked", TextTable::LEFT
, TextTable::RIGHT
);
2196 for (auto i
= blocked_by_sum
.begin();
2197 i
!= blocked_by_sum
.end();
2201 tab
<< TextTable::endrow
;
2208 * update aggregated delta
2210 * @param cct ceph context
2211 * @param ts Timestamp for the stats being delta'ed
2212 * @param old_pool_sum Previous stats sum
2213 * @param last_ts Last timestamp for pool
2214 * @param result_pool_sum Resulting stats
2215 * @param result_pool_delta Resulting pool delta
2216 * @param result_ts_delta Resulting timestamp delta
2217 * @param delta_avg_list List of last N computed deltas, used to average
2219 void PGMap::update_delta(
2222 const pool_stat_t
& old_pool_sum
,
2224 const pool_stat_t
& current_pool_sum
,
2225 pool_stat_t
*result_pool_delta
,
2226 utime_t
*result_ts_delta
,
2227 mempool::pgmap::list
<pair
<pool_stat_t
,utime_t
> > *delta_avg_list
)
2229 /* @p ts is the timestamp we want to associate with the data
2230 * in @p old_pool_sum, and on which we will base ourselves to
2231 * calculate the delta, stored in 'delta_t'.
2234 delta_t
= ts
; // start with the provided timestamp
2235 delta_t
-= *last_ts
; // take the last timestamp we saw
2236 *last_ts
= ts
; // @p ts becomes the last timestamp we saw
2238 // adjust delta_t, quick start if there is no update in a long period
2239 delta_t
= std::min(delta_t
,
2240 utime_t(2 * (cct
? cct
->_conf
->mon_delta_reset_interval
: 10), 0));
2242 // calculate a delta, and average over the last 6 deltas by default.
2243 /* start by taking a copy of our current @p result_pool_sum, and by
2244 * taking out the stats from @p old_pool_sum. This generates a stats
2245 * delta. Stash this stats delta in @p delta_avg_list, along with the
2246 * timestamp delta for these results.
2248 pool_stat_t d
= current_pool_sum
;
2249 d
.stats
.sub(old_pool_sum
.stats
);
2250 delta_avg_list
->push_back(make_pair(d
,delta_t
));
2251 *result_ts_delta
+= delta_t
;
2253 /* Aggregate current delta, and take out the last seen delta (if any) to
2256 result_pool_delta
->stats
.add(d
.stats
);
2257 size_t s
= MAX(1, cct
? cct
->_conf
->mon_stat_smooth_intervals
: 1);
2258 if (delta_avg_list
->size() > s
) {
2259 result_pool_delta
->stats
.sub(delta_avg_list
->front().first
.stats
);
2260 *result_ts_delta
-= delta_avg_list
->front().second
;
2261 delta_avg_list
->pop_front();
2266 * update aggregated delta
2268 * @param cct ceph context
2269 * @param ts Timestamp
2270 * @param pg_sum_old Old pg_sum
2272 void PGMap::update_global_delta(CephContext
*cct
,
2273 const utime_t ts
, const pool_stat_t
& pg_sum_old
)
2275 update_delta(cct
, ts
, pg_sum_old
, &stamp
, pg_sum
, &pg_sum_delta
,
2276 &stamp_delta
, &pg_sum_deltas
);
2280 * Update a given pool's deltas
2282 * @param cct Ceph Context
2283 * @param ts Timestamp for the stats being delta'ed
2284 * @param pool Pool's id
2285 * @param old_pool_sum Previous stats sum
2287 void PGMap::update_one_pool_delta(
2290 const uint64_t pool
,
2291 const pool_stat_t
& old_pool_sum
)
2293 if (per_pool_sum_deltas
.count(pool
) == 0) {
2294 assert(per_pool_sum_deltas_stamps
.count(pool
) == 0);
2295 assert(per_pool_sum_delta
.count(pool
) == 0);
2298 auto& sum_delta
= per_pool_sum_delta
[pool
];
2300 update_delta(cct
, ts
, old_pool_sum
, &sum_delta
.second
, pg_pool_sum
[pool
],
2301 &sum_delta
.first
, &per_pool_sum_deltas_stamps
[pool
],
2302 &per_pool_sum_deltas
[pool
]);
2306 * Update pools' deltas
2308 * @param cct CephContext
2309 * @param ts Timestamp for the stats being delta'ed
2310 * @param pg_pool_sum_old Map of pool stats for delta calcs.
2312 void PGMap::update_pool_deltas(
2313 CephContext
*cct
, const utime_t ts
,
2314 const mempool::pgmap::unordered_map
<uint64_t,pool_stat_t
>& pg_pool_sum_old
)
2316 for (auto it
= pg_pool_sum_old
.begin();
2317 it
!= pg_pool_sum_old
.end(); ++it
) {
2318 update_one_pool_delta(cct
, ts
, it
->first
, it
->second
);
2322 void PGMap::clear_delta()
2324 pg_sum_delta
= pool_stat_t();
2325 pg_sum_deltas
.clear();
2326 stamp_delta
= utime_t();
2329 void PGMap::generate_test_instances(list
<PGMap
*>& o
)
2331 o
.push_back(new PGMap
);
2332 list
<Incremental
*> inc
;
2333 Incremental::generate_test_instances(inc
);
2336 while (!inc
.empty()) {
2337 PGMap
*pmp
= new PGMap();
2340 o
.back()->apply_incremental(NULL
, *inc
.front());
2346 void PGMap::get_filtered_pg_stats(uint32_t state
, int64_t poolid
, int64_t osdid
,
2347 bool primary
, set
<pg_t
>& pgs
) const
2349 for (auto i
= pg_stat
.begin();
2352 if ((poolid
>= 0) && (uint64_t(poolid
) != i
->first
.pool()))
2354 if ((osdid
>= 0) && !(i
->second
.is_acting_osd(osdid
,primary
)))
2356 if (!(i
->second
.state
& state
))
2358 pgs
.insert(i
->first
);
2362 void PGMap::dump_filtered_pg_stats(Formatter
*f
, set
<pg_t
>& pgs
) const
2364 f
->open_array_section("pg_stats");
2365 for (auto i
= pgs
.begin(); i
!= pgs
.end(); ++i
) {
2366 const pg_stat_t
& st
= pg_stat
.at(*i
);
2367 f
->open_object_section("pg_stat");
2368 f
->dump_stream("pgid") << *i
;
2375 void PGMap::dump_filtered_pg_stats(ostream
& ss
, set
<pg_t
>& pgs
) const
2379 tab
.define_column("PG_STAT", TextTable::LEFT
, TextTable::LEFT
);
2380 tab
.define_column("OBJECTS", TextTable::LEFT
, TextTable::RIGHT
);
2381 tab
.define_column("MISSING_ON_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
2382 tab
.define_column("DEGRADED", TextTable::LEFT
, TextTable::RIGHT
);
2383 tab
.define_column("MISPLACED", TextTable::LEFT
, TextTable::RIGHT
);
2384 tab
.define_column("UNFOUND", TextTable::LEFT
, TextTable::RIGHT
);
2385 tab
.define_column("BYTES", TextTable::LEFT
, TextTable::RIGHT
);
2386 tab
.define_column("LOG", TextTable::LEFT
, TextTable::RIGHT
);
2387 tab
.define_column("DISK_LOG", TextTable::LEFT
, TextTable::RIGHT
);
2388 tab
.define_column("STATE", TextTable::LEFT
, TextTable::RIGHT
);
2389 tab
.define_column("STATE_STAMP", TextTable::LEFT
, TextTable::RIGHT
);
2390 tab
.define_column("VERSION", TextTable::LEFT
, TextTable::RIGHT
);
2391 tab
.define_column("REPORTED", TextTable::LEFT
, TextTable::RIGHT
);
2392 tab
.define_column("UP", TextTable::LEFT
, TextTable::RIGHT
);
2393 tab
.define_column("UP_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
2394 tab
.define_column("ACTING", TextTable::LEFT
, TextTable::RIGHT
);
2395 tab
.define_column("ACTING_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
2396 tab
.define_column("LAST_SCRUB", TextTable::LEFT
, TextTable::RIGHT
);
2397 tab
.define_column("SCRUB_STAMP", TextTable::LEFT
, TextTable::RIGHT
);
2398 tab
.define_column("LAST_DEEP_SCRUB", TextTable::LEFT
, TextTable::RIGHT
);
2399 tab
.define_column("DEEP_SCRUB_STAMP", TextTable::LEFT
, TextTable::RIGHT
);
2401 for (auto i
= pgs
.begin(); i
!= pgs
.end(); ++i
) {
2402 const pg_stat_t
& st
= pg_stat
.at(*i
);
2404 ostringstream reported
;
2405 reported
<< st
.reported_epoch
<< ":" << st
.reported_seq
;
2408 << st
.stats
.sum
.num_objects
2409 << st
.stats
.sum
.num_objects_missing_on_primary
2410 << st
.stats
.sum
.num_objects_degraded
2411 << st
.stats
.sum
.num_objects_misplaced
2412 << st
.stats
.sum
.num_objects_unfound
2413 << st
.stats
.sum
.num_bytes
2415 << st
.ondisk_log_size
2416 << pg_state_string(st
.state
)
2423 << st
.acting_primary
2425 << st
.last_scrub_stamp
2426 << st
.last_deep_scrub
2427 << st
.last_deep_scrub_stamp
2428 << TextTable::endrow
;
2436 // Only called with a single bit set in "what"
2437 static void note_stuck_detail(
2439 mempool::pgmap::unordered_map
<pg_t
,pg_stat_t
>& stuck_pgs
,
2441 list
<pair
<health_status_t
,string
> > *detail
)
2444 for (auto p
= stuck_pgs
.begin();
2445 p
!= stuck_pgs
.end();
2449 const char *whatname
= 0;
2451 case PGMap::STUCK_INACTIVE
:
2452 since
= p
->second
.last_active
;
2453 whatname
= "inactive";
2455 case PGMap::STUCK_UNCLEAN
:
2456 since
= p
->second
.last_clean
;
2457 whatname
= "unclean";
2459 case PGMap::STUCK_DEGRADED
:
2460 since
= p
->second
.last_undegraded
;
2461 whatname
= "degraded";
2463 case PGMap::STUCK_UNDERSIZED
:
2464 since
= p
->second
.last_fullsized
;
2465 whatname
= "undersized";
2467 case PGMap::STUCK_STALE
:
2468 since
= p
->second
.last_unstale
;
2474 if (--max_detail
== 0) {
2476 ss
<< (stuck_pgs
.size() - n
) << " more pgs are also stuck " << whatname
;
2477 detail
->push_back(make_pair(HEALTH_WARN
, ss
.str()));
2481 ss
<< "pg " << p
->first
<< " is stuck " << whatname
;
2482 if (since
== utime_t()) {
2483 ss
<< " since forever";
2485 utime_t dur
= ceph_clock_now() - since
;
2486 ss
<< " for " << dur
;
2488 ss
<< ", current state " << pg_state_string(p
->second
.state
)
2489 << ", last acting " << p
->second
.acting
;
2490 detail
->push_back(make_pair(HEALTH_WARN
, ss
.str()));
2494 static pair
<int,int> _warn_slow_request_histogram(
2496 const pow2_hist_t
& h
,
2498 list
<pair
<health_status_t
,string
> >& summary
,
2499 list
<pair
<health_status_t
,string
> > *detail
)
2502 return make_pair(0, 0);
2504 unsigned warn
= 0, error
= 0;
2506 cct
->_conf
->mon_osd_warn_op_age
* cct
->_conf
->mon_osd_err_op_age_ratio
;
2507 for (unsigned i
= h
.h
.size() - 1; i
> 0; --i
) {
2508 float ub
= (float)(1 << i
) / 1000.0;
2509 if (ub
< cct
->_conf
->mon_osd_warn_op_age
)
2512 auto sev
= HEALTH_WARN
;
2521 ss
<< h
.h
[i
] << " ops are blocked > " << ub
<< " sec" << suffix
;
2522 detail
->push_back(make_pair(sev
, ss
.str()));
2526 return make_pair(warn
, error
);
2530 enum class scrubbed_or_deepscrubbed_t
{ SCRUBBED
, DEEPSCRUBBED
};
2532 void print_unscrubbed_detailed(
2533 const std::pair
<const pg_t
,pg_stat_t
> &pg_entry
,
2534 list
<pair
<health_status_t
,string
> > *detail
,
2535 scrubbed_or_deepscrubbed_t how_scrubbed
)
2537 std::stringstream ss
;
2538 const auto& pg_stat(pg_entry
.second
);
2540 ss
<< "pg " << pg_entry
.first
<< " is not ";
2541 if (how_scrubbed
== scrubbed_or_deepscrubbed_t::SCRUBBED
) {
2542 ss
<< "scrubbed, last_scrub_stamp "
2543 << pg_stat
.last_scrub_stamp
;
2544 } else if (how_scrubbed
== scrubbed_or_deepscrubbed_t::DEEPSCRUBBED
) {
2545 ss
<< "deep-scrubbed, last_deep_scrub_stamp "
2546 << pg_stat
.last_deep_scrub_stamp
;
2549 detail
->push_back(make_pair(HEALTH_WARN
, ss
.str()));
2552 using pg_stat_map_t
= const mempool::pgmap::unordered_map
<pg_t
,pg_stat_t
>;
2554 void print_unscrubbed_pgs(
2555 pg_stat_map_t
& pg_stats
,
2556 list
<pair
<health_status_t
,string
> > &summary
,
2557 list
<pair
<health_status_t
,string
> > *detail
,
2558 const CephContext
* cct
)
2560 if (cct
->_conf
->mon_warn_not_scrubbed
== 0 &&
2561 cct
->_conf
->mon_warn_not_deep_scrubbed
== 0)
2565 const utime_t now
= ceph_clock_now();
2566 for (const auto& pg_entry
: pg_stats
) {
2567 const auto& pg_stat(pg_entry
.second
);
2568 const utime_t time_since_ls
= now
- pg_stat
.last_scrub_stamp
;
2569 const utime_t time_since_lds
= now
- pg_stat
.last_deep_scrub_stamp
;
2571 const int mon_warn_not_scrubbed
=
2572 cct
->_conf
->mon_warn_not_scrubbed
+ cct
->_conf
->mon_scrub_interval
;
2574 const int mon_warn_not_deep_scrubbed
=
2575 cct
->_conf
->mon_warn_not_deep_scrubbed
+ cct
->_conf
->osd_deep_scrub_interval
;
2577 bool not_scrubbed
= (time_since_ls
>= mon_warn_not_scrubbed
&&
2578 cct
->_conf
->mon_warn_not_scrubbed
!= 0);
2580 bool not_deep_scrubbed
= (time_since_lds
>= mon_warn_not_deep_scrubbed
&&
2581 cct
->_conf
->mon_warn_not_deep_scrubbed
!= 0);
2583 if (detail
!= nullptr) {
2585 print_unscrubbed_detailed(pg_entry
,
2587 scrubbed_or_deepscrubbed_t::SCRUBBED
);
2589 if (not_deep_scrubbed
) {
2590 print_unscrubbed_detailed(pg_entry
,
2592 scrubbed_or_deepscrubbed_t::DEEPSCRUBBED
);
2595 if (not_scrubbed
|| not_deep_scrubbed
) {
2600 if (pgs_count
> 0) {
2601 std::stringstream ss
;
2602 ss
<< pgs_count
<< " unscrubbed pgs";
2603 summary
.push_back(make_pair(HEALTH_WARN
, ss
.str()));
2608 void PGMap::get_health_checks(
2610 const OSDMap
& osdmap
,
2611 health_check_map_t
*checks
) const
2613 utime_t now
= ceph_clock_now();
2614 const unsigned max
= cct
->_conf
->mon_health_max_detail
;
2615 const auto& pools
= osdmap
.get_pools();
2617 typedef enum pg_consequence_t
{
2618 UNAVAILABLE
= 1, // Client IO to the pool may block
2619 DEGRADED
= 2, // Fewer than the requested number of replicas are present
2620 DEGRADED_FULL
= 3, // Fewer than the request number of replicas may be present
2621 // and insufficiet resources are present to fix this
2622 DAMAGED
= 4 // The data may be missing or inconsistent on disk and
2626 // For a given PG state, how should it be reported at the pool level?
2627 class PgStateResponse
{
2629 pg_consequence_t consequence
;
2630 typedef std::function
< utime_t(const pg_stat_t
&) > stuck_cb
;
2631 stuck_cb stuck_since
;
2634 PgStateResponse(const pg_consequence_t
&c
, stuck_cb s
)
2635 : consequence(c
), stuck_since(s
), invert(false)
2639 PgStateResponse(const pg_consequence_t
&c
, stuck_cb s
, bool i
)
2640 : consequence(c
), stuck_since(s
), invert(i
)
2645 // Record the PG state counts that contributed to a reported pool state
2648 // Map of PG_STATE_* to number of pgs in that state.
2649 std::map
<unsigned, unsigned> states
;
2651 // List of all PG IDs that had a state contributing
2652 // to this health condition.
2655 std::map
<pg_t
, std::string
> pg_messages
;
2658 // Map of PG state to how to respond to it
2659 std::map
<unsigned, PgStateResponse
> state_to_response
= {
2660 // Immediate reports
2661 { PG_STATE_INCONSISTENT
, {DAMAGED
, {}} },
2662 { PG_STATE_INCOMPLETE
, {UNAVAILABLE
, {}} },
2663 { PG_STATE_REPAIR
, {DAMAGED
, {}} },
2664 { PG_STATE_SNAPTRIM_ERROR
, {DAMAGED
, {}} },
2665 { PG_STATE_BACKFILL_TOOFULL
, {DEGRADED_FULL
, {}} },
2666 { PG_STATE_RECOVERY_TOOFULL
, {DEGRADED_FULL
, {}} },
2667 { PG_STATE_DEGRADED
, {DEGRADED
, {}} },
2668 { PG_STATE_DOWN
, {UNAVAILABLE
, {}} },
2669 // Delayed (wait until stuck) reports
2670 { PG_STATE_PEERING
, {UNAVAILABLE
, [](const pg_stat_t
&p
){return p
.last_peered
;} } },
2671 { PG_STATE_UNDERSIZED
, {DEGRADED
, [](const pg_stat_t
&p
){return p
.last_fullsized
;} } },
2672 { PG_STATE_STALE
, {UNAVAILABLE
, [](const pg_stat_t
&p
){return p
.last_unstale
;} } },
2673 // Delayed and inverted reports
2674 { PG_STATE_ACTIVE
, {UNAVAILABLE
, [](const pg_stat_t
&p
){return p
.last_active
;}, true} },
2675 { PG_STATE_CLEAN
, {DEGRADED
, [](const pg_stat_t
&p
){return p
.last_clean
;}, true} }
2678 // Specialized state printer that takes account of inversion of
2679 // ACTIVE, CLEAN checks.
2680 auto state_name
= [](const uint32_t &state
) {
2681 // Special cases for the states that are inverted checks
2682 if (state
== PG_STATE_CLEAN
) {
2683 return std::string("unclean");
2684 } else if (state
== PG_STATE_ACTIVE
) {
2685 return std::string("inactive");
2687 return pg_state_string(state
);
2691 // Map of what is wrong to information about why, implicitly also stores
2692 // the list of what is wrong.
2693 std::map
<pg_consequence_t
, PgCauses
> detected
;
2695 // Optimisation: trim down the number of checks to apply based on
2696 // the summary counters
2697 std::map
<unsigned, PgStateResponse
> possible_responses
;
2698 for (const auto &i
: num_pg_by_state
) {
2699 for (const auto &j
: state_to_response
) {
2700 if (!j
.second
.invert
) {
2701 // Check for normal tests by seeing if any pgs have the flag
2702 if (i
.first
& j
.first
) {
2703 possible_responses
.insert(j
);
2709 for (const auto &j
: state_to_response
) {
2710 if (j
.second
.invert
) {
2711 // Check for inverted tests by seeing if not-all pgs have the flag
2712 const auto &found
= num_pg_by_state
.find(j
.first
);
2713 if (found
== num_pg_by_state
.end() || found
->second
!= num_pg
) {
2714 possible_responses
.insert(j
);
2719 utime_t cutoff
= now
- utime_t(cct
->_conf
->mon_pg_stuck_threshold
, 0);
2720 // Loop over all PGs, if there are any possibly-unhealthy states in there
2721 if (!possible_responses
.empty()) {
2722 for (const auto& i
: pg_stat
) {
2723 const auto &pg_id
= i
.first
;
2724 const auto &pg_info
= i
.second
;
2726 for (const auto &j
: state_to_response
) {
2727 const auto &pg_response_state
= j
.first
;
2728 const auto &pg_response
= j
.second
;
2730 // Apply the state test
2731 if (!(bool(pg_info
.state
& pg_response_state
) != pg_response
.invert
)) {
2735 // Apply stuckness test if needed
2736 if (pg_response
.stuck_since
) {
2737 // Delayed response, check for stuckness
2738 utime_t last_whatever
= pg_response
.stuck_since(pg_info
);
2739 if (last_whatever
>= cutoff
) {
2740 // Not stuck enough, ignore.
2747 auto &causes
= detected
[pg_response
.consequence
];
2748 causes
.states
[pg_response_state
]++;
2749 causes
.pgs
.insert(pg_id
);
2751 // Don't bother composing detail string if we have already recorded
2753 if (causes
.pg_messages
.size() > max
) {
2757 std::ostringstream ss
;
2758 if (pg_response
.stuck_since
) {
2759 utime_t since
= pg_response
.stuck_since(pg_info
);
2760 ss
<< "pg " << pg_id
<< " is stuck " << state_name(pg_response_state
);
2761 if (since
== utime_t()) {
2762 ss
<< " since forever";
2764 utime_t dur
= now
- since
;
2765 ss
<< " for " << dur
;
2767 ss
<< ", current state " << pg_state_string(pg_info
.state
)
2768 << ", last acting " << pg_info
.acting
;
2770 ss
<< "pg " << pg_id
<< " is "
2771 << pg_state_string(pg_info
.state
);
2772 ss
<< ", acting " << pg_info
.acting
;
2773 if (pg_info
.stats
.sum
.num_objects_unfound
) {
2774 ss
<< ", " << pg_info
.stats
.sum
.num_objects_unfound
2779 if (pg_info
.state
& PG_STATE_INCOMPLETE
) {
2780 const pg_pool_t
*pi
= osdmap
.get_pg_pool(pg_id
.pool());
2781 if (pi
&& pi
->min_size
> 1) {
2782 ss
<< " (reducing pool "
2783 << osdmap
.get_pool_name(pg_id
.pool())
2784 << " min_size from " << (int)pi
->min_size
2785 << " may help; search ceph.com/docs for 'incomplete')";
2789 causes
.pg_messages
[pg_id
] = ss
.str();
2793 dout(10) << __func__
<< " skipping loop over PGs: counters look OK" << dendl
;
2796 for (const auto &i
: detected
) {
2797 std::string health_code
;
2798 health_status_t sev
;
2799 std::string summary
;
2802 health_code
= "PG_AVAILABILITY";
2804 summary
= "Reduced data availability: ";
2807 health_code
= "PG_DEGRADED";
2808 summary
= "Degraded data redundancy: ";
2812 health_code
= "PG_DEGRADED_FULL";
2813 summary
= "Degraded data redundancy (low space): ";
2817 health_code
= "PG_DAMAGED";
2818 summary
= "Possible data damage: ";
2825 if (i
.first
== DEGRADED
) {
2826 if (pg_sum
.stats
.sum
.num_objects_degraded
&&
2827 pg_sum
.stats
.sum
.num_object_copies
> 0) {
2828 double pc
= (double)pg_sum
.stats
.sum
.num_objects_degraded
/
2829 (double)pg_sum
.stats
.sum
.num_object_copies
* (double)100.0;
2831 snprintf(b
, sizeof(b
), "%.3lf", pc
);
2833 ss
<< pg_sum
.stats
.sum
.num_objects_degraded
2834 << "/" << pg_sum
.stats
.sum
.num_object_copies
<< " objects degraded ("
2837 // Throw in a comma for the benefit of the following PG counts
2838 summary
+= ss
.str() + ", ";
2842 // Compose summary message saying how many PGs in what states led
2843 // to this health check failing
2844 std::vector
<std::string
> pg_msgs
;
2845 for (const auto &j
: i
.second
.states
) {
2846 std::ostringstream msg
;
2847 msg
<< j
.second
<< (j
.second
> 1 ? " pgs " : " pg ") << state_name(j
.first
);
2848 pg_msgs
.push_back(msg
.str());
2850 summary
+= joinify(pg_msgs
.begin(), pg_msgs
.end(), std::string(", "));
2854 health_check_t
*check
= &checks
->add(
2859 // Compose list of PGs contributing to this health check failing
2860 for (const auto &j
: i
.second
.pg_messages
) {
2861 check
->detail
.push_back(j
.second
);
2866 if (pg_sum
.stats
.sum
.num_scrub_errors
) {
2868 ss
<< pg_sum
.stats
.sum
.num_scrub_errors
<< " scrub errors";
2869 checks
->add("OSD_SCRUB_ERRORS", HEALTH_ERR
, ss
.str());
2872 // CACHE_POOL_NEAR_FULL
2874 list
<string
> detail
;
2875 unsigned num_pools
= 0;
2876 for (auto& p
: pools
) {
2877 if ((!p
.second
.target_max_objects
&& !p
.second
.target_max_bytes
) ||
2878 !pg_pool_sum
.count(p
.first
)) {
2881 bool nearfull
= false;
2882 const string
& name
= osdmap
.get_pool_name(p
.first
);
2883 const pool_stat_t
& st
= get_pg_pool_sum_stat(p
.first
);
2884 uint64_t ratio
= p
.second
.cache_target_full_ratio_micro
+
2885 ((1000000 - p
.second
.cache_target_full_ratio_micro
) *
2886 cct
->_conf
->mon_cache_target_full_warn_ratio
);
2887 if (p
.second
.target_max_objects
&&
2888 (uint64_t)(st
.stats
.sum
.num_objects
-
2889 st
.stats
.sum
.num_objects_hit_set_archive
) >
2890 p
.second
.target_max_objects
* (ratio
/ 1000000.0)) {
2892 ss
<< "cache pool '" << name
<< "' with "
2893 << si_t(st
.stats
.sum
.num_objects
)
2894 << " objects at/near target max "
2895 << si_t(p
.second
.target_max_objects
) << " objects";
2896 detail
.push_back(ss
.str());
2899 if (p
.second
.target_max_bytes
&&
2900 (uint64_t)(st
.stats
.sum
.num_bytes
-
2901 st
.stats
.sum
.num_bytes_hit_set_archive
) >
2902 p
.second
.target_max_bytes
* (ratio
/ 1000000.0)) {
2904 ss
<< "cache pool '" << name
2905 << "' with " << si_t(st
.stats
.sum
.num_bytes
)
2906 << "B at/near target max "
2907 << si_t(p
.second
.target_max_bytes
) << "B";
2908 detail
.push_back(ss
.str());
2915 if (!detail
.empty()) {
2917 ss
<< num_pools
<< " cache pools at or near target size";
2918 auto& d
= checks
->add("CACHE_POOL_NEAR_FULL", HEALTH_WARN
, ss
.str());
2919 d
.detail
.swap(detail
);
2924 int num_in
= osdmap
.get_num_in_osds();
2925 int sum_pg_up
= MAX(pg_sum
.up
, static_cast<int32_t>(pg_stat
.size()));
2927 cct
->_conf
->mon_pg_warn_min_per_osd
> 0 &&
2928 osdmap
.get_pools().size() > 0) {
2929 int per
= sum_pg_up
/ num_in
;
2930 if (per
< cct
->_conf
->mon_pg_warn_min_per_osd
&& per
) {
2932 ss
<< "too few PGs per OSD (" << per
2933 << " < min " << cct
->_conf
->mon_pg_warn_min_per_osd
<< ")";
2934 checks
->add("TOO_FEW_PGS", HEALTH_WARN
, ss
.str());
2939 if (num_in
&& cct
->_conf
->mon_pg_warn_max_per_osd
> 0) {
2940 int per
= sum_pg_up
/ num_in
;
2941 if (per
> cct
->_conf
->mon_pg_warn_max_per_osd
) {
2943 ss
<< "too many PGs per OSD (" << per
2944 << " > max " << cct
->_conf
->mon_pg_warn_max_per_osd
<< ")";
2945 checks
->add("TOO_MANY_PGS", HEALTH_WARN
, ss
.str());
2950 // MANY_OBJECTS_PER_PG
2951 if (!pg_stat
.empty()) {
2952 list
<string
> pgp_detail
, many_detail
;
2953 for (auto p
= pg_pool_sum
.begin();
2954 p
!= pg_pool_sum
.end();
2956 const pg_pool_t
*pi
= osdmap
.get_pg_pool(p
->first
);
2958 continue; // in case osdmap changes haven't propagated to PGMap yet
2959 const string
& name
= osdmap
.get_pool_name(p
->first
);
2960 if (pi
->get_pg_num() > pi
->get_pgp_num() &&
2961 !(name
.find(".DELETED") != string::npos
&&
2962 cct
->_conf
->mon_fake_pool_delete
)) {
2964 ss
<< "pool " << name
<< " pg_num "
2965 << pi
->get_pg_num() << " > pgp_num " << pi
->get_pgp_num();
2966 pgp_detail
.push_back(ss
.str());
2968 int average_objects_per_pg
= pg_sum
.stats
.sum
.num_objects
/ pg_stat
.size();
2969 if (average_objects_per_pg
> 0 &&
2970 pg_sum
.stats
.sum
.num_objects
>= cct
->_conf
->mon_pg_warn_min_objects
&&
2971 p
->second
.stats
.sum
.num_objects
>=
2972 cct
->_conf
->mon_pg_warn_min_pool_objects
) {
2973 int objects_per_pg
= p
->second
.stats
.sum
.num_objects
/ pi
->get_pg_num();
2974 float ratio
= (float)objects_per_pg
/ (float)average_objects_per_pg
;
2975 if (cct
->_conf
->mon_pg_warn_max_object_skew
> 0 &&
2976 ratio
> cct
->_conf
->mon_pg_warn_max_object_skew
) {
2978 ss
<< "pool " << name
<< " objects per pg ("
2979 << objects_per_pg
<< ") is more than " << ratio
2980 << " times cluster average ("
2981 << average_objects_per_pg
<< ")";
2982 many_detail
.push_back(ss
.str());
2986 if (!pgp_detail
.empty()) {
2988 ss
<< pgp_detail
.size() << " pools have pg_num > pgp_num";
2989 auto& d
= checks
->add("SMALLER_PGP_NUM", HEALTH_WARN
, ss
.str());
2990 d
.detail
.swap(pgp_detail
);
2992 if (!many_detail
.empty()) {
2994 ss
<< many_detail
.size() << " pools have many more objects per pg than"
2996 auto& d
= checks
->add("MANY_OBJECTS_PER_PG", HEALTH_WARN
, ss
.str());
2997 d
.detail
.swap(many_detail
);
3004 float warn_threshold
= (float)g_conf
->mon_pool_quota_warn_threshold
/100;
3005 float crit_threshold
= (float)g_conf
->mon_pool_quota_crit_threshold
/100;
3006 list
<string
> full_detail
, nearfull_detail
;
3007 unsigned full_pools
= 0, nearfull_pools
= 0;
3008 for (auto it
: pools
) {
3009 auto it2
= pg_pool_sum
.find(it
.first
);
3010 if (it2
== pg_pool_sum
.end()) {
3013 const pool_stat_t
*pstat
= &it2
->second
;
3014 const object_stat_sum_t
& sum
= pstat
->stats
.sum
;
3015 const string
& pool_name
= osdmap
.get_pool_name(it
.first
);
3016 const pg_pool_t
&pool
= it
.second
;
3017 bool full
= false, nearfull
= false;
3018 if (pool
.quota_max_objects
> 0) {
3020 if ((uint64_t)sum
.num_objects
>= pool
.quota_max_objects
) {
3021 } else if (crit_threshold
> 0 &&
3022 sum
.num_objects
>= pool
.quota_max_objects
*crit_threshold
) {
3023 ss
<< "pool '" << pool_name
3024 << "' has " << sum
.num_objects
<< " objects"
3025 << " (max " << pool
.quota_max_objects
<< ")";
3026 full_detail
.push_back(ss
.str());
3028 } else if (warn_threshold
> 0 &&
3029 sum
.num_objects
>= pool
.quota_max_objects
*warn_threshold
) {
3030 ss
<< "pool '" << pool_name
3031 << "' has " << sum
.num_objects
<< " objects"
3032 << " (max " << pool
.quota_max_objects
<< ")";
3033 nearfull_detail
.push_back(ss
.str());
3037 if (pool
.quota_max_bytes
> 0) {
3039 if ((uint64_t)sum
.num_bytes
>= pool
.quota_max_bytes
) {
3040 } else if (crit_threshold
> 0 &&
3041 sum
.num_bytes
>= pool
.quota_max_bytes
*crit_threshold
) {
3042 ss
<< "pool '" << pool_name
3043 << "' has " << si_t(sum
.num_bytes
) << " bytes"
3044 << " (max " << si_t(pool
.quota_max_bytes
) << ")";
3045 full_detail
.push_back(ss
.str());
3047 } else if (warn_threshold
> 0 &&
3048 sum
.num_bytes
>= pool
.quota_max_bytes
*warn_threshold
) {
3049 ss
<< "pool '" << pool_name
3050 << "' has " << si_t(sum
.num_bytes
) << " bytes"
3051 << " (max " << si_t(pool
.quota_max_bytes
) << ")";
3052 nearfull_detail
.push_back(ss
.str());
3065 ss
<< full_pools
<< " pools full";
3066 auto& d
= checks
->add("POOL_FULL", HEALTH_ERR
, ss
.str());
3067 d
.detail
.swap(full_detail
);
3069 if (nearfull_pools
) {
3071 ss
<< nearfull_pools
<< " pools full";
3072 auto& d
= checks
->add("POOL_NEAR_FULL", HEALTH_WARN
, ss
.str());
3073 d
.detail
.swap(nearfull_detail
);
3078 if (pg_sum
.stats
.sum
.num_objects_misplaced
&&
3079 pg_sum
.stats
.sum
.num_object_copies
> 0) {
3080 double pc
= (double)pg_sum
.stats
.sum
.num_objects_misplaced
/
3081 (double)pg_sum
.stats
.sum
.num_object_copies
* (double)100.0;
3083 snprintf(b
, sizeof(b
), "%.3lf", pc
);
3085 ss
<< pg_sum
.stats
.sum
.num_objects_misplaced
3086 << "/" << pg_sum
.stats
.sum
.num_object_copies
<< " objects misplaced ("
3088 checks
->add("OBJECT_MISPLACED", HEALTH_WARN
, ss
.str());
3092 if (pg_sum
.stats
.sum
.num_objects_unfound
&&
3093 pg_sum
.stats
.sum
.num_objects
) {
3094 double pc
= (double)pg_sum
.stats
.sum
.num_objects_unfound
/
3095 (double)pg_sum
.stats
.sum
.num_objects
* (double)100.0;
3097 snprintf(b
, sizeof(b
), "%.3lf", pc
);
3099 ss
<< pg_sum
.stats
.sum
.num_objects_unfound
3100 << "/" << pg_sum
.stats
.sum
.num_objects
<< " objects unfound (" << b
<< "%)";
3101 auto& d
= checks
->add("OBJECT_UNFOUND", HEALTH_WARN
, ss
.str());
3103 for (auto& p
: pg_stat
) {
3104 if (p
.second
.stats
.sum
.num_objects_unfound
) {
3106 ss
<< "pg " << p
.first
3107 << " has " << p
.second
.stats
.sum
.num_objects_unfound
3108 << " unfound objects";
3109 d
.detail
.push_back(ss
.str());
3110 if (d
.detail
.size() > max
) {
3111 d
.detail
.push_back("(additional pgs left out for brevity)");
3120 if (cct
->_conf
->mon_osd_warn_op_age
> 0 &&
3121 !osd_sum
.op_queue_age_hist
.h
.empty() &&
3122 osd_sum
.op_queue_age_hist
.upper_bound() / 1000.0 >
3123 cct
->_conf
->mon_osd_warn_op_age
) {
3124 list
<string
> warn_detail
, error_detail
;
3125 unsigned warn
= 0, error
= 0;
3127 cct
->_conf
->mon_osd_warn_op_age
* cct
->_conf
->mon_osd_err_op_age_ratio
;
3128 const pow2_hist_t
& h
= osd_sum
.op_queue_age_hist
;
3129 for (unsigned i
= h
.h
.size() - 1; i
> 0; --i
) {
3130 float ub
= (float)(1 << i
) / 1000.0;
3131 if (ub
< cct
->_conf
->mon_osd_warn_op_age
)
3135 ss
<< h
.h
[i
] << " ops are blocked > " << ub
<< " sec";
3138 error_detail
.push_back(ss
.str());
3141 warn_detail
.push_back(ss
.str());
3146 map
<float,set
<int>> warn_osd_by_max
; // max -> osds
3147 map
<float,set
<int>> error_osd_by_max
; // max -> osds
3148 if (!warn_detail
.empty() || !error_detail
.empty()) {
3149 for (auto& p
: osd_stat
) {
3150 const pow2_hist_t
& h
= p
.second
.op_queue_age_hist
;
3151 for (unsigned i
= h
.h
.size() - 1; i
> 0; --i
) {
3152 float ub
= (float)(1 << i
) / 1000.0;
3153 if (ub
< cct
->_conf
->mon_osd_warn_op_age
)
3157 error_osd_by_max
[ub
].insert(p
.first
);
3159 warn_osd_by_max
[ub
].insert(p
.first
);
3167 if (!warn_detail
.empty()) {
3169 ss
<< warn
<< " slow requests are blocked > "
3170 << cct
->_conf
->mon_osd_warn_op_age
<< " sec";
3171 auto& d
= checks
->add("REQUEST_SLOW", HEALTH_WARN
, ss
.str());
3172 d
.detail
.swap(warn_detail
);
3174 for (auto& p
: warn_osd_by_max
) {
3176 if (p
.second
.size() > 1) {
3177 ss
<< "osds " << p
.second
3178 << " have blocked requests > " << p
.first
<< " sec";
3180 ss
<< "osd." << *p
.second
.begin()
3181 << " has blocked requests > " << p
.first
<< " sec";
3183 d
.detail
.push_back(ss
.str());
3189 if (!error_detail
.empty()) {
3191 ss
<< error
<< " stuck requests are blocked > "
3192 << err_age
<< " sec";
3193 auto& d
= checks
->add("REQUEST_STUCK", HEALTH_ERR
, ss
.str());
3194 d
.detail
.swap(error_detail
);
3196 for (auto& p
: error_osd_by_max
) {
3198 if (p
.second
.size() > 1) {
3199 ss
<< "osds " << p
.second
3200 << " have stuck requests > " << p
.first
<< " sec";
3202 ss
<< "osd." << *p
.second
.begin()
3203 << " has stuck requests > " << p
.first
<< " sec";
3205 d
.detail
.push_back(ss
.str());
3214 // PG_NOT_DEEP_SCRUBBED
3216 if (cct
->_conf
->mon_warn_not_scrubbed
||
3217 cct
->_conf
->mon_warn_not_deep_scrubbed
) {
3218 list
<string
> detail
, deep_detail
;
3219 const double age
= cct
->_conf
->mon_warn_not_scrubbed
+
3220 cct
->_conf
->mon_scrub_interval
;
3221 utime_t cutoff
= now
;
3223 const double deep_age
= cct
->_conf
->mon_warn_not_deep_scrubbed
+
3224 cct
->_conf
->osd_deep_scrub_interval
;
3225 utime_t deep_cutoff
= now
;
3226 deep_cutoff
-= deep_age
;
3227 for (auto& p
: pg_stat
) {
3228 if (cct
->_conf
->mon_warn_not_scrubbed
&&
3229 p
.second
.last_scrub_stamp
< cutoff
) {
3231 ss
<< "pg " << p
.first
<< " not scrubbed since "
3232 << p
.second
.last_scrub_stamp
;
3233 detail
.push_back(ss
.str());
3235 if (cct
->_conf
->mon_warn_not_deep_scrubbed
&&
3236 p
.second
.last_deep_scrub_stamp
< deep_cutoff
) {
3238 ss
<< "pg " << p
.first
<< " not deep-scrubbed since "
3239 << p
.second
.last_deep_scrub_stamp
;
3240 deep_detail
.push_back(ss
.str());
3243 if (!detail
.empty()) {
3245 ss
<< detail
.size() << " pgs not scrubbed for " << age
;
3246 auto& d
= checks
->add("PG_NOT_SCRUBBED", HEALTH_WARN
, ss
.str());
3247 d
.detail
.swap(detail
);
3249 if (!deep_detail
.empty()) {
3251 ss
<< deep_detail
.size() << " pgs not deep-scrubbed for " << deep_age
;
3252 auto& d
= checks
->add("PG_NOT_DEEP_SCRUBBED", HEALTH_WARN
, ss
.str());
3253 d
.detail
.swap(deep_detail
);
3259 if (g_conf
->get_val
<bool>("mon_warn_on_pool_no_app")) {
3260 list
<string
> detail
;
3261 for (auto &it
: pools
) {
3262 const pg_pool_t
&pool
= it
.second
;
3263 const string
& pool_name
= osdmap
.get_pool_name(it
.first
);
3264 auto it2
= pg_pool_sum
.find(it
.first
);
3265 if (it2
== pg_pool_sum
.end()) {
3268 const pool_stat_t
*pstat
= &it2
->second
;
3269 if (pstat
== nullptr) {
3272 const object_stat_sum_t
& sum
= pstat
->stats
.sum
;
3273 // application metadata is not encoded until luminous is minimum
3275 if (osdmap
.require_osd_release
>= CEPH_RELEASE_LUMINOUS
&&
3276 sum
.num_objects
> 0 && pool
.application_metadata
.empty() &&
3277 !pool
.is_tier() && !g_conf
->mon_debug_no_require_luminous
) {
3279 ss
<< "application not enabled on pool '" << pool_name
<< "'";
3280 detail
.push_back(ss
.str());
3283 if (!detail
.empty()) {
3285 ss
<< "application not enabled on " << detail
.size() << " pool(s)";
3286 auto& d
= checks
->add("POOL_APP_NOT_ENABLED", HEALTH_WARN
, ss
.str());
3288 tip
<< "use 'ceph osd pool application enable <pool-name> "
3289 << "<app-name>', where <app-name> is 'cephfs', 'rbd', 'rgw', "
3290 << "or freeform for custom applications.";
3291 detail
.push_back(tip
.str());
3292 d
.detail
.swap(detail
);
3297 void PGMap::get_health(
3299 const OSDMap
& osdmap
,
3300 list
<pair
<health_status_t
,string
> >& summary
,
3301 list
<pair
<health_status_t
,string
> > *detail
) const
3303 map
<string
,int> note
;
3304 auto p
= num_pg_by_state
.begin();
3305 auto p_end
= num_pg_by_state
.end();
3306 for (; p
!= p_end
; ++p
) {
3307 if (p
->first
& PG_STATE_STALE
)
3308 note
["stale"] += p
->second
;
3309 if (p
->first
& PG_STATE_DOWN
)
3310 note
["down"] += p
->second
;
3311 if (p
->first
& PG_STATE_UNDERSIZED
)
3312 note
["undersized"] += p
->second
;
3313 if (p
->first
& PG_STATE_DEGRADED
)
3314 note
["degraded"] += p
->second
;
3315 if (p
->first
& PG_STATE_INCONSISTENT
)
3316 note
["inconsistent"] += p
->second
;
3317 if (p
->first
& PG_STATE_PEERING
)
3318 note
["peering"] += p
->second
;
3319 if (p
->first
& PG_STATE_REPAIR
)
3320 note
["repair"] += p
->second
;
3321 if (p
->first
& PG_STATE_RECOVERING
)
3322 note
["recovering"] += p
->second
;
3323 if (p
->first
& PG_STATE_RECOVERY_WAIT
)
3324 note
["recovery_wait"] += p
->second
;
3325 if (p
->first
& PG_STATE_INCOMPLETE
)
3326 note
["incomplete"] += p
->second
;
3327 if (p
->first
& PG_STATE_BACKFILL_WAIT
)
3328 note
["backfill_wait"] += p
->second
;
3329 if (p
->first
& PG_STATE_BACKFILL
)
3330 note
["backfilling"] += p
->second
;
3331 if (p
->first
& PG_STATE_BACKFILL_TOOFULL
)
3332 note
["backfill_toofull"] += p
->second
;
3333 if (p
->first
& PG_STATE_RECOVERY_TOOFULL
)
3334 note
["recovery_toofull"] += p
->second
;
3335 if (p
->first
& PG_STATE_SNAPTRIM_ERROR
)
3336 note
["snaptrim_error"] += p
->second
;
3339 mempool::pgmap::unordered_map
<pg_t
, pg_stat_t
> stuck_pgs
;
3340 utime_t
now(ceph_clock_now());
3341 utime_t cutoff
= now
- utime_t(cct
->_conf
->mon_pg_stuck_threshold
, 0);
3342 uint64_t num_inactive_pgs
= 0;
3345 // we need to collect details of stuck pgs, first do a quick check
3346 // whether this will yield any results
3347 if (get_stuck_counts(cutoff
, note
)) {
3349 // there are stuck pgs. gather details for specified statuses
3350 // only if we know that there are pgs stuck in that status
3352 if (note
.find("stuck inactive") != note
.end()) {
3353 get_stuck_stats(PGMap::STUCK_INACTIVE
, cutoff
, stuck_pgs
);
3354 note
["stuck inactive"] = stuck_pgs
.size();
3355 num_inactive_pgs
+= stuck_pgs
.size();
3356 note_stuck_detail(PGMap::STUCK_INACTIVE
, stuck_pgs
,
3357 cct
->_conf
->mon_health_max_detail
, detail
);
3361 if (note
.find("stuck unclean") != note
.end()) {
3362 get_stuck_stats(PGMap::STUCK_UNCLEAN
, cutoff
, stuck_pgs
);
3363 note
["stuck unclean"] = stuck_pgs
.size();
3364 note_stuck_detail(PGMap::STUCK_UNCLEAN
, stuck_pgs
,
3365 cct
->_conf
->mon_health_max_detail
, detail
);
3369 if (note
.find("stuck undersized") != note
.end()) {
3370 get_stuck_stats(PGMap::STUCK_UNDERSIZED
, cutoff
, stuck_pgs
);
3371 note
["stuck undersized"] = stuck_pgs
.size();
3372 note_stuck_detail(PGMap::STUCK_UNDERSIZED
, stuck_pgs
,
3373 cct
->_conf
->mon_health_max_detail
, detail
);
3377 if (note
.find("stuck degraded") != note
.end()) {
3378 get_stuck_stats(PGMap::STUCK_DEGRADED
, cutoff
, stuck_pgs
);
3379 note
["stuck degraded"] = stuck_pgs
.size();
3380 note_stuck_detail(PGMap::STUCK_DEGRADED
, stuck_pgs
,
3381 cct
->_conf
->mon_health_max_detail
, detail
);
3385 if (note
.find("stuck stale") != note
.end()) {
3386 get_stuck_stats(PGMap::STUCK_STALE
, cutoff
, stuck_pgs
);
3387 note
["stuck stale"] = stuck_pgs
.size();
3388 num_inactive_pgs
+= stuck_pgs
.size();
3389 note_stuck_detail(PGMap::STUCK_STALE
, stuck_pgs
,
3390 cct
->_conf
->mon_health_max_detail
, detail
);
3394 get_stuck_counts(cutoff
, note
);
3395 auto p
= note
.find("stuck inactive");
3396 if (p
!= note
.end())
3397 num_inactive_pgs
+= p
->second
;
3398 p
= note
.find("stuck stale");
3399 if (p
!= note
.end())
3400 num_inactive_pgs
+= p
->second
;
3403 if (cct
->_conf
->mon_pg_min_inactive
> 0 &&
3404 num_inactive_pgs
>= cct
->_conf
->mon_pg_min_inactive
) {
3406 ss
<< num_inactive_pgs
<< " pgs are stuck inactive for more than " << cct
->_conf
->mon_pg_stuck_threshold
<< " seconds";
3407 summary
.push_back(make_pair(HEALTH_ERR
, ss
.str()));
3410 if (!note
.empty()) {
3411 for (auto p
= note
.begin(); p
!= note
.end(); ++p
) {
3413 ss
<< p
->second
<< " pgs " << p
->first
;
3414 summary
.push_back(make_pair(HEALTH_WARN
, ss
.str()));
3417 int n
= 0, more
= 0;
3418 int max
= cct
->_conf
->mon_health_max_detail
;
3419 for (auto p
= pg_stat
.begin();
3422 if ((p
->second
.state
& (PG_STATE_STALE
|
3424 PG_STATE_UNDERSIZED
|
3426 PG_STATE_INCONSISTENT
|
3429 PG_STATE_RECOVERING
|
3430 PG_STATE_RECOVERY_WAIT
|
3431 PG_STATE_RECOVERY_TOOFULL
|
3432 PG_STATE_INCOMPLETE
|
3433 PG_STATE_BACKFILL_WAIT
|
3435 PG_STATE_BACKFILL_TOOFULL
)) &&
3436 stuck_pgs
.count(p
->first
) == 0) {
3445 ss
<< "pg " << p
->first
<< " is " << pg_state_string(p
->second
.state
);
3446 ss
<< ", acting " << p
->second
.acting
;
3447 if (p
->second
.stats
.sum
.num_objects_unfound
)
3448 ss
<< ", " << p
->second
.stats
.sum
.num_objects_unfound
<< " unfound";
3449 if (p
->second
.state
& PG_STATE_INCOMPLETE
) {
3450 const pg_pool_t
*pi
= osdmap
.get_pg_pool(p
->first
.pool());
3451 if (pi
&& pi
->min_size
> 1) {
3452 ss
<< " (reducing pool " << osdmap
.get_pool_name(p
->first
.pool())
3453 << " min_size from " << (int)pi
->min_size
3454 << " may help; search ceph.com/docs for 'incomplete')";
3457 detail
->push_back(make_pair(HEALTH_WARN
, ss
.str()));
3462 ss
<< more
<< " more pgs are also unhealthy";
3463 detail
->push_back(make_pair(HEALTH_WARN
, ss
.str()));
3469 if (cct
->_conf
->mon_osd_warn_op_age
> 0 &&
3470 osd_sum
.op_queue_age_hist
.upper_bound() / 1000.0 >
3471 cct
->_conf
->mon_osd_warn_op_age
) {
3472 auto sum
= _warn_slow_request_histogram(
3473 cct
, osd_sum
.op_queue_age_hist
, "", summary
, NULL
);
3474 if (sum
.first
> 0 || sum
.second
> 0) {
3475 if (sum
.first
> 0) {
3477 ss
<< sum
.first
<< " requests are blocked > "
3478 << cct
->_conf
->mon_osd_warn_op_age
3480 summary
.push_back(make_pair(HEALTH_WARN
, ss
.str()));
3482 if (sum
.second
> 0) {
3484 ss
<< sum
.second
<< " requests are blocked > "
3485 << (cct
->_conf
->mon_osd_warn_op_age
*
3486 cct
->_conf
->mon_osd_err_op_age_ratio
)
3488 summary
.push_back(make_pair(HEALTH_ERR
, ss
.str()));
3492 unsigned num_warn
= 0, num_err
= 0;
3493 // do per-osd warnings
3494 for (auto p
= osd_stat
.begin();
3495 p
!= osd_stat
.end();
3497 auto sum
= _warn_slow_request_histogram(
3499 p
->second
.op_queue_age_hist
,
3500 string(" on osd.") + stringify(p
->first
),
3509 ss2
<< num_err
<< " osds have very slow requests";
3510 summary
.push_back(make_pair(HEALTH_ERR
, ss2
.str()));
3511 detail
->push_back(make_pair(HEALTH_ERR
, ss2
.str()));
3515 ss2
<< num_warn
<< " osds have slow requests";
3516 summary
.push_back(make_pair(HEALTH_WARN
, ss2
.str()));
3517 detail
->push_back(make_pair(HEALTH_WARN
, ss2
.str()));
3525 overall_recovery_summary(NULL
, &sl
);
3526 for (auto p
= sl
.begin(); p
!= sl
.end(); ++p
) {
3527 summary
.push_back(make_pair(HEALTH_WARN
, "recovery " + *p
));
3529 detail
->push_back(make_pair(HEALTH_WARN
, "recovery " + *p
));
3532 // near-target max pools
3533 auto& pools
= osdmap
.get_pools();
3534 for (auto p
= pools
.begin();
3535 p
!= pools
.end(); ++p
) {
3536 if ((!p
->second
.target_max_objects
&& !p
->second
.target_max_bytes
) ||
3537 !pg_pool_sum
.count(p
->first
))
3539 bool nearfull
= false;
3540 const string
& name
= osdmap
.get_pool_name(p
->first
);
3541 const pool_stat_t
& st
= get_pg_pool_sum_stat(p
->first
);
3542 uint64_t ratio
= p
->second
.cache_target_full_ratio_micro
+
3543 ((1000000 - p
->second
.cache_target_full_ratio_micro
) *
3544 cct
->_conf
->mon_cache_target_full_warn_ratio
);
3545 if (p
->second
.target_max_objects
&&
3546 (uint64_t)(st
.stats
.sum
.num_objects
-
3547 st
.stats
.sum
.num_objects_hit_set_archive
) >
3548 p
->second
.target_max_objects
* (ratio
/ 1000000.0)) {
3552 ss
<< "cache pool '" << name
<< "' with "
3553 << si_t(st
.stats
.sum
.num_objects
)
3554 << " objects at/near target max "
3555 << si_t(p
->second
.target_max_objects
) << " objects";
3556 detail
->push_back(make_pair(HEALTH_WARN
, ss
.str()));
3559 if (p
->second
.target_max_bytes
&&
3560 (uint64_t)(st
.stats
.sum
.num_bytes
-
3561 st
.stats
.sum
.num_bytes_hit_set_archive
) >
3562 p
->second
.target_max_bytes
* (ratio
/ 1000000.0)) {
3566 ss
<< "cache pool '" << name
3567 << "' with " << si_t(st
.stats
.sum
.num_bytes
)
3568 << "B at/near target max "
3569 << si_t(p
->second
.target_max_bytes
) << "B";
3570 detail
->push_back(make_pair(HEALTH_WARN
, ss
.str()));
3575 ss
<< "'" << name
<< "' at/near target max";
3576 summary
.push_back(make_pair(HEALTH_WARN
, ss
.str()));
3581 if (pg_sum
.stats
.sum
.num_scrub_errors
) {
3583 ss
<< pg_sum
.stats
.sum
.num_scrub_errors
<< " scrub errors";
3584 summary
.push_back(make_pair(HEALTH_ERR
, ss
.str()));
3586 detail
->push_back(make_pair(HEALTH_ERR
, ss
.str()));
3591 int num_in
= osdmap
.get_num_in_osds();
3592 int sum_pg_up
= MAX(pg_sum
.up
, static_cast<int32_t>(pg_stat
.size()));
3593 int sum_objects
= pg_sum
.stats
.sum
.num_objects
;
3594 if (sum_objects
< cct
->_conf
->mon_pg_warn_min_objects
) {
3597 if (num_in
&& cct
->_conf
->mon_pg_warn_min_per_osd
> 0) {
3598 int per
= sum_pg_up
/ num_in
;
3599 if (per
< cct
->_conf
->mon_pg_warn_min_per_osd
&& per
) {
3601 ss
<< "too few PGs per OSD (" << per
<< " < min " << cct
->_conf
->mon_pg_warn_min_per_osd
<< ")";
3602 summary
.push_back(make_pair(HEALTH_WARN
, ss
.str()));
3604 detail
->push_back(make_pair(HEALTH_WARN
, ss
.str()));
3607 if (num_in
&& cct
->_conf
->mon_pg_warn_max_per_osd
> 0) {
3608 int per
= sum_pg_up
/ num_in
;
3609 if (per
> cct
->_conf
->mon_pg_warn_max_per_osd
) {
3611 ss
<< "too many PGs per OSD (" << per
<< " > max " << cct
->_conf
->mon_pg_warn_max_per_osd
<< ")";
3612 summary
.push_back(make_pair(HEALTH_WARN
, ss
.str()));
3614 detail
->push_back(make_pair(HEALTH_WARN
, ss
.str()));
3617 if (!pg_stat
.empty()) {
3618 for (auto p
= pg_pool_sum
.begin();
3619 p
!= pg_pool_sum
.end();
3621 const pg_pool_t
*pi
= osdmap
.get_pg_pool(p
->first
);
3623 continue; // in case osdmap changes haven't propagated to PGMap yet
3624 const string
& name
= osdmap
.get_pool_name(p
->first
);
3625 if (pi
->get_pg_num() > pi
->get_pgp_num() &&
3626 !(name
.find(".DELETED") != string::npos
&&
3627 cct
->_conf
->mon_fake_pool_delete
)) {
3629 ss
<< "pool " << name
<< " pg_num "
3630 << pi
->get_pg_num() << " > pgp_num " << pi
->get_pgp_num();
3631 summary
.push_back(make_pair(HEALTH_WARN
, ss
.str()));
3633 detail
->push_back(make_pair(HEALTH_WARN
, ss
.str()));
3635 int average_objects_per_pg
= pg_sum
.stats
.sum
.num_objects
/ pg_stat
.size();
3636 if (average_objects_per_pg
> 0 &&
3637 pg_sum
.stats
.sum
.num_objects
>= cct
->_conf
->mon_pg_warn_min_objects
&&
3638 p
->second
.stats
.sum
.num_objects
>= cct
->_conf
->mon_pg_warn_min_pool_objects
) {
3639 int objects_per_pg
= p
->second
.stats
.sum
.num_objects
/ pi
->get_pg_num();
3640 float ratio
= (float)objects_per_pg
/ (float)average_objects_per_pg
;
3641 if (cct
->_conf
->mon_pg_warn_max_object_skew
> 0 &&
3642 ratio
> cct
->_conf
->mon_pg_warn_max_object_skew
) {
3644 ss
<< "pool " << name
<< " has many more objects per pg than average (too few pgs?)";
3645 summary
.push_back(make_pair(HEALTH_WARN
, ss
.str()));
3648 ss
<< "pool " << name
<< " objects per pg ("
3649 << objects_per_pg
<< ") is more than " << ratio
<< " times cluster average ("
3650 << average_objects_per_pg
<< ")";
3651 detail
->push_back(make_pair(HEALTH_WARN
, ss
.str()));
3658 for (auto it
: pools
) {
3659 auto it2
= pg_pool_sum
.find(it
.first
);
3660 if (it2
== pg_pool_sum
.end()) {
3663 const pool_stat_t
*pstat
= &it2
->second
;
3664 const object_stat_sum_t
& sum
= pstat
->stats
.sum
;
3665 const string
& pool_name
= osdmap
.get_pool_name(it
.first
);
3666 const pg_pool_t
&pool
= it
.second
;
3668 float warn_threshold
= (float)g_conf
->mon_pool_quota_warn_threshold
/100;
3669 float crit_threshold
= (float)g_conf
->mon_pool_quota_crit_threshold
/100;
3671 if (pool
.quota_max_objects
> 0) {
3673 health_status_t status
= HEALTH_OK
;
3674 if ((uint64_t)sum
.num_objects
>= pool
.quota_max_objects
) {
3675 } else if (crit_threshold
> 0 &&
3676 sum
.num_objects
>= pool
.quota_max_objects
*crit_threshold
) {
3677 ss
<< "pool '" << pool_name
3678 << "' has " << sum
.num_objects
<< " objects"
3679 << " (max " << pool
.quota_max_objects
<< ")";
3680 status
= HEALTH_ERR
;
3681 } else if (warn_threshold
> 0 &&
3682 sum
.num_objects
>= pool
.quota_max_objects
*warn_threshold
) {
3683 ss
<< "pool '" << pool_name
3684 << "' has " << sum
.num_objects
<< " objects"
3685 << " (max " << pool
.quota_max_objects
<< ")";
3686 status
= HEALTH_WARN
;
3688 if (status
!= HEALTH_OK
) {
3689 pair
<health_status_t
,string
> s(status
, ss
.str());
3690 summary
.push_back(s
);
3692 detail
->push_back(s
);
3696 if (pool
.quota_max_bytes
> 0) {
3697 health_status_t status
= HEALTH_OK
;
3699 if ((uint64_t)sum
.num_bytes
>= pool
.quota_max_bytes
) {
3700 } else if (crit_threshold
> 0 &&
3701 sum
.num_bytes
>= pool
.quota_max_bytes
*crit_threshold
) {
3702 ss
<< "pool '" << pool_name
3703 << "' has " << si_t(sum
.num_bytes
) << " bytes"
3704 << " (max " << si_t(pool
.quota_max_bytes
) << ")";
3705 status
= HEALTH_ERR
;
3706 } else if (warn_threshold
> 0 &&
3707 sum
.num_bytes
>= pool
.quota_max_bytes
*warn_threshold
) {
3708 ss
<< "pool '" << pool_name
3709 << "' has " << si_t(sum
.num_bytes
) << " bytes"
3710 << " (max " << si_t(pool
.quota_max_bytes
) << ")";
3711 status
= HEALTH_WARN
;
3713 if (status
!= HEALTH_OK
) {
3714 pair
<health_status_t
,string
> s(status
, ss
.str());
3715 summary
.push_back(s
);
3717 detail
->push_back(s
);
3722 print_unscrubbed_pgs(pg_stat
, summary
, detail
, cct
);
3725 int process_pg_map_command(
3726 const string
& orig_prefix
,
3727 const map
<string
,cmd_vartype
>& orig_cmdmap
,
3728 const PGMap
& pg_map
,
3729 const OSDMap
& osdmap
,
3734 string prefix
= orig_prefix
;
3735 map
<string
,cmd_vartype
> cmdmap
= orig_cmdmap
;
3737 // perhaps these would be better in the parsing, but it's weird
3738 bool primary
= false;
3739 if (prefix
== "pg dump_json") {
3741 v
.push_back(string("all"));
3742 cmd_putval(g_ceph_context
, cmdmap
, "format", string("json"));
3743 cmd_putval(g_ceph_context
, cmdmap
, "dumpcontents", v
);
3745 } else if (prefix
== "pg dump_pools_json") {
3747 v
.push_back(string("pools"));
3748 cmd_putval(g_ceph_context
, cmdmap
, "format", string("json"));
3749 cmd_putval(g_ceph_context
, cmdmap
, "dumpcontents", v
);
3751 } else if (prefix
== "pg ls-by-primary") {
3754 } else if (prefix
== "pg ls-by-osd") {
3756 } else if (prefix
== "pg ls-by-pool") {
3759 cmd_getval(g_ceph_context
, cmdmap
, "poolstr", poolstr
);
3760 int64_t pool
= osdmap
.lookup_pg_pool_name(poolstr
.c_str());
3762 *ss
<< "pool " << poolstr
<< " does not exist";
3765 cmd_putval(g_ceph_context
, cmdmap
, "pool", pool
);
3770 if (prefix
== "pg stat") {
3772 f
->open_object_section("pg_summary");
3773 pg_map
.print_oneline_summary(f
, NULL
);
3783 if (prefix
== "pg getmap") {
3784 pg_map
.encode(*odata
);
3785 *ss
<< "got pgmap version " << pg_map
.version
;
3789 if (prefix
== "pg dump") {
3791 vector
<string
> dumpcontents
;
3793 if (cmd_getval(g_ceph_context
, cmdmap
, "dumpcontents", dumpcontents
)) {
3794 copy(dumpcontents
.begin(), dumpcontents
.end(),
3795 inserter(what
, what
.end()));
3800 if (what
.count("all")) {
3801 f
->open_object_section("pg_map");
3804 } else if (what
.count("summary") || what
.count("sum")) {
3805 f
->open_object_section("pg_map");
3806 pg_map
.dump_basic(f
);
3809 if (what
.count("pools")) {
3810 pg_map
.dump_pool_stats(f
);
3812 if (what
.count("osds")) {
3813 pg_map
.dump_osd_stats(f
);
3815 if (what
.count("pgs")) {
3816 pg_map
.dump_pg_stats(f
, false);
3818 if (what
.count("pgs_brief")) {
3819 pg_map
.dump_pg_stats(f
, true);
3821 if (what
.count("delta")) {
3822 f
->open_object_section("delta");
3823 pg_map
.dump_delta(f
);
3829 if (what
.count("all")) {
3831 } else if (what
.count("summary") || what
.count("sum")) {
3832 pg_map
.dump_basic(ds
);
3833 pg_map
.dump_pg_sum_stats(ds
, true);
3834 pg_map
.dump_osd_sum_stats(ds
);
3836 if (what
.count("pgs_brief")) {
3837 pg_map
.dump_pg_stats(ds
, true);
3840 if (what
.count("pgs")) {
3841 pg_map
.dump_pg_stats(ds
, false);
3844 if (what
.count("pools")) {
3845 pg_map
.dump_pool_stats(ds
, header
);
3847 if (what
.count("osds")) {
3848 pg_map
.dump_osd_stats(ds
);
3853 *ss
<< "dumped " << what
;
3857 if (prefix
== "pg ls") {
3860 vector
<string
>states
;
3862 cmd_getval(g_ceph_context
, cmdmap
, "pool", pool
);
3863 cmd_getval(g_ceph_context
, cmdmap
, "osd", osd
);
3864 cmd_getval(g_ceph_context
, cmdmap
, "states", states
);
3865 if (pool
>= 0 && !osdmap
.have_pg_pool(pool
)) {
3866 *ss
<< "pool " << pool
<< " does not exist";
3869 if (osd
>= 0 && !osdmap
.is_up(osd
)) {
3870 *ss
<< "osd " << osd
<< " is not up";
3874 states
.push_back("all");
3878 while (!states
.empty()) {
3879 string state_str
= states
.back();
3881 if (state_str
== "all") {
3885 int filter
= pg_string_state(state_str
);
3887 *ss
<< "'" << state_str
<< "' is not a valid pg state,"
3888 << " available choices: " << pg_state_string(0xFFFFFFFF);
3897 pg_map
.get_filtered_pg_stats(state
, pool
, osd
, primary
, pgs
);
3899 if (f
&& !pgs
.empty()) {
3900 pg_map
.dump_filtered_pg_stats(f
, pgs
);
3902 } else if (!pgs
.empty()) {
3903 pg_map
.dump_filtered_pg_stats(ds
, pgs
);
3909 if (prefix
== "pg dump_stuck") {
3910 vector
<string
> stuckop_vec
;
3911 cmd_getval(g_ceph_context
, cmdmap
, "stuckops", stuckop_vec
);
3912 if (stuckop_vec
.empty())
3913 stuckop_vec
.push_back("unclean");
3915 cmd_getval(g_ceph_context
, cmdmap
, "threshold", threshold
,
3916 int64_t(g_conf
->mon_pg_stuck_threshold
));
3918 r
= pg_map
.dump_stuck_pg_stats(ds
, f
, (int)threshold
, stuckop_vec
);
3927 if (prefix
== "pg debug") {
3929 cmd_getval(g_ceph_context
, cmdmap
, "debugop", debugop
,
3930 string("unfound_objects_exist"));
3931 if (debugop
== "unfound_objects_exist") {
3932 bool unfound_objects_exist
= false;
3933 for (const auto& p
: pg_map
.pg_stat
) {
3934 if (p
.second
.stats
.sum
.num_objects_unfound
> 0) {
3935 unfound_objects_exist
= true;
3939 if (unfound_objects_exist
)
3946 if (debugop
== "degraded_pgs_exist") {
3947 bool degraded_pgs_exist
= false;
3948 for (const auto& p
: pg_map
.pg_stat
) {
3949 if (p
.second
.stats
.sum
.num_objects_degraded
> 0) {
3950 degraded_pgs_exist
= true;
3954 if (degraded_pgs_exist
)
3963 if (prefix
== "osd perf") {
3965 f
->open_object_section("osdstats");
3966 pg_map
.dump_osd_perf_stats(f
);
3970 pg_map
.print_osd_perf_stats(&ds
);
3976 if (prefix
== "osd blocked-by") {
3978 f
->open_object_section("osd_blocked_by");
3979 pg_map
.dump_osd_blocked_by_stats(f
);
3983 pg_map
.print_osd_blocked_by_stats(&ds
);
3989 if (prefix
== "osd pool stats") {
3991 cmd_getval(g_ceph_context
, cmdmap
, "name", pool_name
);
3993 int64_t poolid
= -ENOENT
;
3994 bool one_pool
= false;
3995 if (!pool_name
.empty()) {
3996 poolid
= osdmap
.lookup_pg_pool_name(pool_name
);
3998 assert(poolid
== -ENOENT
);
3999 *ss
<< "unrecognized pool '" << pool_name
<< "'";
4008 f
->open_array_section("pool_stats");
4010 if (osdmap
.get_pools().empty()) {
4011 *ss
<< "there are no pools!";
4016 for (auto& p
: osdmap
.get_pools()) {
4020 pool_name
= osdmap
.get_pool_name(poolid
);
4023 f
->open_object_section("pool");
4024 f
->dump_string("pool_name", pool_name
.c_str());
4025 f
->dump_int("pool_id", poolid
);
4026 f
->open_object_section("recovery");
4031 pg_map
.pool_recovery_summary(f
, &sl
, poolid
);
4032 if (!f
&& !sl
.empty()) {
4034 tss
<< " " << p
<< "\n";
4039 f
->open_object_section("recovery_rate");
4043 pg_map
.pool_recovery_rate_summary(f
, &rss
, poolid
);
4044 if (!f
&& !rss
.str().empty())
4045 tss
<< " recovery io " << rss
.str() << "\n";
4049 f
->open_object_section("client_io_rate");
4054 pg_map
.pool_client_io_rate_summary(f
, &rss
, poolid
);
4055 if (!f
&& !rss
.str().empty())
4056 tss
<< " client io " << rss
.str() << "\n";
4058 // dump cache tier IO rate for cache pool
4059 const pg_pool_t
*pool
= osdmap
.get_pg_pool(poolid
);
4060 if (pool
->is_tier()) {
4063 f
->open_object_section("cache_io_rate");
4068 pg_map
.pool_cache_io_rate_summary(f
, &rss
, poolid
);
4069 if (!f
&& !rss
.str().empty())
4070 tss
<< " cache tier io " << rss
.str() << "\n";
4076 rs
<< "pool " << pool_name
<< " id " << poolid
<< "\n";
4077 if (!tss
.str().empty())
4078 rs
<< tss
.str() << "\n";
4080 rs
<< " nothing is going on\n\n";
4092 odata
->append(rs
.str());
4100 void PGMapUpdater::check_osd_map(const OSDMap::Incremental
&osd_inc
,
4101 std::set
<int> *need_check_down_pg_osds
,
4102 std::map
<int,utime_t
> *last_osd_report
,
4104 PGMap::Incremental
*pending_inc
)
4106 for (const auto &p
: osd_inc
.new_weight
) {
4107 if (p
.second
== CEPH_OSD_OUT
) {
4108 dout(10) << __func__
<< " osd." << p
.first
<< " went OUT" << dendl
;
4109 auto j
= pg_map
->osd_epochs
.find(p
.first
);
4110 if (j
!= pg_map
->osd_epochs
.end())
4111 pending_inc
->stat_osd_out(p
.first
, j
->second
);
4115 // this is conservative: we want to know if any osds (maybe) got marked down.
4116 for (const auto &p
: osd_inc
.new_state
) {
4117 if (p
.second
& CEPH_OSD_UP
) { // true if marked up OR down,
4118 // but we're too lazy to check
4120 need_check_down_pg_osds
->insert(p
.first
);
4122 // clear out the last_osd_report for this OSD
4123 auto report
= last_osd_report
->find(p
.first
);
4124 if (report
!= last_osd_report
->end()) {
4125 last_osd_report
->erase(report
);
4128 // clear out osd_stat slow request histogram
4129 dout(20) << __func__
<< " clearing osd." << p
.first
4130 << " request histogram" << dendl
;
4131 pending_inc
->stat_osd_down_up(p
.first
, osd_inc
.epoch
, *pg_map
);
4134 if (p
.second
& CEPH_OSD_EXISTS
) {
4135 // whether it was created *or* destroyed, we can safely drop
4136 // it's osd_stat_t record.
4137 dout(10) << __func__
<< " osd." << p
.first
4138 << " created or destroyed" << dendl
;
4139 pending_inc
->rm_stat(p
.first
);
4141 // and adjust full, nearfull set
4142 pg_map
->nearfull_osds
.erase(p
.first
);
4143 pg_map
->full_osds
.erase(p
.first
);
4148 void PGMapUpdater::check_osd_map(
4150 const OSDMap
& osdmap
,
4152 PGMap::Incremental
*pending_inc
)
4154 for (auto& p
: pgmap
.osd_stat
) {
4155 if (!osdmap
.exists(p
.first
)) {
4157 pending_inc
->rm_stat(p
.first
);
4158 } else if (osdmap
.is_out(p
.first
)) {
4160 if (p
.second
.kb
!= 0) {
4161 auto j
= pgmap
.osd_epochs
.find(p
.first
);
4162 if (j
!= pgmap
.osd_epochs
.end()) {
4163 pending_inc
->stat_osd_out(p
.first
, j
->second
);
4166 } else if (!osdmap
.is_up(p
.first
)) {
4167 // zero the op_queue_age_hist
4168 if (!p
.second
.op_queue_age_hist
.empty()) {
4169 pending_inc
->stat_osd_down_up(p
.first
, osdmap
.get_epoch(), pgmap
);
4174 // deleted pgs (pools)?
4175 for (auto& p
: pgmap
.pg_pool_sum
) {
4176 if (!osdmap
.have_pg_pool(p
.first
)) {
4177 ldout(cct
, 10) << __func__
<< " pool " << p
.first
<< " gone, removing pgs"
4179 for (auto& q
: pgmap
.pg_stat
) {
4180 if (q
.first
.pool() == (uint64_t)p
.first
) {
4181 pending_inc
->pg_remove
.insert(q
.first
);
4184 auto q
= pending_inc
->pg_stat_updates
.begin();
4185 while (q
!= pending_inc
->pg_stat_updates
.end()) {
4186 if (q
->first
.pool() == (uint64_t)p
.first
) {
4187 q
= pending_inc
->pg_stat_updates
.erase(q
);
4195 // new pgs (split or new pool)?
4196 for (auto& p
: osdmap
.get_pools()) {
4197 int64_t poolid
= p
.first
;
4198 const pg_pool_t
& pi
= p
.second
;
4199 auto q
= pgmap
.num_pg_by_pool
.find(poolid
);
4200 unsigned my_pg_num
= 0;
4201 if (q
!= pgmap
.num_pg_by_pool
.end())
4202 my_pg_num
= q
->second
;
4203 unsigned pg_num
= pi
.get_pg_num();
4204 if (my_pg_num
!= pg_num
) {
4205 ldout(cct
,10) << __func__
<< " pool " << poolid
<< " pg_num " << pg_num
4206 << " != my pg_num " << my_pg_num
<< dendl
;
4207 for (unsigned ps
= my_pg_num
; ps
< pg_num
; ++ps
) {
4208 pg_t
pgid(ps
, poolid
);
4209 if (pending_inc
->pg_stat_updates
.count(pgid
) == 0) {
4210 ldout(cct
,20) << __func__
<< " adding " << pgid
<< dendl
;
4211 pg_stat_t
&stats
= pending_inc
->pg_stat_updates
[pgid
];
4212 stats
.last_fresh
= osdmap
.get_modified();
4213 stats
.last_active
= osdmap
.get_modified();
4214 stats
.last_change
= osdmap
.get_modified();
4215 stats
.last_peered
= osdmap
.get_modified();
4216 stats
.last_clean
= osdmap
.get_modified();
4217 stats
.last_unstale
= osdmap
.get_modified();
4218 stats
.last_undegraded
= osdmap
.get_modified();
4219 stats
.last_fullsized
= osdmap
.get_modified();
4220 stats
.last_scrub_stamp
= osdmap
.get_modified();
4221 stats
.last_deep_scrub_stamp
= osdmap
.get_modified();
4222 stats
.last_clean_scrub_stamp
= osdmap
.get_modified();
4229 void PGMapUpdater::register_pg(
4230 const OSDMap
&osd_map
,
4231 pg_t pgid
, epoch_t epoch
,
4233 const PGMap
&pg_map
,
4234 PGMap::Incremental
*pending_inc
)
4238 auto parent_stat
= pg_map
.pg_stat
.end();
4242 // remove most significant bit
4243 int msb
= cbits(parent
.ps());
4246 parent
.set_ps(parent
.ps() & ~(1<<(msb
-1)));
4248 dout(30) << " is " << pgid
<< " parent " << parent
<< " ?" << dendl
;
4249 parent_stat
= pg_map
.pg_stat
.find(parent
);
4250 if (parent_stat
!= pg_map
.pg_stat
.end() &&
4251 parent_stat
->second
.state
!= PG_STATE_CREATING
) {
4252 dout(10) << " parent is " << parent
<< dendl
;
4258 pg_stat_t
&stats
= pending_inc
->pg_stat_updates
[pgid
];
4259 stats
.state
= PG_STATE_CREATING
;
4260 stats
.created
= epoch
;
4261 stats
.parent
= parent
;
4262 stats
.parent_split_bits
= split_bits
;
4263 stats
.mapping_epoch
= epoch
;
4265 if (parent_stat
!= pg_map
.pg_stat
.end()) {
4266 const pg_stat_t
&ps
= parent_stat
->second
;
4267 stats
.last_fresh
= ps
.last_fresh
;
4268 stats
.last_active
= ps
.last_active
;
4269 stats
.last_change
= ps
.last_change
;
4270 stats
.last_peered
= ps
.last_peered
;
4271 stats
.last_clean
= ps
.last_clean
;
4272 stats
.last_unstale
= ps
.last_unstale
;
4273 stats
.last_undegraded
= ps
.last_undegraded
;
4274 stats
.last_fullsized
= ps
.last_fullsized
;
4275 stats
.last_scrub_stamp
= ps
.last_scrub_stamp
;
4276 stats
.last_deep_scrub_stamp
= ps
.last_deep_scrub_stamp
;
4277 stats
.last_clean_scrub_stamp
= ps
.last_clean_scrub_stamp
;
4279 utime_t now
= osd_map
.get_modified();
4280 stats
.last_fresh
= now
;
4281 stats
.last_active
= now
;
4282 stats
.last_change
= now
;
4283 stats
.last_peered
= now
;
4284 stats
.last_clean
= now
;
4285 stats
.last_unstale
= now
;
4286 stats
.last_undegraded
= now
;
4287 stats
.last_fullsized
= now
;
4288 stats
.last_scrub_stamp
= now
;
4289 stats
.last_deep_scrub_stamp
= now
;
4290 stats
.last_clean_scrub_stamp
= now
;
4293 osd_map
.pg_to_up_acting_osds(
4298 &stats
.acting_primary
);
4300 if (split_bits
== 0) {
4301 dout(10) << __func__
<< " will create " << pgid
4302 << " primary " << stats
.acting_primary
4303 << " acting " << stats
.acting
4306 dout(10) << __func__
<< " will create " << pgid
4307 << " primary " << stats
.acting_primary
4308 << " acting " << stats
.acting
4309 << " parent " << parent
4310 << " by " << split_bits
<< " bits"
4315 void PGMapUpdater::register_new_pgs(
4316 const OSDMap
&osd_map
,
4317 const PGMap
&pg_map
,
4318 PGMap::Incremental
*pending_inc
)
4320 epoch_t epoch
= osd_map
.get_epoch();
4321 dout(10) << __func__
<< " checking pg pools for osdmap epoch " << epoch
4322 << ", last_pg_scan " << pg_map
.last_pg_scan
<< dendl
;
4325 const auto &pools
= osd_map
.get_pools();
4327 for (const auto &p
: pools
) {
4328 int64_t poolid
= p
.first
;
4329 const pg_pool_t
&pool
= p
.second
;
4330 int ruleno
= osd_map
.crush
->find_rule(pool
.get_crush_rule(),
4331 pool
.get_type(), pool
.get_size());
4332 if (ruleno
< 0 || !osd_map
.crush
->rule_exists(ruleno
))
4335 if (pool
.get_last_change() <= pg_map
.last_pg_scan
||
4336 pool
.get_last_change() <= pending_inc
->pg_scan
) {
4337 dout(10) << " no change in pool " << poolid
<< " " << pool
<< dendl
;
4341 dout(10) << __func__
<< " scanning pool " << poolid
4342 << " " << pool
<< dendl
;
4344 // first pgs in this pool
4345 bool new_pool
= pg_map
.pg_pool_sum
.count(poolid
) == 0;
4347 for (ps_t ps
= 0; ps
< pool
.get_pg_num(); ps
++) {
4348 pg_t
pgid(ps
, poolid
, -1);
4349 if (pg_map
.pg_stat
.count(pgid
)) {
4350 dout(20) << "register_new_pgs have " << pgid
<< dendl
;
4354 register_pg(osd_map
, pgid
, pool
.get_last_change(), new_pool
,
4355 pg_map
, pending_inc
);
4360 for (const auto &p
: pg_map
.creating_pgs
) {
4361 if (p
.preferred() >= 0) {
4362 dout(20) << " removing creating_pg " << p
4363 << " because it is localized and obsolete" << dendl
;
4364 pending_inc
->pg_remove
.insert(p
);
4366 } else if (!osd_map
.have_pg_pool(p
.pool())) {
4367 dout(20) << " removing creating_pg " << p
4368 << " because containing pool deleted" << dendl
;
4369 pending_inc
->pg_remove
.insert(p
);
4375 for (const auto &p
: pg_map
.pg_stat
) {
4376 if (!osd_map
.have_pg_pool(p
.first
.pool())) {
4377 dout(20) << " removing pg_stat " << p
.first
<< " because "
4378 << "containing pool deleted" << dendl
;
4379 pending_inc
->pg_remove
.insert(p
.first
);
4381 } else if (p
.first
.preferred() >= 0) {
4382 dout(20) << " removing localized pg " << p
.first
<< dendl
;
4383 pending_inc
->pg_remove
.insert(p
.first
);
4388 // we don't want to redo this work if we can avoid it.
4389 pending_inc
->pg_scan
= epoch
;
4391 dout(10) << "register_new_pgs registered " << created
<< " new pgs, removed "
4392 << removed
<< " uncreated pgs" << dendl
;
4396 void PGMapUpdater::update_creating_pgs(
4397 const OSDMap
&osd_map
,
4398 const PGMap
&pg_map
,
4399 PGMap::Incremental
*pending_inc
)
4401 dout(10) << __func__
<< " to " << pg_map
.creating_pgs
.size()
4402 << " pgs, osdmap epoch " << osd_map
.get_epoch()
4405 unsigned changed
= 0;
4406 for (auto p
= pg_map
.creating_pgs
.begin();
4407 p
!= pg_map
.creating_pgs
.end();
4411 auto q
= pg_map
.pg_stat
.find(pgid
);
4412 assert(q
!= pg_map
.pg_stat
.end());
4413 const pg_stat_t
*s
= &q
->second
;
4415 if (s
->parent_split_bits
)
4418 vector
<int> up
, acting
;
4419 int up_primary
, acting_primary
;
4420 osd_map
.pg_to_up_acting_osds(
4428 up_primary
!= s
->up_primary
||
4429 acting
!= s
->acting
||
4430 acting_primary
!= s
->acting_primary
) {
4431 pg_stat_t
*ns
= &pending_inc
->pg_stat_updates
[pgid
];
4432 if (osd_map
.get_epoch() > ns
->reported_epoch
) {
4433 dout(20) << __func__
<< " " << pgid
<< " "
4434 << " acting_primary: " << s
->acting_primary
4435 << " -> " << acting_primary
4436 << " acting: " << s
->acting
<< " -> " << acting
4437 << " up_primary: " << s
->up_primary
<< " -> " << up_primary
4438 << " up: " << s
->up
<< " -> " << up
4441 // only initialize if it wasn't already a pending update
4442 if (ns
->reported_epoch
== 0)
4445 // note epoch if the target of the create message changed
4446 if (acting_primary
!= ns
->acting_primary
)
4447 ns
->mapping_epoch
= osd_map
.get_epoch();
4450 ns
->up_primary
= up_primary
;
4451 ns
->acting
= acting
;
4452 ns
->acting_primary
= acting_primary
;
4456 dout(20) << __func__
<< " " << pgid
<< " has pending update from newer"
4457 << " epoch " << ns
->reported_epoch
4463 dout(10) << __func__
<< " " << changed
<< " pgs changed primary" << dendl
;
4467 static void _try_mark_pg_stale(
4468 const OSDMap
& osdmap
,
4470 const pg_stat_t
& cur
,
4471 PGMap::Incremental
*pending_inc
)
4473 if ((cur
.state
& PG_STATE_STALE
) == 0 &&
4474 cur
.acting_primary
!= -1 &&
4475 osdmap
.is_down(cur
.acting_primary
)) {
4477 auto q
= pending_inc
->pg_stat_updates
.find(pgid
);
4478 if (q
!= pending_inc
->pg_stat_updates
.end()) {
4479 if ((q
->second
.acting_primary
== cur
.acting_primary
) ||
4480 ((q
->second
.state
& PG_STATE_STALE
) == 0 &&
4481 q
->second
.acting_primary
!= -1 &&
4482 osdmap
.is_down(q
->second
.acting_primary
))) {
4483 newstat
= &q
->second
;
4485 // pending update is no longer down or already stale
4489 newstat
= &pending_inc
->pg_stat_updates
[pgid
];
4492 dout(10) << __func__
<< " marking pg " << pgid
4493 << " stale (acting_primary " << newstat
->acting_primary
4495 newstat
->state
|= PG_STATE_STALE
;
4496 newstat
->last_unstale
= ceph_clock_now();
4500 void PGMapUpdater::check_down_pgs(
4501 const OSDMap
&osdmap
,
4502 const PGMap
&pg_map
,
4504 const set
<int>& need_check_down_pg_osds
,
4505 PGMap::Incremental
*pending_inc
)
4507 // if a large number of osds changed state, just iterate over the whole
4509 if (need_check_down_pg_osds
.size() > (unsigned)osdmap
.get_num_osds() *
4510 g_conf
->mon_pg_check_down_all_threshold
) {
4515 for (const auto& p
: pg_map
.pg_stat
) {
4516 _try_mark_pg_stale(osdmap
, p
.first
, p
.second
, pending_inc
);
4519 for (auto osd
: need_check_down_pg_osds
) {
4520 if (osdmap
.is_down(osd
)) {
4521 auto p
= pg_map
.pg_by_osd
.find(osd
);
4522 if (p
== pg_map
.pg_by_osd
.end()) {
4525 for (auto pgid
: p
->second
) {
4526 const pg_stat_t
&stat
= pg_map
.pg_stat
.at(pgid
);
4527 assert(stat
.acting_primary
== osd
);
4528 _try_mark_pg_stale(osdmap
, pgid
, stat
, pending_inc
);
4535 int reweight::by_utilization(
4536 const OSDMap
&osdmap
,
4541 bool by_pg
, const set
<int64_t> *pools
,
4543 mempool::osdmap::map
<int32_t, uint32_t>* new_weights
,
4544 std::stringstream
*ss
,
4545 std::string
*out_str
,
4549 *ss
<< "You must give a percentage higher than 100. "
4550 "The reweighting threshold will be calculated as <average-utilization> "
4551 "times <input-percentage>. For example, an argument of 200 would "
4552 "reweight OSDs which are twice as utilized as the average OSD.\n";
4556 vector
<int> pgs_by_osd(osdmap
.get_max_osd());
4558 // Avoid putting a small number (or 0) in the denominator when calculating
4560 double average_util
;
4563 double weight_sum
= 0.0; // sum up the crush weights
4564 unsigned num_pg_copies
= 0;
4566 for (const auto& pg
: pgm
.pg_stat
) {
4567 if (pools
&& pools
->count(pg
.first
.pool()) == 0)
4569 for (const auto acting
: pg
.second
.acting
) {
4570 if (!osdmap
.exists(acting
)) {
4573 if (acting
>= (int)pgs_by_osd
.size())
4574 pgs_by_osd
.resize(acting
);
4575 if (pgs_by_osd
[acting
] == 0) {
4576 if (osdmap
.crush
->get_item_weightf(acting
) <= 0) {
4577 //skip if we currently can not identify item
4580 weight_sum
+= osdmap
.crush
->get_item_weightf(acting
);
4583 ++pgs_by_osd
[acting
];
4588 if (!num_osds
|| (num_pg_copies
/ num_osds
< g_conf
->mon_reweight_min_pgs_per_osd
)) {
4589 *ss
<< "Refusing to reweight: we only have " << num_pg_copies
4590 << " PGs across " << num_osds
<< " osds!\n";
4594 average_util
= (double)num_pg_copies
/ weight_sum
;
4596 // by osd utilization
4597 int num_osd
= MAX(1, pgm
.osd_stat
.size());
4598 if ((uint64_t)pgm
.osd_sum
.kb
* 1024 / num_osd
4599 < g_conf
->mon_reweight_min_bytes_per_osd
) {
4600 *ss
<< "Refusing to reweight: we only have " << pgm
.osd_sum
.kb
4601 << " kb across all osds!\n";
4604 if ((uint64_t)pgm
.osd_sum
.kb_used
* 1024 / num_osd
4605 < g_conf
->mon_reweight_min_bytes_per_osd
) {
4606 *ss
<< "Refusing to reweight: we only have " << pgm
.osd_sum
.kb_used
4607 << " kb used across all osds!\n";
4611 average_util
= (double)pgm
.osd_sum
.kb_used
/ (double)pgm
.osd_sum
.kb
;
4614 // adjust down only if we are above the threshold
4615 const double overload_util
= average_util
* (double)oload
/ 100.0;
4617 // but aggressively adjust weights up whenever possible.
4618 const double underload_util
= average_util
;
4620 const unsigned max_change
= (unsigned)(max_changef
* (double)0x10000);
4624 f
->open_object_section("reweight_by_utilization");
4625 f
->dump_int("overload_min", oload
);
4626 f
->dump_float("max_change", max_changef
);
4627 f
->dump_int("max_change_osds", max_osds
);
4628 f
->dump_float("average_utilization", average_util
);
4629 f
->dump_float("overload_utilization", overload_util
);
4631 oss
<< "oload " << oload
<< "\n";
4632 oss
<< "max_change " << max_changef
<< "\n";
4633 oss
<< "max_change_osds " << max_osds
<< "\n";
4635 oss
<< "average_utilization " << std::fixed
<< average_util
<< "\n";
4636 oss
<< "overload_utilization " << overload_util
<< "\n";
4638 int num_changed
= 0;
4640 // precompute util for each OSD
4641 std::vector
<std::pair
<int, float> > util_by_osd
;
4642 for (const auto& p
: pgm
.osd_stat
) {
4643 std::pair
<int, float> osd_util
;
4644 osd_util
.first
= p
.first
;
4646 if (p
.first
>= (int)pgs_by_osd
.size() ||
4647 pgs_by_osd
[p
.first
] == 0) {
4648 // skip if this OSD does not contain any pg
4649 // belonging to the specified pool(s).
4653 if (osdmap
.crush
->get_item_weightf(p
.first
) <= 0) {
4654 // skip if we are unable to locate item.
4658 osd_util
.second
= pgs_by_osd
[p
.first
] / osdmap
.crush
->get_item_weightf(p
.first
);
4660 osd_util
.second
= (double)p
.second
.kb_used
/ (double)p
.second
.kb
;
4662 util_by_osd
.push_back(osd_util
);
4665 // sort by absolute deviation from the mean utilization,
4666 // in descending order.
4667 std::sort(util_by_osd
.begin(), util_by_osd
.end(),
4668 [average_util
](std::pair
<int, float> l
, std::pair
<int, float> r
) {
4669 return abs(l
.second
- average_util
) > abs(r
.second
- average_util
);
4674 f
->open_array_section("reweights");
4676 for (const auto& p
: util_by_osd
) {
4677 unsigned weight
= osdmap
.get_weight(p
.first
);
4679 // skip if OSD is currently out
4682 float util
= p
.second
;
4684 if (util
>= overload_util
) {
4685 // Assign a lower weight to overloaded OSDs. The current weight
4686 // is a factor to take into account the original weights,
4687 // to represent e.g. differing storage capacities
4688 unsigned new_weight
= (unsigned)((average_util
/ util
) * (float)weight
);
4689 if (weight
> max_change
)
4690 new_weight
= MAX(new_weight
, weight
- max_change
);
4691 new_weights
->insert({p
.first
, new_weight
});
4693 f
->open_object_section("osd");
4694 f
->dump_int("osd", p
.first
);
4695 f
->dump_float("weight", (float)weight
/ (float)0x10000);
4696 f
->dump_float("new_weight", (float)new_weight
/ (float)0x10000);
4699 oss
<< "osd." << p
.first
<< " weight "
4700 << (float)weight
/ (float)0x10000 << " -> "
4701 << (float)new_weight
/ (float)0x10000 << "\n";
4703 if (++num_changed
>= max_osds
)
4706 if (!no_increasing
&& util
<= underload_util
) {
4707 // assign a higher weight.. if we can.
4708 unsigned new_weight
= (unsigned)((average_util
/ util
) * (float)weight
);
4709 new_weight
= MIN(new_weight
, weight
+ max_change
);
4710 if (new_weight
> 0x10000)
4711 new_weight
= 0x10000;
4712 if (new_weight
> weight
) {
4713 new_weights
->insert({p
.first
, new_weight
});
4714 oss
<< "osd." << p
.first
<< " weight "
4715 << (float)weight
/ (float)0x10000 << " -> "
4716 << (float)new_weight
/ (float)0x10000 << "\n";
4717 if (++num_changed
>= max_osds
)
4727 newmap
.deepish_copy_from(osdmap
);
4728 OSDMap::Incremental newinc
;
4729 newinc
.fsid
= newmap
.get_fsid();
4730 newinc
.epoch
= newmap
.get_epoch() + 1;
4731 newinc
.new_weight
= *new_weights
;
4732 newmap
.apply_incremental(newinc
);
4734 osdmap
.summarize_mapping_stats(&newmap
, pools
, out_str
, f
);
4740 *out_str
+= oss
.str();