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/Clock.h"
11 #include "common/Formatter.h"
12 #include "global/global_context.h"
13 #include "include/ceph_features.h"
14 #include "include/stringify.h"
16 #include "osd/osd_types.h"
17 #include "osd/OSDMap.h"
18 #include <boost/range/adaptor/reversed.hpp>
20 #define dout_context g_ceph_context
27 using std::ostringstream
;
30 using std::stringstream
;
33 using ceph::bufferlist
;
34 using TOPNSPC::common::cmd_getval
;
36 MEMPOOL_DEFINE_OBJECT_FACTORY(PGMapDigest
, pgmap_digest
, pgmap
);
37 MEMPOOL_DEFINE_OBJECT_FACTORY(PGMap
, pgmap
, pgmap
);
38 MEMPOOL_DEFINE_OBJECT_FACTORY(PGMap::Incremental
, pgmap_inc
, pgmap
);
41 // ---------------------
44 void PGMapDigest::encode(bufferlist
& bl
, uint64_t features
) const
46 // NOTE: see PGMap::encode_digest
48 if (!HAVE_FEATURE(features
, SERVER_MIMIC
)) {
50 } else if (!HAVE_FEATURE(features
, SERVER_NAUTILUS
)) {
53 ENCODE_START(v
, 1, bl
);
55 encode(num_pg_active
, bl
);
56 encode(num_pg_unknown
, bl
);
58 encode(pg_pool_sum
, bl
, features
);
59 encode(pg_sum
, bl
, features
);
60 encode(osd_sum
, bl
, features
);
62 encode(num_pg_by_state
, bl
);
64 uint32_t n
= num_pg_by_state
.size();
66 for (auto p
: num_pg_by_state
) {
67 encode((int32_t)p
.first
, bl
);
71 encode(num_pg_by_osd
, bl
);
72 encode(num_pg_by_pool
, bl
);
73 encode(osd_last_seq
, bl
);
74 encode(per_pool_sum_delta
, bl
, features
);
75 encode(per_pool_sum_deltas_stamps
, bl
);
76 encode(pg_sum_delta
, bl
, features
);
77 encode(stamp_delta
, bl
);
78 encode(avail_space_by_rule
, bl
);
80 encode(purged_snaps
, bl
);
83 encode(osd_sum_by_class
, bl
, features
);
88 void PGMapDigest::decode(bufferlist::const_iterator
& p
)
92 decode(num_pg_active
, p
);
93 decode(num_pg_unknown
, p
);
95 decode(pg_pool_sum
, p
);
99 decode(num_pg_by_state
, p
);
101 map
<int32_t, int32_t> nps
;
103 num_pg_by_state
.clear();
105 num_pg_by_state
[i
.first
] = i
.second
;
108 decode(num_pg_by_osd
, p
);
109 decode(num_pg_by_pool
, p
);
110 decode(osd_last_seq
, p
);
111 decode(per_pool_sum_delta
, p
);
112 decode(per_pool_sum_deltas_stamps
, p
);
113 decode(pg_sum_delta
, p
);
114 decode(stamp_delta
, p
);
115 decode(avail_space_by_rule
, p
);
117 decode(purged_snaps
, p
);
120 decode(osd_sum_by_class
, p
);
125 void PGMapDigest::dump(ceph::Formatter
*f
) const
127 f
->dump_unsigned("num_pg", num_pg
);
128 f
->dump_unsigned("num_pg_active", num_pg_active
);
129 f
->dump_unsigned("num_pg_unknown", num_pg_unknown
);
130 f
->dump_unsigned("num_osd", num_osd
);
131 f
->dump_object("pool_sum", pg_sum
);
132 f
->dump_object("osd_sum", osd_sum
);
134 f
->open_object_section("osd_sum_by_class");
135 for (auto& i
: osd_sum_by_class
) {
136 f
->dump_object(i
.first
.c_str(), i
.second
);
140 f
->open_array_section("pool_stats");
141 for (auto& p
: pg_pool_sum
) {
142 f
->open_object_section("pool_stat");
143 f
->dump_int("poolid", p
.first
);
144 auto q
= num_pg_by_pool
.find(p
.first
);
145 if (q
!= num_pg_by_pool
.end())
146 f
->dump_unsigned("num_pg", q
->second
);
151 f
->open_array_section("osd_stats");
153 // TODO: this isn't really correct since we can dump non-existent OSDs
154 // I dunno what osd_last_seq is set to in that case...
155 for (auto& p
: osd_last_seq
) {
156 f
->open_object_section("osd_stat");
157 f
->dump_int("osd", i
);
158 f
->dump_unsigned("seq", p
);
163 f
->open_array_section("num_pg_by_state");
164 for (auto& p
: num_pg_by_state
) {
165 f
->open_object_section("count");
166 f
->dump_string("state", pg_state_string(p
.first
));
167 f
->dump_unsigned("num", p
.second
);
171 f
->open_array_section("num_pg_by_osd");
172 for (auto& p
: num_pg_by_osd
) {
173 f
->open_object_section("count");
174 f
->dump_unsigned("osd", p
.first
);
175 f
->dump_unsigned("num_primary_pg", p
.second
.primary
);
176 f
->dump_unsigned("num_acting_pg", p
.second
.acting
);
177 f
->dump_unsigned("num_up_not_acting_pg", p
.second
.up_not_acting
);
181 f
->open_array_section("purged_snaps");
182 for (auto& j
: purged_snaps
) {
183 f
->open_object_section("pool");
184 f
->dump_int("pool", j
.first
);
185 f
->open_object_section("purged_snaps");
186 for (auto i
= j
.second
.begin(); i
!= j
.second
.end(); ++i
) {
187 f
->open_object_section("interval");
188 f
->dump_stream("start") << i
.get_start();
189 f
->dump_stream("length") << i
.get_len();
198 void PGMapDigest::generate_test_instances(list
<PGMapDigest
*>& ls
)
200 ls
.push_back(new PGMapDigest
);
203 inline std::string
percentify(const float& a
) {
204 std::stringstream ss
;
208 ss
<< std::fixed
<< std::setprecision(2) << a
;
212 void PGMapDigest::print_summary(ceph::Formatter
*f
, ostream
*out
) const
215 f
->open_array_section("pgs_by_state");
217 // list is descending numeric order (by count)
218 std::multimap
<int,uint64_t> state_by_count
; // count -> state
219 for (auto p
= num_pg_by_state
.begin();
220 p
!= num_pg_by_state
.end();
222 state_by_count
.insert(make_pair(p
->second
, p
->first
));
225 for (auto p
= state_by_count
.rbegin();
226 p
!= state_by_count
.rend();
229 f
->open_object_section("pgs_by_state_element");
230 f
->dump_string("state_name", pg_state_string(p
->second
));
231 f
->dump_unsigned("count", p
->first
);
239 f
->dump_unsigned("num_pgs", num_pg
);
240 f
->dump_unsigned("num_pools", pg_pool_sum
.size());
241 f
->dump_unsigned("num_objects", pg_sum
.stats
.sum
.num_objects
);
242 f
->dump_unsigned("data_bytes", pg_sum
.stats
.sum
.num_bytes
);
243 f
->dump_unsigned("bytes_used", osd_sum
.statfs
.get_used_raw());
244 f
->dump_unsigned("bytes_avail", osd_sum
.statfs
.available
);
245 f
->dump_unsigned("bytes_total", osd_sum
.statfs
.total
);
247 *out
<< " pools: " << pg_pool_sum
.size() << " pools, "
248 << num_pg
<< " pgs\n";
249 *out
<< " objects: " << si_u_t(pg_sum
.stats
.sum
.num_objects
) << " objects, "
250 << byte_u_t(pg_sum
.stats
.sum
.num_bytes
) << "\n";
252 << byte_u_t(osd_sum
.statfs
.get_used_raw()) << " used, "
253 << byte_u_t(osd_sum
.statfs
.available
) << " / "
254 << byte_u_t(osd_sum
.statfs
.total
) << " avail\n";
260 if (num_pg_unknown
> 0) {
261 float p
= (float)num_pg_unknown
/ (float)num_pg
;
263 f
->dump_float("unknown_pgs_ratio", p
);
266 snprintf(b
, sizeof(b
), "%.3lf", p
* 100.0);
267 *out
<< b
<< "% pgs unknown\n";
272 int num_pg_inactive
= num_pg
- num_pg_active
- num_pg_unknown
;
273 if (num_pg_inactive
> 0) {
274 float p
= (float)num_pg_inactive
/ (float)num_pg
;
276 f
->dump_float("inactive_pgs_ratio", p
);
282 snprintf(b
, sizeof(b
), "%.3f", p
* 100.0);
283 *out
<< b
<< "% pgs not active\n";
289 overall_recovery_summary(f
, &sl
);
290 if (!f
&& !sl
.empty()) {
291 for (auto p
= sl
.begin(); p
!= sl
.end(); ++p
) {
302 unsigned max_width
= 1;
303 for (auto p
= state_by_count
.rbegin(); p
!= state_by_count
.rend(); ++p
)
305 std::stringstream ss
;
307 max_width
= std::max
<size_t>(ss
.str().size(), max_width
);
310 for (auto p
= state_by_count
.rbegin(); p
!= state_by_count
.rend(); ++p
)
316 out
->setf(std::ios::left
);
317 *out
<< std::setw(max_width
) << p
->first
318 << " " << pg_state_string(p
->second
) << "\n";
319 out
->unsetf(std::ios::left
);
323 ostringstream ss_rec_io
;
324 overall_recovery_rate_summary(f
, &ss_rec_io
);
325 ostringstream ss_client_io
;
326 overall_client_io_rate_summary(f
, &ss_client_io
);
327 ostringstream ss_cache_io
;
328 overall_cache_io_rate_summary(f
, &ss_cache_io
);
330 if (!f
&& (ss_client_io
.str().length() || ss_rec_io
.str().length()
331 || ss_cache_io
.str().length())) {
336 if (!f
&& ss_client_io
.str().length())
337 *out
<< " client: " << ss_client_io
.str() << "\n";
338 if (!f
&& ss_rec_io
.str().length())
339 *out
<< " recovery: " << ss_rec_io
.str() << "\n";
340 if (!f
&& ss_cache_io
.str().length())
341 *out
<< " cache: " << ss_cache_io
.str() << "\n";
344 void PGMapDigest::print_oneline_summary(ceph::Formatter
*f
, ostream
*out
) const
346 std::stringstream ss
;
349 f
->open_array_section("num_pg_by_state");
350 for (auto p
= num_pg_by_state
.begin();
351 p
!= num_pg_by_state
.end();
354 f
->open_object_section("state");
355 f
->dump_string("name", pg_state_string(p
->first
));
356 f
->dump_unsigned("num", p
->second
);
359 if (p
!= num_pg_by_state
.begin())
361 ss
<< p
->second
<< " " << pg_state_string(p
->first
);
366 string states
= ss
.str();
368 *out
<< num_pg
<< " pgs: "
370 << byte_u_t(pg_sum
.stats
.sum
.num_bytes
) << " data, "
371 << byte_u_t(osd_sum
.statfs
.get_used()) << " used, "
372 << byte_u_t(osd_sum
.statfs
.available
) << " / "
373 << byte_u_t(osd_sum
.statfs
.total
) << " avail";
375 f
->dump_unsigned("num_pgs", num_pg
);
376 f
->dump_unsigned("num_bytes", pg_sum
.stats
.sum
.num_bytes
);
377 f
->dump_int("total_bytes", osd_sum
.statfs
.total
);
378 f
->dump_int("total_avail_bytes", osd_sum
.statfs
.available
);
379 f
->dump_int("total_used_bytes", osd_sum
.statfs
.get_used());
380 f
->dump_int("total_used_raw_bytes", osd_sum
.statfs
.get_used_raw());
383 // make non-negative; we can get negative values if osds send
384 // uncommitted stats and then "go backward" or if they are just
386 pool_stat_t pos_delta
= pg_sum_delta
;
388 if (pos_delta
.stats
.sum
.num_rd
||
389 pos_delta
.stats
.sum
.num_wr
) {
392 if (pos_delta
.stats
.sum
.num_rd
) {
393 int64_t rd
= (pos_delta
.stats
.sum
.num_rd_kb
<< 10) / (double)stamp_delta
;
395 *out
<< byte_u_t(rd
) << "/s rd, ";
397 f
->dump_unsigned("read_bytes_sec", rd
);
399 if (pos_delta
.stats
.sum
.num_wr
) {
400 int64_t wr
= (pos_delta
.stats
.sum
.num_wr_kb
<< 10) / (double)stamp_delta
;
402 *out
<< byte_u_t(wr
) << "/s wr, ";
404 f
->dump_unsigned("write_bytes_sec", wr
);
406 int64_t iops
= (pos_delta
.stats
.sum
.num_rd
+ pos_delta
.stats
.sum
.num_wr
) / (double)stamp_delta
;
408 *out
<< si_u_t(iops
) << " op/s";
410 f
->dump_unsigned("io_sec", iops
);
414 overall_recovery_summary(f
, &sl
);
416 for (auto p
= sl
.begin(); p
!= sl
.end(); ++p
)
418 std::stringstream ssr
;
419 overall_recovery_rate_summary(f
, &ssr
);
420 if (out
&& ssr
.str().length())
421 *out
<< "; " << ssr
.str() << " recovering";
424 void PGMapDigest::get_recovery_stats(
425 double *misplaced_ratio
,
426 double *degraded_ratio
,
427 double *inactive_pgs_ratio
,
428 double *unknown_pgs_ratio
) const
430 if (pg_sum
.stats
.sum
.num_objects_degraded
&&
431 pg_sum
.stats
.sum
.num_object_copies
> 0) {
432 *degraded_ratio
= (double)pg_sum
.stats
.sum
.num_objects_degraded
/
433 (double)pg_sum
.stats
.sum
.num_object_copies
;
437 if (pg_sum
.stats
.sum
.num_objects_misplaced
&&
438 pg_sum
.stats
.sum
.num_object_copies
> 0) {
439 *misplaced_ratio
= (double)pg_sum
.stats
.sum
.num_objects_misplaced
/
440 (double)pg_sum
.stats
.sum
.num_object_copies
;
442 *misplaced_ratio
= 0;
445 int num_pg_inactive
= num_pg
- num_pg_active
- num_pg_unknown
;
446 *inactive_pgs_ratio
= (double)num_pg_inactive
/ (double)num_pg
;
447 *unknown_pgs_ratio
= (double)num_pg_unknown
/ (double)num_pg
;
449 *inactive_pgs_ratio
= 0;
450 *unknown_pgs_ratio
= 0;
454 void PGMapDigest::recovery_summary(ceph::Formatter
*f
, list
<string
> *psl
,
455 const pool_stat_t
& pool_sum
) const
457 if (pool_sum
.stats
.sum
.num_objects_degraded
&& pool_sum
.stats
.sum
.num_object_copies
> 0) {
458 double pc
= (double)pool_sum
.stats
.sum
.num_objects_degraded
/
459 (double)pool_sum
.stats
.sum
.num_object_copies
* (double)100.0;
461 snprintf(b
, sizeof(b
), "%.3lf", pc
);
463 f
->dump_unsigned("degraded_objects", pool_sum
.stats
.sum
.num_objects_degraded
);
464 f
->dump_unsigned("degraded_total", pool_sum
.stats
.sum
.num_object_copies
);
465 f
->dump_float("degraded_ratio", pc
/ 100.0);
468 ss
<< pool_sum
.stats
.sum
.num_objects_degraded
469 << "/" << pool_sum
.stats
.sum
.num_object_copies
<< " objects degraded (" << b
<< "%)";
470 psl
->push_back(ss
.str());
473 if (pool_sum
.stats
.sum
.num_objects_misplaced
&& pool_sum
.stats
.sum
.num_object_copies
> 0) {
474 double pc
= (double)pool_sum
.stats
.sum
.num_objects_misplaced
/
475 (double)pool_sum
.stats
.sum
.num_object_copies
* (double)100.0;
477 snprintf(b
, sizeof(b
), "%.3lf", pc
);
479 f
->dump_unsigned("misplaced_objects", pool_sum
.stats
.sum
.num_objects_misplaced
);
480 f
->dump_unsigned("misplaced_total", pool_sum
.stats
.sum
.num_object_copies
);
481 f
->dump_float("misplaced_ratio", pc
/ 100.0);
484 ss
<< pool_sum
.stats
.sum
.num_objects_misplaced
485 << "/" << pool_sum
.stats
.sum
.num_object_copies
<< " objects misplaced (" << b
<< "%)";
486 psl
->push_back(ss
.str());
489 if (pool_sum
.stats
.sum
.num_objects_unfound
&& pool_sum
.stats
.sum
.num_objects
) {
490 double pc
= (double)pool_sum
.stats
.sum
.num_objects_unfound
/
491 (double)pool_sum
.stats
.sum
.num_objects
* (double)100.0;
493 snprintf(b
, sizeof(b
), "%.3lf", pc
);
495 f
->dump_unsigned("unfound_objects", pool_sum
.stats
.sum
.num_objects_unfound
);
496 f
->dump_unsigned("unfound_total", pool_sum
.stats
.sum
.num_objects
);
497 f
->dump_float("unfound_ratio", pc
/ 100.0);
500 ss
<< pool_sum
.stats
.sum
.num_objects_unfound
501 << "/" << pool_sum
.stats
.sum
.num_objects
<< " objects unfound (" << b
<< "%)";
502 psl
->push_back(ss
.str());
507 void PGMapDigest::recovery_rate_summary(ceph::Formatter
*f
, ostream
*out
,
508 const pool_stat_t
& delta_sum
,
509 utime_t delta_stamp
) const
511 // make non-negative; we can get negative values if osds send
512 // uncommitted stats and then "go backward" or if they are just
514 pool_stat_t pos_delta
= delta_sum
;
516 if (pos_delta
.stats
.sum
.num_objects_recovered
||
517 pos_delta
.stats
.sum
.num_bytes_recovered
||
518 pos_delta
.stats
.sum
.num_keys_recovered
) {
519 int64_t objps
= pos_delta
.stats
.sum
.num_objects_recovered
/ (double)delta_stamp
;
520 int64_t bps
= pos_delta
.stats
.sum
.num_bytes_recovered
/ (double)delta_stamp
;
521 int64_t kps
= pos_delta
.stats
.sum
.num_keys_recovered
/ (double)delta_stamp
;
523 f
->dump_int("recovering_objects_per_sec", objps
);
524 f
->dump_int("recovering_bytes_per_sec", bps
);
525 f
->dump_int("recovering_keys_per_sec", kps
);
526 f
->dump_int("num_objects_recovered", pos_delta
.stats
.sum
.num_objects_recovered
);
527 f
->dump_int("num_bytes_recovered", pos_delta
.stats
.sum
.num_bytes_recovered
);
528 f
->dump_int("num_keys_recovered", pos_delta
.stats
.sum
.num_keys_recovered
);
530 *out
<< byte_u_t(bps
) << "/s";
531 if (pos_delta
.stats
.sum
.num_keys_recovered
)
532 *out
<< ", " << si_u_t(kps
) << " keys/s";
533 *out
<< ", " << si_u_t(objps
) << " objects/s";
538 void PGMapDigest::overall_recovery_rate_summary(ceph::Formatter
*f
, ostream
*out
) const
540 recovery_rate_summary(f
, out
, pg_sum_delta
, stamp_delta
);
543 void PGMapDigest::overall_recovery_summary(ceph::Formatter
*f
, list
<string
> *psl
) const
545 recovery_summary(f
, psl
, pg_sum
);
548 void PGMapDigest::pool_recovery_rate_summary(ceph::Formatter
*f
, ostream
*out
,
549 uint64_t poolid
) const
551 auto p
= per_pool_sum_delta
.find(poolid
);
552 if (p
== per_pool_sum_delta
.end())
555 auto ts
= per_pool_sum_deltas_stamps
.find(p
->first
);
556 ceph_assert(ts
!= per_pool_sum_deltas_stamps
.end());
557 recovery_rate_summary(f
, out
, p
->second
.first
, ts
->second
);
560 void PGMapDigest::pool_recovery_summary(ceph::Formatter
*f
, list
<string
> *psl
,
561 uint64_t poolid
) const
563 auto p
= pg_pool_sum
.find(poolid
);
564 if (p
== pg_pool_sum
.end())
567 recovery_summary(f
, psl
, p
->second
);
570 void PGMapDigest::client_io_rate_summary(ceph::Formatter
*f
, ostream
*out
,
571 const pool_stat_t
& delta_sum
,
572 utime_t delta_stamp
) const
574 pool_stat_t pos_delta
= delta_sum
;
576 if (pos_delta
.stats
.sum
.num_rd
||
577 pos_delta
.stats
.sum
.num_wr
) {
578 if (pos_delta
.stats
.sum
.num_rd
) {
579 int64_t rd
= (pos_delta
.stats
.sum
.num_rd_kb
<< 10) / (double)delta_stamp
;
581 f
->dump_int("read_bytes_sec", rd
);
583 *out
<< byte_u_t(rd
) << "/s rd, ";
586 if (pos_delta
.stats
.sum
.num_wr
) {
587 int64_t wr
= (pos_delta
.stats
.sum
.num_wr_kb
<< 10) / (double)delta_stamp
;
589 f
->dump_int("write_bytes_sec", wr
);
591 *out
<< byte_u_t(wr
) << "/s wr, ";
594 int64_t iops_rd
= pos_delta
.stats
.sum
.num_rd
/ (double)delta_stamp
;
595 int64_t iops_wr
= pos_delta
.stats
.sum
.num_wr
/ (double)delta_stamp
;
597 f
->dump_int("read_op_per_sec", iops_rd
);
598 f
->dump_int("write_op_per_sec", iops_wr
);
600 *out
<< si_u_t(iops_rd
) << " op/s rd, " << si_u_t(iops_wr
) << " op/s wr";
605 void PGMapDigest::overall_client_io_rate_summary(ceph::Formatter
*f
, ostream
*out
) const
607 client_io_rate_summary(f
, out
, pg_sum_delta
, stamp_delta
);
610 void PGMapDigest::pool_client_io_rate_summary(ceph::Formatter
*f
, ostream
*out
,
611 uint64_t poolid
) const
613 auto p
= per_pool_sum_delta
.find(poolid
);
614 if (p
== per_pool_sum_delta
.end())
617 auto ts
= per_pool_sum_deltas_stamps
.find(p
->first
);
618 ceph_assert(ts
!= per_pool_sum_deltas_stamps
.end());
619 client_io_rate_summary(f
, out
, p
->second
.first
, ts
->second
);
622 void PGMapDigest::cache_io_rate_summary(ceph::Formatter
*f
, ostream
*out
,
623 const pool_stat_t
& delta_sum
,
624 utime_t delta_stamp
) const
626 pool_stat_t pos_delta
= delta_sum
;
628 bool have_output
= false;
630 if (pos_delta
.stats
.sum
.num_flush
) {
631 int64_t flush
= (pos_delta
.stats
.sum
.num_flush_kb
<< 10) / (double)delta_stamp
;
633 f
->dump_int("flush_bytes_sec", flush
);
635 *out
<< byte_u_t(flush
) << "/s flush";
639 if (pos_delta
.stats
.sum
.num_evict
) {
640 int64_t evict
= (pos_delta
.stats
.sum
.num_evict_kb
<< 10) / (double)delta_stamp
;
642 f
->dump_int("evict_bytes_sec", evict
);
646 *out
<< byte_u_t(evict
) << "/s evict";
650 if (pos_delta
.stats
.sum
.num_promote
) {
651 int64_t promote
= pos_delta
.stats
.sum
.num_promote
/ (double)delta_stamp
;
653 f
->dump_int("promote_op_per_sec", promote
);
657 *out
<< si_u_t(promote
) << " op/s promote";
661 if (pos_delta
.stats
.sum
.num_flush_mode_low
) {
663 f
->dump_int("num_flush_mode_low", pos_delta
.stats
.sum
.num_flush_mode_low
);
667 *out
<< si_u_t(pos_delta
.stats
.sum
.num_flush_mode_low
) << " PGs flushing";
671 if (pos_delta
.stats
.sum
.num_flush_mode_high
) {
673 f
->dump_int("num_flush_mode_high", pos_delta
.stats
.sum
.num_flush_mode_high
);
677 *out
<< si_u_t(pos_delta
.stats
.sum
.num_flush_mode_high
) << " PGs flushing (high)";
681 if (pos_delta
.stats
.sum
.num_evict_mode_some
) {
683 f
->dump_int("num_evict_mode_some", pos_delta
.stats
.sum
.num_evict_mode_some
);
687 *out
<< si_u_t(pos_delta
.stats
.sum
.num_evict_mode_some
) << " PGs evicting";
691 if (pos_delta
.stats
.sum
.num_evict_mode_full
) {
693 f
->dump_int("num_evict_mode_full", pos_delta
.stats
.sum
.num_evict_mode_full
);
697 *out
<< si_u_t(pos_delta
.stats
.sum
.num_evict_mode_full
) << " PGs evicting (full)";
702 void PGMapDigest::overall_cache_io_rate_summary(ceph::Formatter
*f
, ostream
*out
) const
704 cache_io_rate_summary(f
, out
, pg_sum_delta
, stamp_delta
);
707 void PGMapDigest::pool_cache_io_rate_summary(ceph::Formatter
*f
, ostream
*out
,
708 uint64_t poolid
) const
710 auto p
= per_pool_sum_delta
.find(poolid
);
711 if (p
== per_pool_sum_delta
.end())
714 auto ts
= per_pool_sum_deltas_stamps
.find(p
->first
);
715 ceph_assert(ts
!= per_pool_sum_deltas_stamps
.end());
716 cache_io_rate_summary(f
, out
, p
->second
.first
, ts
->second
);
719 ceph_statfs
PGMapDigest::get_statfs(OSDMap
&osdmap
,
720 boost::optional
<int64_t> data_pool
) const
724 object_stat_sum_t sum
;
727 auto i
= pg_pool_sum
.find(*data_pool
);
728 if (i
!= pg_pool_sum
.end()) {
729 sum
= i
->second
.stats
.sum
;
735 statfs
.kb_used
= (sum
.num_bytes
>> 10);
736 statfs
.kb_avail
= get_pool_free_space(osdmap
, *data_pool
) >> 10;
737 statfs
.num_objects
= sum
.num_objects
;
738 statfs
.kb
= statfs
.kb_used
+ statfs
.kb_avail
;
741 statfs
.kb
= osd_sum
.statfs
.kb();
742 statfs
.kb_used
= osd_sum
.statfs
.kb_used_raw();
743 statfs
.kb_avail
= osd_sum
.statfs
.kb_avail();
744 statfs
.num_objects
= pg_sum
.stats
.sum
.num_objects
;
750 void PGMapDigest::dump_pool_stats_full(
751 const OSDMap
&osd_map
,
759 f
->open_array_section("pools");
761 tbl
.define_column("POOL", TextTable::LEFT
, TextTable::LEFT
);
762 tbl
.define_column("ID", TextTable::LEFT
, TextTable::RIGHT
);
763 tbl
.define_column("STORED", TextTable::LEFT
, TextTable::RIGHT
);
765 tbl
.define_column("(DATA)", TextTable::LEFT
, TextTable::RIGHT
);
766 tbl
.define_column("(OMAP)", TextTable::LEFT
, TextTable::RIGHT
);
768 tbl
.define_column("OBJECTS", TextTable::LEFT
, TextTable::RIGHT
);
769 tbl
.define_column("USED", TextTable::LEFT
, TextTable::RIGHT
);
771 tbl
.define_column("(DATA)", TextTable::LEFT
, TextTable::RIGHT
);
772 tbl
.define_column("(OMAP)", TextTable::LEFT
, TextTable::RIGHT
);
774 tbl
.define_column("%USED", TextTable::LEFT
, TextTable::RIGHT
);
775 tbl
.define_column("MAX AVAIL", TextTable::LEFT
, TextTable::RIGHT
);
778 tbl
.define_column("QUOTA OBJECTS", TextTable::LEFT
, TextTable::LEFT
);
779 tbl
.define_column("QUOTA BYTES", TextTable::LEFT
, TextTable::LEFT
);
780 tbl
.define_column("DIRTY", TextTable::LEFT
, TextTable::RIGHT
);
781 tbl
.define_column("USED COMPR", TextTable::LEFT
, TextTable::RIGHT
);
782 tbl
.define_column("UNDER COMPR", TextTable::LEFT
, TextTable::RIGHT
);
786 map
<int,uint64_t> avail_by_rule
;
787 for (auto p
= osd_map
.get_pools().begin();
788 p
!= osd_map
.get_pools().end(); ++p
) {
789 int64_t pool_id
= p
->first
;
790 if ((pool_id
< 0) || (pg_pool_sum
.count(pool_id
) == 0))
793 const string
& pool_name
= osd_map
.get_pool_name(pool_id
);
794 const pool_stat_t
&stat
= pg_pool_sum
.at(pool_id
);
796 const pg_pool_t
*pool
= osd_map
.get_pg_pool(pool_id
);
797 int ruleno
= osd_map
.crush
->find_rule(pool
->get_crush_rule(),
801 if (avail_by_rule
.count(ruleno
) == 0) {
802 // FIXME: we don't guarantee avail_space_by_rule is up-to-date before this function is invoked
803 avail
= get_rule_avail(ruleno
);
806 avail_by_rule
[ruleno
] = avail
;
808 avail
= avail_by_rule
[ruleno
];
811 f
->open_object_section("pool");
812 f
->dump_string("name", pool_name
);
813 f
->dump_int("id", pool_id
);
814 f
->open_object_section("stats");
819 float raw_used_rate
= osd_map
.pool_raw_used_rate(pool_id
);
820 bool per_pool
= use_per_pool_stats();
821 bool per_pool_omap
= use_per_pool_omap_stats();
822 dump_object_stat_sum(tbl
, f
, stat
, avail
, raw_used_rate
, verbose
, per_pool
,
823 per_pool_omap
, pool
);
825 f
->close_section(); // stats
826 f
->close_section(); // pool
828 tbl
<< TextTable::endrow
;
834 ceph_assert(ss
!= nullptr);
835 *ss
<< "--- POOLS ---\n";
840 void PGMapDigest::dump_cluster_stats(stringstream
*ss
,
845 f
->open_object_section("stats");
846 f
->dump_int("total_bytes", osd_sum
.statfs
.total
);
847 f
->dump_int("total_avail_bytes", osd_sum
.statfs
.available
);
848 f
->dump_int("total_used_bytes", osd_sum
.statfs
.get_used());
849 f
->dump_int("total_used_raw_bytes", osd_sum
.statfs
.get_used_raw());
850 f
->dump_float("total_used_raw_ratio", osd_sum
.statfs
.get_used_raw_ratio());
851 f
->dump_unsigned("num_osds", osd_sum
.num_osds
);
852 f
->dump_unsigned("num_per_pool_osds", osd_sum
.num_per_pool_osds
);
853 f
->dump_unsigned("num_per_pool_omap_osds", osd_sum
.num_per_pool_omap_osds
);
855 f
->open_object_section("stats_by_class");
856 for (auto& i
: osd_sum_by_class
) {
857 f
->open_object_section(i
.first
.c_str());
858 f
->dump_int("total_bytes", i
.second
.statfs
.total
);
859 f
->dump_int("total_avail_bytes", i
.second
.statfs
.available
);
860 f
->dump_int("total_used_bytes", i
.second
.statfs
.get_used());
861 f
->dump_int("total_used_raw_bytes", i
.second
.statfs
.get_used_raw());
862 f
->dump_float("total_used_raw_ratio",
863 i
.second
.statfs
.get_used_raw_ratio());
868 ceph_assert(ss
!= nullptr);
870 tbl
.define_column("CLASS", TextTable::LEFT
, TextTable::LEFT
);
871 tbl
.define_column("SIZE", TextTable::LEFT
, TextTable::RIGHT
);
872 tbl
.define_column("AVAIL", TextTable::LEFT
, TextTable::RIGHT
);
873 tbl
.define_column("USED", TextTable::LEFT
, TextTable::RIGHT
);
874 tbl
.define_column("RAW USED", TextTable::LEFT
, TextTable::RIGHT
);
875 tbl
.define_column("%RAW USED", TextTable::LEFT
, TextTable::RIGHT
);
878 for (auto& i
: osd_sum_by_class
) {
880 tbl
<< stringify(byte_u_t(i
.second
.statfs
.total
))
881 << stringify(byte_u_t(i
.second
.statfs
.available
))
882 << stringify(byte_u_t(i
.second
.statfs
.get_used()))
883 << stringify(byte_u_t(i
.second
.statfs
.get_used_raw()))
884 << percentify(i
.second
.statfs
.get_used_raw_ratio()*100.0)
885 << TextTable::endrow
;
888 tbl
<< stringify(byte_u_t(osd_sum
.statfs
.total
))
889 << stringify(byte_u_t(osd_sum
.statfs
.available
))
890 << stringify(byte_u_t(osd_sum
.statfs
.get_used()))
891 << stringify(byte_u_t(osd_sum
.statfs
.get_used_raw()))
892 << percentify(osd_sum
.statfs
.get_used_raw_ratio()*100.0)
893 << TextTable::endrow
;
895 *ss
<< "--- RAW STORAGE ---\n";
900 void PGMapDigest::dump_object_stat_sum(
901 TextTable
&tbl
, ceph::Formatter
*f
,
902 const pool_stat_t
&pool_stat
, uint64_t avail
,
903 float raw_used_rate
, bool verbose
, bool per_pool
, bool per_pool_omap
,
904 const pg_pool_t
*pool
)
906 const object_stat_sum_t
&sum
= pool_stat
.stats
.sum
;
907 const store_statfs_t statfs
= pool_stat
.store_stats
;
909 if (sum
.num_object_copies
> 0) {
910 raw_used_rate
*= (float)(sum
.num_object_copies
- sum
.num_objects_degraded
) / sum
.num_object_copies
;
913 uint64_t used_data_bytes
= pool_stat
.get_allocated_data_bytes(per_pool
);
914 uint64_t used_omap_bytes
= pool_stat
.get_allocated_omap_bytes(per_pool_omap
);
915 uint64_t used_bytes
= used_data_bytes
+ used_omap_bytes
;
918 // note avail passed in is raw_avail, calc raw_used here.
921 used
/= used
+ avail
;
922 } else if (used_bytes
) {
925 auto avail_res
= raw_used_rate
? avail
/ raw_used_rate
: 0;
926 // an approximation for actually stored user data
927 auto stored_data_normalized
= pool_stat
.get_user_data_bytes(
928 raw_used_rate
, per_pool
);
929 auto stored_omap_normalized
= pool_stat
.get_user_omap_bytes(
930 raw_used_rate
, per_pool_omap
);
931 auto stored_normalized
= stored_data_normalized
+ stored_omap_normalized
;
932 // same, amplied by replication or EC
933 auto stored_raw
= stored_normalized
* raw_used_rate
;
935 f
->dump_int("stored", stored_normalized
);
937 f
->dump_int("stored_data", stored_data_normalized
);
938 f
->dump_int("stored_omap", stored_omap_normalized
);
940 f
->dump_int("objects", sum
.num_objects
);
941 f
->dump_int("kb_used", shift_round_up(used_bytes
, 10));
942 f
->dump_int("bytes_used", used_bytes
);
944 f
->dump_int("data_bytes_used", used_data_bytes
);
945 f
->dump_int("omap_bytes_used", used_omap_bytes
);
947 f
->dump_float("percent_used", used
);
948 f
->dump_unsigned("max_avail", avail_res
);
950 f
->dump_int("quota_objects", pool
->quota_max_objects
);
951 f
->dump_int("quota_bytes", pool
->quota_max_bytes
);
952 f
->dump_int("dirty", sum
.num_objects_dirty
);
953 f
->dump_int("rd", sum
.num_rd
);
954 f
->dump_int("rd_bytes", sum
.num_rd_kb
* 1024ull);
955 f
->dump_int("wr", sum
.num_wr
);
956 f
->dump_int("wr_bytes", sum
.num_wr_kb
* 1024ull);
957 f
->dump_int("compress_bytes_used", statfs
.data_compressed_allocated
);
958 f
->dump_int("compress_under_bytes", statfs
.data_compressed_original
);
959 // Stored by user amplified by replication
960 f
->dump_int("stored_raw", stored_raw
);
963 tbl
<< stringify(byte_u_t(stored_normalized
));
965 tbl
<< stringify(byte_u_t(stored_data_normalized
));
966 tbl
<< stringify(byte_u_t(stored_omap_normalized
));
968 tbl
<< stringify(si_u_t(sum
.num_objects
));
969 tbl
<< stringify(byte_u_t(used_bytes
));
971 tbl
<< stringify(byte_u_t(used_data_bytes
));
972 tbl
<< stringify(byte_u_t(used_omap_bytes
));
974 tbl
<< percentify(used
*100);
975 tbl
<< stringify(byte_u_t(avail_res
));
977 if (pool
->quota_max_objects
== 0)
980 tbl
<< stringify(si_u_t(pool
->quota_max_objects
));
982 if (pool
->quota_max_bytes
== 0)
985 tbl
<< stringify(byte_u_t(pool
->quota_max_bytes
));
987 tbl
<< stringify(si_u_t(sum
.num_objects_dirty
))
988 << stringify(byte_u_t(statfs
.data_compressed_allocated
))
989 << stringify(byte_u_t(statfs
.data_compressed_original
))
995 int64_t PGMapDigest::get_pool_free_space(const OSDMap
&osd_map
,
996 int64_t poolid
) const
998 const pg_pool_t
*pool
= osd_map
.get_pg_pool(poolid
);
999 int ruleno
= osd_map
.crush
->find_rule(pool
->get_crush_rule(),
1003 avail
= get_rule_avail(ruleno
);
1007 return avail
/ osd_map
.pool_raw_used_rate(poolid
);
1010 int64_t PGMap::get_rule_avail(const OSDMap
& osdmap
, int ruleno
) const
1013 int r
= osdmap
.crush
->get_rule_weight_osd_map(ruleno
, &wm
);
1021 float fratio
= osdmap
.get_full_ratio();
1024 for (auto p
= wm
.begin(); p
!= wm
.end(); ++p
) {
1025 auto osd_info
= osd_stat
.find(p
->first
);
1026 if (osd_info
!= osd_stat
.end()) {
1027 if (osd_info
->second
.statfs
.total
== 0 || p
->second
== 0) {
1028 // osd must be out, hence its stats have been zeroed
1029 // (unless we somehow managed to have a disk with size 0...)
1031 // (p->second == 0), if osd weight is 0, no need to
1032 // calculate proj below.
1035 double unusable
= (double)osd_info
->second
.statfs
.kb() *
1037 double avail
= std::max(0.0, (double)osd_info
->second
.statfs
.kb_avail() - unusable
);
1039 int64_t proj
= (int64_t)(avail
/ (double)p
->second
);
1040 if (min
< 0 || proj
< min
) {
1044 if (osdmap
.is_up(p
->first
)) {
1045 // This is a level 4 rather than an error, because we might have
1046 // only just started, and not received the first stats message yet.
1047 dout(4) << "OSD " << p
->first
<< " is up, but has no stats" << dendl
;
1054 void PGMap::get_rules_avail(const OSDMap
& osdmap
,
1055 std::map
<int,int64_t> *avail_map
) const
1058 for (auto p
: osdmap
.get_pools()) {
1059 int64_t pool_id
= p
.first
;
1060 if ((pool_id
< 0) || (pg_pool_sum
.count(pool_id
) == 0))
1062 const pg_pool_t
*pool
= osdmap
.get_pg_pool(pool_id
);
1063 int ruleno
= osdmap
.crush
->find_rule(pool
->get_crush_rule(),
1066 if (avail_map
->count(ruleno
) == 0)
1067 (*avail_map
)[ruleno
] = get_rule_avail(osdmap
, ruleno
);
1071 // ---------------------
1074 void PGMap::Incremental::dump(ceph::Formatter
*f
) const
1076 f
->dump_unsigned("version", version
);
1077 f
->dump_stream("stamp") << stamp
;
1078 f
->dump_unsigned("osdmap_epoch", osdmap_epoch
);
1079 f
->dump_unsigned("pg_scan_epoch", pg_scan
);
1081 f
->open_array_section("pg_stat_updates");
1082 for (auto p
= pg_stat_updates
.begin(); p
!= pg_stat_updates
.end(); ++p
) {
1083 f
->open_object_section("pg_stat");
1084 f
->dump_stream("pgid") << p
->first
;
1090 f
->open_array_section("osd_stat_updates");
1091 for (auto p
= osd_stat_updates
.begin(); p
!= osd_stat_updates
.end(); ++p
) {
1092 f
->open_object_section("osd_stat");
1093 f
->dump_int("osd", p
->first
);
1098 f
->open_array_section("pool_statfs_updates");
1099 for (auto p
= pool_statfs_updates
.begin(); p
!= pool_statfs_updates
.end(); ++p
) {
1100 f
->open_object_section("pool_statfs");
1101 f
->dump_stream("poolid/osd") << p
->first
;
1107 f
->open_array_section("osd_stat_removals");
1108 for (auto p
= osd_stat_rm
.begin(); p
!= osd_stat_rm
.end(); ++p
)
1109 f
->dump_int("osd", *p
);
1112 f
->open_array_section("pg_removals");
1113 for (auto p
= pg_remove
.begin(); p
!= pg_remove
.end(); ++p
)
1114 f
->dump_stream("pgid") << *p
;
1118 void PGMap::Incremental::generate_test_instances(list
<PGMap::Incremental
*>& o
)
1120 o
.push_back(new Incremental
);
1121 o
.push_back(new Incremental
);
1122 o
.back()->version
= 1;
1123 o
.back()->stamp
= utime_t(123,345);
1124 o
.push_back(new Incremental
);
1125 o
.back()->version
= 2;
1126 o
.back()->pg_stat_updates
[pg_t(1,2)] = pg_stat_t();
1127 o
.back()->osd_stat_updates
[5] = osd_stat_t();
1128 o
.push_back(new Incremental
);
1129 o
.back()->version
= 3;
1130 o
.back()->osdmap_epoch
= 1;
1131 o
.back()->pg_scan
= 2;
1132 o
.back()->pg_stat_updates
[pg_t(4,5)] = pg_stat_t();
1133 o
.back()->osd_stat_updates
[6] = osd_stat_t();
1134 o
.back()->pg_remove
.insert(pg_t(1,2));
1135 o
.back()->osd_stat_rm
.insert(5);
1136 o
.back()->pool_statfs_updates
[std::make_pair(1234,4)] = store_statfs_t();
1141 void PGMap::apply_incremental(CephContext
*cct
, const Incremental
& inc
)
1143 ceph_assert(inc
.version
== version
+1);
1146 pool_stat_t pg_sum_old
= pg_sum
;
1147 mempool::pgmap::unordered_map
<int32_t, pool_stat_t
> pg_pool_sum_old
;
1148 pg_pool_sum_old
= pg_pool_sum
;
1150 for (auto p
= inc
.pg_stat_updates
.begin();
1151 p
!= inc
.pg_stat_updates
.end();
1153 const pg_t
&update_pg(p
->first
);
1154 auto update_pool
= update_pg
.pool();
1155 const pg_stat_t
&update_stat(p
->second
);
1157 auto pg_stat_iter
= pg_stat
.find(update_pg
);
1158 pool_stat_t
&pool_sum_ref
= pg_pool_sum
[update_pool
];
1159 if (pg_stat_iter
== pg_stat
.end()) {
1160 pg_stat
.insert(make_pair(update_pg
, update_stat
));
1162 stat_pg_sub(update_pg
, pg_stat_iter
->second
);
1163 pool_sum_ref
.sub(pg_stat_iter
->second
);
1164 pg_stat_iter
->second
= update_stat
;
1166 stat_pg_add(update_pg
, update_stat
);
1167 pool_sum_ref
.add(update_stat
);
1170 for (auto p
= inc
.pool_statfs_updates
.begin();
1171 p
!= inc
.pool_statfs_updates
.end();
1173 auto update_pool
= p
->first
.first
;
1174 auto update_osd
= p
->first
.second
;
1175 auto& statfs_inc
= p
->second
;
1177 auto pool_statfs_iter
=
1178 pool_statfs
.find(std::make_pair(update_pool
, update_osd
));
1179 if (pg_pool_sum
.count(update_pool
)) {
1180 pool_stat_t
&pool_sum_ref
= pg_pool_sum
[update_pool
];
1181 if (pool_statfs_iter
== pool_statfs
.end()) {
1182 pool_statfs
.emplace(std::make_pair(update_pool
, update_osd
), statfs_inc
);
1184 pool_sum_ref
.sub(pool_statfs_iter
->second
);
1185 pool_statfs_iter
->second
= statfs_inc
;
1187 pool_sum_ref
.add(statfs_inc
);
1191 for (auto p
= inc
.get_osd_stat_updates().begin();
1192 p
!= inc
.get_osd_stat_updates().end();
1195 const osd_stat_t
&new_stats(p
->second
);
1197 auto t
= osd_stat
.find(osd
);
1198 if (t
== osd_stat
.end()) {
1199 osd_stat
.insert(make_pair(osd
, new_stats
));
1201 stat_osd_sub(t
->first
, t
->second
);
1202 t
->second
= new_stats
;
1204 stat_osd_add(osd
, new_stats
);
1206 set
<int64_t> deleted_pools
;
1207 for (auto p
= inc
.pg_remove
.begin();
1208 p
!= inc
.pg_remove
.end();
1210 const pg_t
&removed_pg(*p
);
1211 auto s
= pg_stat
.find(removed_pg
);
1212 bool pool_erased
= false;
1213 if (s
!= pg_stat
.end()) {
1214 pool_erased
= stat_pg_sub(removed_pg
, s
->second
);
1217 deleted_pools
.insert(removed_pg
.pool());
1222 for (auto p
= inc
.get_osd_stat_rm().begin();
1223 p
!= inc
.get_osd_stat_rm().end();
1225 auto t
= osd_stat
.find(*p
);
1226 if (t
!= osd_stat
.end()) {
1227 stat_osd_sub(t
->first
, t
->second
);
1230 for (auto i
= pool_statfs
.begin(); i
!= pool_statfs
.end(); ++i
) {
1231 if (i
->first
.second
== *p
) {
1232 pg_pool_sum
[i
->first
.first
].sub(i
->second
);
1233 pool_statfs
.erase(i
);
1238 // skip calculating delta while sum was not synchronized
1239 if (!stamp
.is_zero() && !pg_sum_old
.stats
.sum
.is_zero()) {
1241 delta_t
= inc
.stamp
;
1243 // calculate a delta, and average over the last 2 deltas.
1244 pool_stat_t d
= pg_sum
;
1245 d
.stats
.sub(pg_sum_old
.stats
);
1246 pg_sum_deltas
.push_back(make_pair(d
, delta_t
));
1247 stamp_delta
+= delta_t
;
1248 pg_sum_delta
.stats
.add(d
.stats
);
1249 auto smooth_intervals
=
1250 cct
? cct
->_conf
.get_val
<uint64_t>("mon_stat_smooth_intervals") : 1;
1251 while (pg_sum_deltas
.size() > smooth_intervals
) {
1252 pg_sum_delta
.stats
.sub(pg_sum_deltas
.front().first
.stats
);
1253 stamp_delta
-= pg_sum_deltas
.front().second
;
1254 pg_sum_deltas
.pop_front();
1259 update_pool_deltas(cct
, inc
.stamp
, pg_pool_sum_old
);
1261 for (auto p
: deleted_pools
) {
1263 dout(20) << " deleted pool " << p
<< dendl
;
1267 if (inc
.osdmap_epoch
)
1268 last_osdmap_epoch
= inc
.osdmap_epoch
;
1270 last_pg_scan
= inc
.pg_scan
;
1273 void PGMap::calc_stats()
1279 pg_pool_sum
.clear();
1280 num_pg_by_pool
.clear();
1282 pg_sum
= pool_stat_t();
1283 osd_sum
= osd_stat_t();
1284 osd_sum_by_class
.clear();
1285 num_pg_by_state
.clear();
1286 num_pg_by_pool_state
.clear();
1287 num_pg_by_osd
.clear();
1289 for (auto p
= pg_stat
.begin();
1293 stat_pg_add(pg
, p
->second
);
1294 pg_pool_sum
[pg
.pool()].add(p
->second
);
1296 for (auto p
= pool_statfs
.begin();
1297 p
!= pool_statfs
.end();
1299 auto pool
= p
->first
.first
;
1300 pg_pool_sum
[pool
].add(p
->second
);
1302 for (auto p
= osd_stat
.begin();
1303 p
!= osd_stat
.end();
1305 stat_osd_add(p
->first
, p
->second
);
1308 void PGMap::stat_pg_add(const pg_t
&pgid
, const pg_stat_t
&s
,
1311 auto pool
= pgid
.pool();
1315 num_pg_by_state
[s
.state
]++;
1316 num_pg_by_pool_state
[pgid
.pool()][s
.state
]++;
1317 num_pg_by_pool
[pool
]++;
1319 if ((s
.state
& PG_STATE_CREATING
) &&
1320 s
.parent_split_bits
== 0) {
1321 creating_pgs
.insert(pgid
);
1322 if (s
.acting_primary
>= 0) {
1323 creating_pgs_by_osd_epoch
[s
.acting_primary
][s
.mapping_epoch
].insert(pgid
);
1327 if (s
.state
& PG_STATE_ACTIVE
) {
1337 for (auto p
= s
.blocked_by
.begin();
1338 p
!= s
.blocked_by
.end();
1340 ++blocked_by_sum
[*p
];
1343 for (auto p
= s
.acting
.begin(); p
!= s
.acting
.end(); ++p
) {
1344 pg_by_osd
[*p
].insert(pgid
);
1345 num_pg_by_osd
[*p
].acting
++;
1347 for (auto p
= s
.up
.begin(); p
!= s
.up
.end(); ++p
) {
1348 auto& t
= pg_by_osd
[*p
];
1349 if (t
.find(pgid
) == t
.end()) {
1351 num_pg_by_osd
[*p
].up_not_acting
++;
1355 if (s
.up_primary
>= 0) {
1356 num_pg_by_osd
[s
.up_primary
].primary
++;
1360 bool PGMap::stat_pg_sub(const pg_t
&pgid
, const pg_stat_t
&s
,
1363 bool pool_erased
= false;
1367 int end
= --num_pg_by_state
[s
.state
];
1368 ceph_assert(end
>= 0);
1370 num_pg_by_state
.erase(s
.state
);
1371 if (--num_pg_by_pool_state
[pgid
.pool()][s
.state
] == 0) {
1372 num_pg_by_pool_state
[pgid
.pool()].erase(s
.state
);
1374 end
= --num_pg_by_pool
[pgid
.pool()];
1379 if ((s
.state
& PG_STATE_CREATING
) &&
1380 s
.parent_split_bits
== 0) {
1381 creating_pgs
.erase(pgid
);
1382 if (s
.acting_primary
>= 0) {
1383 map
<epoch_t
,set
<pg_t
> >& r
= creating_pgs_by_osd_epoch
[s
.acting_primary
];
1384 r
[s
.mapping_epoch
].erase(pgid
);
1385 if (r
[s
.mapping_epoch
].empty())
1386 r
.erase(s
.mapping_epoch
);
1388 creating_pgs_by_osd_epoch
.erase(s
.acting_primary
);
1392 if (s
.state
& PG_STATE_ACTIVE
) {
1402 for (auto p
= s
.blocked_by
.begin();
1403 p
!= s
.blocked_by
.end();
1405 auto q
= blocked_by_sum
.find(*p
);
1406 ceph_assert(q
!= blocked_by_sum
.end());
1409 blocked_by_sum
.erase(q
);
1412 set
<int32_t> actingset
;
1413 for (auto p
= s
.acting
.begin(); p
!= s
.acting
.end(); ++p
) {
1414 actingset
.insert(*p
);
1415 auto& oset
= pg_by_osd
[*p
];
1418 pg_by_osd
.erase(*p
);
1419 auto it
= num_pg_by_osd
.find(*p
);
1420 if (it
!= num_pg_by_osd
.end() && it
->second
.acting
> 0)
1421 it
->second
.acting
--;
1423 for (auto p
= s
.up
.begin(); p
!= s
.up
.end(); ++p
) {
1424 auto& oset
= pg_by_osd
[*p
];
1427 pg_by_osd
.erase(*p
);
1428 if (actingset
.count(*p
))
1430 auto it
= num_pg_by_osd
.find(*p
);
1431 if (it
!= num_pg_by_osd
.end() && it
->second
.up_not_acting
> 0)
1432 it
->second
.up_not_acting
--;
1435 if (s
.up_primary
>= 0) {
1436 auto it
= num_pg_by_osd
.find(s
.up_primary
);
1437 if (it
!= num_pg_by_osd
.end() && it
->second
.primary
> 0)
1438 it
->second
.primary
--;
1443 void PGMap::calc_purged_snaps()
1445 purged_snaps
.clear();
1446 set
<int64_t> unknown
;
1447 for (auto& i
: pg_stat
) {
1448 if (i
.second
.state
== 0) {
1449 unknown
.insert(i
.first
.pool());
1450 purged_snaps
.erase(i
.first
.pool());
1452 } else if (unknown
.count(i
.first
.pool())) {
1455 auto j
= purged_snaps
.find(i
.first
.pool());
1456 if (j
== purged_snaps
.end()) {
1458 purged_snaps
[i
.first
.pool()] = i
.second
.purged_snaps
;
1460 j
->second
.intersection_of(i
.second
.purged_snaps
);
1465 void PGMap::calc_osd_sum_by_class(const OSDMap
& osdmap
)
1467 osd_sum_by_class
.clear();
1468 for (auto& i
: osd_stat
) {
1469 const char *class_name
= osdmap
.crush
->get_item_class(i
.first
);
1471 osd_sum_by_class
[class_name
].add(i
.second
);
1476 void PGMap::stat_osd_add(int osd
, const osd_stat_t
&s
)
1480 if (osd
>= (int)osd_last_seq
.size()) {
1481 osd_last_seq
.resize(osd
+ 1);
1483 osd_last_seq
[osd
] = s
.seq
;
1486 void PGMap::stat_osd_sub(int osd
, const osd_stat_t
&s
)
1490 ceph_assert(osd
< (int)osd_last_seq
.size());
1491 osd_last_seq
[osd
] = 0;
1494 void PGMap::encode_digest(const OSDMap
& osdmap
,
1495 bufferlist
& bl
, uint64_t features
)
1497 get_rules_avail(osdmap
, &avail_space_by_rule
);
1498 calc_osd_sum_by_class(osdmap
);
1499 calc_purged_snaps();
1500 PGMapDigest::encode(bl
, features
);
1503 void PGMap::encode(bufferlist
&bl
, uint64_t features
) const
1505 ENCODE_START(8, 8, bl
);
1506 encode(version
, bl
);
1507 encode(pg_stat
, bl
);
1508 encode(osd_stat
, bl
, features
);
1509 encode(last_osdmap_epoch
, bl
);
1510 encode(last_pg_scan
, bl
);
1512 encode(pool_statfs
, bl
, features
);
1516 void PGMap::decode(bufferlist::const_iterator
&bl
)
1518 DECODE_START(8, bl
);
1519 decode(version
, bl
);
1520 decode(pg_stat
, bl
);
1521 decode(osd_stat
, bl
);
1522 decode(last_osdmap_epoch
, bl
);
1523 decode(last_pg_scan
, bl
);
1525 decode(pool_statfs
, bl
);
1531 void PGMap::dump(ceph::Formatter
*f
, bool with_net
) const
1534 dump_pg_stats(f
, false);
1536 dump_osd_stats(f
, with_net
);
1539 void PGMap::dump_basic(ceph::Formatter
*f
) const
1541 f
->dump_unsigned("version", version
);
1542 f
->dump_stream("stamp") << stamp
;
1543 f
->dump_unsigned("last_osdmap_epoch", last_osdmap_epoch
);
1544 f
->dump_unsigned("last_pg_scan", last_pg_scan
);
1546 f
->open_object_section("pg_stats_sum");
1550 f
->open_object_section("osd_stats_sum");
1557 void PGMap::dump_delta(ceph::Formatter
*f
) const
1559 f
->open_object_section("pg_stats_delta");
1560 pg_sum_delta
.dump(f
);
1561 f
->dump_stream("stamp_delta") << stamp_delta
;
1565 void PGMap::dump_pg_stats(ceph::Formatter
*f
, bool brief
) const
1567 f
->open_array_section("pg_stats");
1568 for (auto i
= pg_stat
.begin();
1571 f
->open_object_section("pg_stat");
1572 f
->dump_stream("pgid") << i
->first
;
1574 i
->second
.dump_brief(f
);
1582 void PGMap::dump_pool_stats(ceph::Formatter
*f
) const
1584 f
->open_array_section("pool_stats");
1585 for (auto p
= pg_pool_sum
.begin();
1586 p
!= pg_pool_sum
.end();
1588 f
->open_object_section("pool_stat");
1589 f
->dump_int("poolid", p
->first
);
1590 auto q
= num_pg_by_pool
.find(p
->first
);
1591 if (q
!= num_pg_by_pool
.end())
1592 f
->dump_unsigned("num_pg", q
->second
);
1599 void PGMap::dump_osd_stats(ceph::Formatter
*f
, bool with_net
) const
1601 f
->open_array_section("osd_stats");
1602 for (auto q
= osd_stat
.begin();
1603 q
!= osd_stat
.end();
1605 f
->open_object_section("osd_stat");
1606 f
->dump_int("osd", q
->first
);
1607 q
->second
.dump(f
, with_net
);
1613 void PGMap::dump_osd_ping_times(ceph::Formatter
*f
) const
1615 f
->open_array_section("osd_ping_times");
1616 for (auto& [osd
, stat
] : osd_stat
) {
1617 f
->open_object_section("osd_ping_time");
1618 f
->dump_int("osd", osd
);
1619 stat
.dump_ping_time(f
);
1625 void PGMap::dump_pg_stats_plain(
1627 const mempool::pgmap::unordered_map
<pg_t
, pg_stat_t
>& pg_stats
,
1633 tab
.define_column("PG_STAT", TextTable::LEFT
, TextTable::LEFT
);
1634 tab
.define_column("STATE", TextTable::LEFT
, TextTable::RIGHT
);
1635 tab
.define_column("UP", TextTable::LEFT
, TextTable::RIGHT
);
1636 tab
.define_column("UP_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1637 tab
.define_column("ACTING", TextTable::LEFT
, TextTable::RIGHT
);
1638 tab
.define_column("ACTING_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1641 tab
.define_column("PG_STAT", TextTable::LEFT
, TextTable::LEFT
);
1642 tab
.define_column("OBJECTS", TextTable::LEFT
, TextTable::RIGHT
);
1643 tab
.define_column("MISSING_ON_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1644 tab
.define_column("DEGRADED", TextTable::LEFT
, TextTable::RIGHT
);
1645 tab
.define_column("MISPLACED", TextTable::LEFT
, TextTable::RIGHT
);
1646 tab
.define_column("UNFOUND", TextTable::LEFT
, TextTable::RIGHT
);
1647 tab
.define_column("BYTES", TextTable::LEFT
, TextTable::RIGHT
);
1648 tab
.define_column("OMAP_BYTES*", TextTable::LEFT
, TextTable::RIGHT
);
1649 tab
.define_column("OMAP_KEYS*", TextTable::LEFT
, TextTable::RIGHT
);
1650 tab
.define_column("LOG", TextTable::LEFT
, TextTable::RIGHT
);
1651 tab
.define_column("DISK_LOG", TextTable::LEFT
, TextTable::RIGHT
);
1652 tab
.define_column("STATE", TextTable::LEFT
, TextTable::RIGHT
);
1653 tab
.define_column("STATE_STAMP", TextTable::LEFT
, TextTable::RIGHT
);
1654 tab
.define_column("VERSION", TextTable::LEFT
, TextTable::RIGHT
);
1655 tab
.define_column("REPORTED", TextTable::LEFT
, TextTable::RIGHT
);
1656 tab
.define_column("UP", TextTable::LEFT
, TextTable::RIGHT
);
1657 tab
.define_column("UP_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1658 tab
.define_column("ACTING", TextTable::LEFT
, TextTable::RIGHT
);
1659 tab
.define_column("ACTING_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1660 tab
.define_column("LAST_SCRUB", TextTable::LEFT
, TextTable::RIGHT
);
1661 tab
.define_column("SCRUB_STAMP", TextTable::LEFT
, TextTable::RIGHT
);
1662 tab
.define_column("LAST_DEEP_SCRUB", TextTable::LEFT
, TextTable::RIGHT
);
1663 tab
.define_column("DEEP_SCRUB_STAMP", TextTable::LEFT
, TextTable::RIGHT
);
1664 tab
.define_column("SNAPTRIMQ_LEN", TextTable::LEFT
, TextTable::RIGHT
);
1667 for (auto i
= pg_stats
.begin();
1668 i
!= pg_stats
.end(); ++i
) {
1669 const pg_stat_t
&st(i
->second
);
1672 << pg_state_string(st
.state
)
1676 << st
.acting_primary
1677 << TextTable::endrow
;
1679 ostringstream reported
;
1680 reported
<< st
.reported_epoch
<< ":" << st
.reported_seq
;
1683 << st
.stats
.sum
.num_objects
1684 << st
.stats
.sum
.num_objects_missing_on_primary
1685 << st
.stats
.sum
.num_objects_degraded
1686 << st
.stats
.sum
.num_objects_misplaced
1687 << st
.stats
.sum
.num_objects_unfound
1688 << st
.stats
.sum
.num_bytes
1689 << st
.stats
.sum
.num_omap_bytes
1690 << st
.stats
.sum
.num_omap_keys
1692 << st
.ondisk_log_size
1693 << pg_state_string(st
.state
)
1697 << pg_vector_string(st
.up
)
1699 << pg_vector_string(st
.acting
)
1700 << st
.acting_primary
1702 << st
.last_scrub_stamp
1703 << st
.last_deep_scrub
1704 << st
.last_deep_scrub_stamp
1706 << TextTable::endrow
;
1713 void PGMap::dump(ostream
& ss
) const
1716 dump_pg_stats(ss
, false);
1717 dump_pool_stats(ss
, false);
1718 dump_pg_sum_stats(ss
, false);
1722 void PGMap::dump_basic(ostream
& ss
) const
1724 ss
<< "version " << version
<< std::endl
;
1725 ss
<< "stamp " << stamp
<< std::endl
;
1726 ss
<< "last_osdmap_epoch " << last_osdmap_epoch
<< std::endl
;
1727 ss
<< "last_pg_scan " << last_pg_scan
<< std::endl
;
1730 void PGMap::dump_pg_stats(ostream
& ss
, bool brief
) const
1732 dump_pg_stats_plain(ss
, pg_stat
, brief
);
1735 void PGMap::dump_pool_stats(ostream
& ss
, bool header
) const
1740 tab
.define_column("POOLID", TextTable::LEFT
, TextTable::LEFT
);
1741 tab
.define_column("OBJECTS", TextTable::LEFT
, TextTable::RIGHT
);
1742 tab
.define_column("MISSING_ON_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1743 tab
.define_column("DEGRADED", TextTable::LEFT
, TextTable::RIGHT
);
1744 tab
.define_column("MISPLACED", TextTable::LEFT
, TextTable::RIGHT
);
1745 tab
.define_column("UNFOUND", TextTable::LEFT
, TextTable::RIGHT
);
1746 tab
.define_column("BYTES", TextTable::LEFT
, TextTable::RIGHT
);
1747 tab
.define_column("OMAP_BYTES*", TextTable::LEFT
, TextTable::RIGHT
);
1748 tab
.define_column("OMAP_KEYS*", TextTable::LEFT
, TextTable::RIGHT
);
1749 tab
.define_column("LOG", TextTable::LEFT
, TextTable::RIGHT
);
1750 tab
.define_column("DISK_LOG", TextTable::LEFT
, TextTable::RIGHT
);
1752 tab
.define_column("", TextTable::LEFT
, TextTable::LEFT
);
1753 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1754 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1755 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1756 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1757 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1758 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1759 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1760 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1761 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1762 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1765 for (auto p
= pg_pool_sum
.begin();
1766 p
!= pg_pool_sum
.end();
1769 << p
->second
.stats
.sum
.num_objects
1770 << p
->second
.stats
.sum
.num_objects_missing_on_primary
1771 << p
->second
.stats
.sum
.num_objects_degraded
1772 << p
->second
.stats
.sum
.num_objects_misplaced
1773 << p
->second
.stats
.sum
.num_objects_unfound
1774 << p
->second
.stats
.sum
.num_bytes
1775 << p
->second
.stats
.sum
.num_omap_bytes
1776 << p
->second
.stats
.sum
.num_omap_keys
1777 << p
->second
.log_size
1778 << p
->second
.ondisk_log_size
1779 << TextTable::endrow
;
1785 void PGMap::dump_pg_sum_stats(ostream
& ss
, bool header
) const
1790 tab
.define_column("PG_STAT", TextTable::LEFT
, TextTable::LEFT
);
1791 tab
.define_column("OBJECTS", TextTable::LEFT
, TextTable::RIGHT
);
1792 tab
.define_column("MISSING_ON_PRIMARY", TextTable::LEFT
, TextTable::RIGHT
);
1793 tab
.define_column("DEGRADED", TextTable::LEFT
, TextTable::RIGHT
);
1794 tab
.define_column("MISPLACED", TextTable::LEFT
, TextTable::RIGHT
);
1795 tab
.define_column("UNFOUND", TextTable::LEFT
, TextTable::RIGHT
);
1796 tab
.define_column("BYTES", TextTable::LEFT
, TextTable::RIGHT
);
1797 tab
.define_column("OMAP_BYTES*", TextTable::LEFT
, TextTable::RIGHT
);
1798 tab
.define_column("OMAP_KEYS*", TextTable::LEFT
, TextTable::RIGHT
);
1799 tab
.define_column("LOG", TextTable::LEFT
, TextTable::RIGHT
);
1800 tab
.define_column("DISK_LOG", TextTable::LEFT
, TextTable::RIGHT
);
1802 tab
.define_column("", TextTable::LEFT
, TextTable::LEFT
);
1803 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1804 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1805 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1806 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1807 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1808 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1809 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1810 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1811 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1812 tab
.define_column("", TextTable::LEFT
, TextTable::RIGHT
);
1816 << pg_sum
.stats
.sum
.num_objects
1817 << pg_sum
.stats
.sum
.num_objects_missing_on_primary
1818 << pg_sum
.stats
.sum
.num_objects_degraded
1819 << pg_sum
.stats
.sum
.num_objects_misplaced
1820 << pg_sum
.stats
.sum
.num_objects_unfound
1821 << pg_sum
.stats
.sum
.num_bytes
1822 << pg_sum
.stats
.sum
.num_omap_bytes
1823 << pg_sum
.stats
.sum
.num_omap_keys
1825 << pg_sum
.ondisk_log_size
1826 << TextTable::endrow
;
1831 void PGMap::dump_osd_stats(ostream
& ss
) const
1835 tab
.define_column("OSD_STAT", TextTable::LEFT
, TextTable::LEFT
);
1836 tab
.define_column("USED", TextTable::LEFT
, TextTable::RIGHT
);
1837 tab
.define_column("AVAIL", TextTable::LEFT
, TextTable::RIGHT
);
1838 tab
.define_column("USED_RAW", TextTable::LEFT
, TextTable::RIGHT
);
1839 tab
.define_column("TOTAL", TextTable::LEFT
, TextTable::RIGHT
);
1840 tab
.define_column("HB_PEERS", TextTable::LEFT
, TextTable::RIGHT
);
1841 tab
.define_column("PG_SUM", TextTable::LEFT
, TextTable::RIGHT
);
1842 tab
.define_column("PRIMARY_PG_SUM", TextTable::LEFT
, TextTable::RIGHT
);
1844 for (auto p
= osd_stat
.begin();
1845 p
!= osd_stat
.end();
1848 << byte_u_t(p
->second
.statfs
.get_used())
1849 << byte_u_t(p
->second
.statfs
.available
)
1850 << byte_u_t(p
->second
.statfs
.get_used_raw())
1851 << byte_u_t(p
->second
.statfs
.total
)
1852 << p
->second
.hb_peers
1853 << get_num_pg_by_osd(p
->first
)
1854 << get_num_primary_pg_by_osd(p
->first
)
1855 << TextTable::endrow
;
1859 << byte_u_t(osd_sum
.statfs
.get_used())
1860 << byte_u_t(osd_sum
.statfs
.available
)
1861 << byte_u_t(osd_sum
.statfs
.get_used_raw())
1862 << byte_u_t(osd_sum
.statfs
.total
)
1863 << TextTable::endrow
;
1868 void PGMap::dump_osd_sum_stats(ostream
& ss
) const
1872 tab
.define_column("OSD_STAT", TextTable::LEFT
, TextTable::LEFT
);
1873 tab
.define_column("USED", TextTable::LEFT
, TextTable::RIGHT
);
1874 tab
.define_column("AVAIL", TextTable::LEFT
, TextTable::RIGHT
);
1875 tab
.define_column("USED_RAW", TextTable::LEFT
, TextTable::RIGHT
);
1876 tab
.define_column("TOTAL", TextTable::LEFT
, TextTable::RIGHT
);
1879 << byte_u_t(osd_sum
.statfs
.get_used())
1880 << byte_u_t(osd_sum
.statfs
.available
)
1881 << byte_u_t(osd_sum
.statfs
.get_used_raw())
1882 << byte_u_t(osd_sum
.statfs
.total
)
1883 << TextTable::endrow
;
1888 void PGMap::get_stuck_stats(
1889 int types
, const utime_t cutoff
,
1890 mempool::pgmap::unordered_map
<pg_t
, pg_stat_t
>& stuck_pgs
) const
1892 ceph_assert(types
!= 0);
1893 for (auto i
= pg_stat
.begin();
1896 utime_t val
= cutoff
; // don't care about >= cutoff so that is infinity
1898 if ((types
& STUCK_INACTIVE
) && !(i
->second
.state
& PG_STATE_ACTIVE
)) {
1899 if (i
->second
.last_active
< val
)
1900 val
= i
->second
.last_active
;
1903 if ((types
& STUCK_UNCLEAN
) && !(i
->second
.state
& PG_STATE_CLEAN
)) {
1904 if (i
->second
.last_clean
< val
)
1905 val
= i
->second
.last_clean
;
1908 if ((types
& STUCK_DEGRADED
) && (i
->second
.state
& PG_STATE_DEGRADED
)) {
1909 if (i
->second
.last_undegraded
< val
)
1910 val
= i
->second
.last_undegraded
;
1913 if ((types
& STUCK_UNDERSIZED
) && (i
->second
.state
& PG_STATE_UNDERSIZED
)) {
1914 if (i
->second
.last_fullsized
< val
)
1915 val
= i
->second
.last_fullsized
;
1918 if ((types
& STUCK_STALE
) && (i
->second
.state
& PG_STATE_STALE
)) {
1919 if (i
->second
.last_unstale
< val
)
1920 val
= i
->second
.last_unstale
;
1923 // val is now the earliest any of the requested stuck states began
1925 stuck_pgs
[i
->first
] = i
->second
;
1930 bool PGMap::get_stuck_counts(const utime_t cutoff
, map
<string
, int>& note
) const
1938 for (auto i
= pg_stat
.begin();
1941 if (! (i
->second
.state
& PG_STATE_ACTIVE
)) {
1942 if (i
->second
.last_active
< cutoff
)
1945 if (! (i
->second
.state
& PG_STATE_CLEAN
)) {
1946 if (i
->second
.last_clean
< cutoff
)
1949 if (i
->second
.state
& PG_STATE_DEGRADED
) {
1950 if (i
->second
.last_undegraded
< cutoff
)
1953 if (i
->second
.state
& PG_STATE_UNDERSIZED
) {
1954 if (i
->second
.last_fullsized
< cutoff
)
1957 if (i
->second
.state
& PG_STATE_STALE
) {
1958 if (i
->second
.last_unstale
< cutoff
)
1964 note
["stuck inactive"] = inactive
;
1967 note
["stuck unclean"] = unclean
;
1970 note
["stuck undersized"] = undersized
;
1973 note
["stuck degraded"] = degraded
;
1976 note
["stuck stale"] = stale
;
1978 return inactive
|| unclean
|| undersized
|| degraded
|| stale
;
1981 void PGMap::dump_stuck(ceph::Formatter
*f
, int types
, utime_t cutoff
) const
1983 mempool::pgmap::unordered_map
<pg_t
, pg_stat_t
> stuck_pg_stats
;
1984 get_stuck_stats(types
, cutoff
, stuck_pg_stats
);
1985 f
->open_array_section("stuck_pg_stats");
1986 for (auto i
= stuck_pg_stats
.begin();
1987 i
!= stuck_pg_stats
.end();
1989 f
->open_object_section("pg_stat");
1990 f
->dump_stream("pgid") << i
->first
;
1997 void PGMap::dump_stuck_plain(ostream
& ss
, int types
, utime_t cutoff
) const
1999 mempool::pgmap::unordered_map
<pg_t
, pg_stat_t
> stuck_pg_stats
;
2000 get_stuck_stats(types
, cutoff
, stuck_pg_stats
);
2001 if (!stuck_pg_stats
.empty())
2002 dump_pg_stats_plain(ss
, stuck_pg_stats
, true);
2005 int PGMap::dump_stuck_pg_stats(
2009 vector
<string
>& args
) const
2011 int stuck_types
= 0;
2013 for (auto i
= args
.begin(); i
!= args
.end(); ++i
) {
2014 if (*i
== "inactive")
2015 stuck_types
|= PGMap::STUCK_INACTIVE
;
2016 else if (*i
== "unclean")
2017 stuck_types
|= PGMap::STUCK_UNCLEAN
;
2018 else if (*i
== "undersized")
2019 stuck_types
|= PGMap::STUCK_UNDERSIZED
;
2020 else if (*i
== "degraded")
2021 stuck_types
|= PGMap::STUCK_DEGRADED
;
2022 else if (*i
== "stale")
2023 stuck_types
|= PGMap::STUCK_STALE
;
2025 ds
<< "Unknown type: " << *i
<< std::endl
;
2030 utime_t
now(ceph_clock_now());
2031 utime_t cutoff
= now
- utime_t(threshold
, 0);
2034 dump_stuck_plain(ds
, stuck_types
, cutoff
);
2036 dump_stuck(f
, stuck_types
, cutoff
);
2043 void PGMap::dump_osd_perf_stats(ceph::Formatter
*f
) const
2045 f
->open_array_section("osd_perf_infos");
2046 for (auto i
= osd_stat
.begin();
2047 i
!= osd_stat
.end();
2049 f
->open_object_section("osd");
2050 f
->dump_int("id", i
->first
);
2052 f
->open_object_section("perf_stats");
2053 i
->second
.os_perf_stat
.dump(f
);
2060 void PGMap::print_osd_perf_stats(std::ostream
*ss
) const
2063 tab
.define_column("osd", TextTable::LEFT
, TextTable::RIGHT
);
2064 tab
.define_column("commit_latency(ms)", TextTable::LEFT
, TextTable::RIGHT
);
2065 tab
.define_column("apply_latency(ms)", TextTable::LEFT
, TextTable::RIGHT
);
2066 for (auto i
= osd_stat
.begin();
2067 i
!= osd_stat
.end();
2070 tab
<< i
->second
.os_perf_stat
.os_commit_latency_ns
/ 1000000ull;
2071 tab
<< i
->second
.os_perf_stat
.os_apply_latency_ns
/ 1000000ull;
2072 tab
<< TextTable::endrow
;
2077 void PGMap::dump_osd_blocked_by_stats(ceph::Formatter
*f
) const
2079 f
->open_array_section("osd_blocked_by_infos");
2080 for (auto i
= blocked_by_sum
.begin();
2081 i
!= blocked_by_sum
.end();
2083 f
->open_object_section("osd");
2084 f
->dump_int("id", i
->first
);
2085 f
->dump_int("num_blocked", i
->second
);
2090 void PGMap::print_osd_blocked_by_stats(std::ostream
*ss
) const
2093 tab
.define_column("osd", TextTable::LEFT
, TextTable::RIGHT
);
2094 tab
.define_column("num_blocked", TextTable::LEFT
, TextTable::RIGHT
);
2095 for (auto i
= blocked_by_sum
.begin();
2096 i
!= blocked_by_sum
.end();
2100 tab
<< TextTable::endrow
;
2107 * update aggregated delta
2109 * @param cct ceph context
2110 * @param ts Timestamp for the stats being delta'ed
2111 * @param old_pool_sum Previous stats sum
2112 * @param last_ts Last timestamp for pool
2113 * @param result_pool_sum Resulting stats
2114 * @param result_pool_delta Resulting pool delta
2115 * @param result_ts_delta Resulting timestamp delta
2116 * @param delta_avg_list List of last N computed deltas, used to average
2118 void PGMap::update_delta(
2121 const pool_stat_t
& old_pool_sum
,
2123 const pool_stat_t
& current_pool_sum
,
2124 pool_stat_t
*result_pool_delta
,
2125 utime_t
*result_ts_delta
,
2126 mempool::pgmap::list
<pair
<pool_stat_t
,utime_t
> > *delta_avg_list
)
2128 /* @p ts is the timestamp we want to associate with the data
2129 * in @p old_pool_sum, and on which we will base ourselves to
2130 * calculate the delta, stored in 'delta_t'.
2133 delta_t
= ts
; // start with the provided timestamp
2134 delta_t
-= *last_ts
; // take the last timestamp we saw
2135 *last_ts
= ts
; // @p ts becomes the last timestamp we saw
2137 // adjust delta_t, quick start if there is no update in a long period
2138 delta_t
= std::min(delta_t
,
2139 utime_t(2 * (cct
? cct
->_conf
->mon_delta_reset_interval
: 10), 0));
2141 // calculate a delta, and average over the last 6 deltas by default.
2142 /* start by taking a copy of our current @p result_pool_sum, and by
2143 * taking out the stats from @p old_pool_sum. This generates a stats
2144 * delta. Stash this stats delta in @p delta_avg_list, along with the
2145 * timestamp delta for these results.
2147 pool_stat_t d
= current_pool_sum
;
2148 d
.stats
.sub(old_pool_sum
.stats
);
2150 /* Aggregate current delta, and take out the last seen delta (if any) to
2152 * Skip calculating delta while sum was not synchronized.
2154 if(!old_pool_sum
.stats
.sum
.is_zero()) {
2155 delta_avg_list
->push_back(make_pair(d
,delta_t
));
2156 *result_ts_delta
+= delta_t
;
2157 result_pool_delta
->stats
.add(d
.stats
);
2159 size_t s
= cct
? cct
->_conf
.get_val
<uint64_t>("mon_stat_smooth_intervals") : 1;
2160 while (delta_avg_list
->size() > s
) {
2161 result_pool_delta
->stats
.sub(delta_avg_list
->front().first
.stats
);
2162 *result_ts_delta
-= delta_avg_list
->front().second
;
2163 delta_avg_list
->pop_front();
2168 * Update a given pool's deltas
2170 * @param cct Ceph Context
2171 * @param ts Timestamp for the stats being delta'ed
2172 * @param pool Pool's id
2173 * @param old_pool_sum Previous stats sum
2175 void PGMap::update_one_pool_delta(
2179 const pool_stat_t
& old_pool_sum
)
2181 if (per_pool_sum_deltas
.count(pool
) == 0) {
2182 ceph_assert(per_pool_sum_deltas_stamps
.count(pool
) == 0);
2183 ceph_assert(per_pool_sum_delta
.count(pool
) == 0);
2186 auto& sum_delta
= per_pool_sum_delta
[pool
];
2188 update_delta(cct
, ts
, old_pool_sum
, &sum_delta
.second
, pg_pool_sum
[pool
],
2189 &sum_delta
.first
, &per_pool_sum_deltas_stamps
[pool
],
2190 &per_pool_sum_deltas
[pool
]);
2194 * Update pools' deltas
2196 * @param cct CephContext
2197 * @param ts Timestamp for the stats being delta'ed
2198 * @param pg_pool_sum_old Map of pool stats for delta calcs.
2200 void PGMap::update_pool_deltas(
2201 CephContext
*cct
, const utime_t ts
,
2202 const mempool::pgmap::unordered_map
<int32_t,pool_stat_t
>& pg_pool_sum_old
)
2204 for (auto it
= pg_pool_sum_old
.begin();
2205 it
!= pg_pool_sum_old
.end(); ++it
) {
2206 update_one_pool_delta(cct
, ts
, it
->first
, it
->second
);
2210 void PGMap::clear_delta()
2212 pg_sum_delta
= pool_stat_t();
2213 pg_sum_deltas
.clear();
2214 stamp_delta
= utime_t();
2217 void PGMap::generate_test_instances(list
<PGMap
*>& o
)
2219 o
.push_back(new PGMap
);
2220 list
<Incremental
*> inc
;
2221 Incremental::generate_test_instances(inc
);
2224 while (!inc
.empty()) {
2225 PGMap
*pmp
= new PGMap();
2228 o
.back()->apply_incremental(NULL
, *inc
.front());
2234 void PGMap::get_filtered_pg_stats(uint64_t state
, int64_t poolid
, int64_t osdid
,
2235 bool primary
, set
<pg_t
>& pgs
) const
2237 for (auto i
= pg_stat
.begin();
2240 if ((poolid
>= 0) && (poolid
!= i
->first
.pool()))
2242 if ((osdid
>= 0) && !(i
->second
.is_acting_osd(osdid
,primary
)))
2244 if (state
== (uint64_t)-1 || // "all"
2245 (i
->second
.state
& state
) || // matches a state bit
2246 (state
== 0 && i
->second
.state
== 0)) { // matches "unknown" (== 0)
2247 pgs
.insert(i
->first
);
2252 void PGMap::dump_filtered_pg_stats(ceph::Formatter
*f
, set
<pg_t
>& pgs
) const
2254 f
->open_array_section("pg_stats");
2255 for (auto i
= pgs
.begin(); i
!= pgs
.end(); ++i
) {
2256 const pg_stat_t
& st
= pg_stat
.at(*i
);
2257 f
->open_object_section("pg_stat");
2258 f
->dump_stream("pgid") << *i
;
2265 void PGMap::dump_filtered_pg_stats(ostream
& ss
, set
<pg_t
>& pgs
) const
2268 utime_t now
= ceph_clock_now();
2270 tab
.define_column("PG", TextTable::LEFT
, TextTable::LEFT
);
2271 tab
.define_column("OBJECTS", TextTable::LEFT
, TextTable::RIGHT
);
2272 tab
.define_column("DEGRADED", TextTable::LEFT
, TextTable::RIGHT
);
2273 tab
.define_column("MISPLACED", TextTable::LEFT
, TextTable::RIGHT
);
2274 tab
.define_column("UNFOUND", TextTable::LEFT
, TextTable::RIGHT
);
2275 tab
.define_column("BYTES", TextTable::LEFT
, TextTable::RIGHT
);
2276 tab
.define_column("OMAP_BYTES*", TextTable::LEFT
, TextTable::RIGHT
);
2277 tab
.define_column("OMAP_KEYS*", TextTable::LEFT
, TextTable::RIGHT
);
2278 tab
.define_column("LOG", TextTable::LEFT
, TextTable::RIGHT
);
2279 tab
.define_column("STATE", TextTable::LEFT
, TextTable::RIGHT
);
2280 tab
.define_column("SINCE", TextTable::LEFT
, TextTable::RIGHT
);
2281 tab
.define_column("VERSION", TextTable::LEFT
, TextTable::RIGHT
);
2282 tab
.define_column("REPORTED", TextTable::LEFT
, TextTable::RIGHT
);
2283 tab
.define_column("UP", TextTable::LEFT
, TextTable::RIGHT
);
2284 tab
.define_column("ACTING", TextTable::LEFT
, TextTable::RIGHT
);
2285 tab
.define_column("SCRUB_STAMP", TextTable::LEFT
, TextTable::RIGHT
);
2286 tab
.define_column("DEEP_SCRUB_STAMP", TextTable::LEFT
, TextTable::RIGHT
);
2288 for (auto i
= pgs
.begin(); i
!= pgs
.end(); ++i
) {
2289 const pg_stat_t
& st
= pg_stat
.at(*i
);
2291 ostringstream reported
;
2292 reported
<< st
.reported_epoch
<< ":" << st
.reported_seq
;
2294 ostringstream upstr
, actingstr
;
2295 upstr
<< pg_vector_string(st
.up
) << 'p' << st
.up_primary
;
2296 actingstr
<< pg_vector_string(st
.acting
) << 'p' << st
.acting_primary
;
2298 << st
.stats
.sum
.num_objects
2299 << st
.stats
.sum
.num_objects_degraded
2300 << st
.stats
.sum
.num_objects_misplaced
2301 << st
.stats
.sum
.num_objects_unfound
2302 << st
.stats
.sum
.num_bytes
2303 << st
.stats
.sum
.num_omap_bytes
2304 << st
.stats
.sum
.num_omap_keys
2306 << pg_state_string(st
.state
)
2307 << utimespan_str(now
- st
.last_change
)
2312 << st
.last_scrub_stamp
2313 << st
.last_deep_scrub_stamp
2314 << TextTable::endrow
;
2320 void PGMap::dump_pool_stats_and_io_rate(int64_t poolid
, const OSDMap
&osd_map
,
2322 stringstream
*rs
) const {
2323 string pool_name
= osd_map
.get_pool_name(poolid
);
2325 f
->open_object_section("pool");
2326 f
->dump_string("pool_name", pool_name
.c_str());
2327 f
->dump_int("pool_id", poolid
);
2328 f
->open_object_section("recovery");
2332 pool_recovery_summary(f
, &sl
, poolid
);
2333 if (!f
&& !sl
.empty()) {
2335 tss
<< " " << p
<< "\n";
2338 f
->close_section(); // object section recovery
2339 f
->open_object_section("recovery_rate");
2342 pool_recovery_rate_summary(f
, &rss
, poolid
);
2343 if (!f
&& !rss
.str().empty())
2344 tss
<< " recovery io " << rss
.str() << "\n";
2346 f
->close_section(); // object section recovery_rate
2347 f
->open_object_section("client_io_rate");
2351 pool_client_io_rate_summary(f
, &rss
, poolid
);
2352 if (!f
&& !rss
.str().empty())
2353 tss
<< " client io " << rss
.str() << "\n";
2354 // dump cache tier IO rate for cache pool
2355 const pg_pool_t
*pool
= osd_map
.get_pg_pool(poolid
);
2356 if (pool
->is_tier()) {
2358 f
->close_section(); // object section client_io_rate
2359 f
->open_object_section("cache_io_rate");
2363 pool_cache_io_rate_summary(f
, &rss
, poolid
);
2364 if (!f
&& !rss
.str().empty())
2365 tss
<< " cache tier io " << rss
.str() << "\n";
2368 f
->close_section(); // object section cache_io_rate
2369 f
->close_section(); // object section pool
2371 *rs
<< "pool " << pool_name
<< " id " << poolid
<< "\n";
2372 if (!tss
.str().empty())
2373 *rs
<< tss
.str() << "\n";
2375 *rs
<< " nothing is going on\n\n";
2379 // Get crush parentage for an osd (skip root)
2380 set
<std::string
> PGMap::osd_parentage(const OSDMap
& osdmap
, int id
) const
2382 set
<std::string
> reporters_by_subtree
;
2383 auto reporter_subtree_level
= g_conf().get_val
<string
>("mon_osd_reporter_subtree_level");
2385 auto loc
= osdmap
.crush
->get_full_location(id
);
2386 for (auto& [parent_bucket_type
, parent_id
] : loc
) {
2387 // Should we show the root? Might not be too informative like "default"
2388 if (parent_bucket_type
!= "root" &&
2389 parent_bucket_type
!= reporter_subtree_level
) {
2390 reporters_by_subtree
.insert(parent_id
);
2393 return reporters_by_subtree
;
2396 void PGMap::get_health_checks(
2398 const OSDMap
& osdmap
,
2399 health_check_map_t
*checks
) const
2401 utime_t now
= ceph_clock_now();
2402 const auto max
= cct
->_conf
.get_val
<uint64_t>("mon_health_max_detail");
2403 const auto& pools
= osdmap
.get_pools();
2405 typedef enum pg_consequence_t
{
2406 UNAVAILABLE
= 1, // Client IO to the pool may block
2407 DEGRADED
= 2, // Fewer than the requested number of replicas are present
2408 BACKFILL_FULL
= 3, // Backfill is blocked for space considerations
2409 // This may or may not be a deadlock condition.
2410 DAMAGED
= 4, // The data may be missing or inconsistent on disk and
2412 RECOVERY_FULL
= 5 // Recovery is blocked because OSDs are full
2415 // For a given PG state, how should it be reported at the pool level?
2416 class PgStateResponse
{
2418 pg_consequence_t consequence
;
2419 typedef std::function
< utime_t(const pg_stat_t
&) > stuck_cb
;
2420 stuck_cb stuck_since
;
2423 PgStateResponse(const pg_consequence_t
& c
, stuck_cb
&& s
)
2424 : consequence(c
), stuck_since(std::move(s
)), invert(false)
2428 PgStateResponse(const pg_consequence_t
& c
, stuck_cb
&& s
, bool i
)
2429 : consequence(c
), stuck_since(std::move(s
)), invert(i
)
2434 // Record the PG state counts that contributed to a reported pool state
2437 // Map of PG_STATE_* to number of pgs in that state.
2438 std::map
<unsigned, unsigned> states
;
2440 // List of all PG IDs that had a state contributing
2441 // to this health condition.
2444 std::map
<pg_t
, std::string
> pg_messages
;
2447 // Map of PG state to how to respond to it
2448 std::map
<unsigned, PgStateResponse
> state_to_response
= {
2449 // Immediate reports
2450 { PG_STATE_INCONSISTENT
, {DAMAGED
, {}} },
2451 { PG_STATE_INCOMPLETE
, {UNAVAILABLE
, {}} },
2452 { PG_STATE_SNAPTRIM_ERROR
, {DAMAGED
, {}} },
2453 { PG_STATE_RECOVERY_UNFOUND
, {DAMAGED
, {}} },
2454 { PG_STATE_BACKFILL_UNFOUND
, {DAMAGED
, {}} },
2455 { PG_STATE_BACKFILL_TOOFULL
, {BACKFILL_FULL
, {}} },
2456 { PG_STATE_RECOVERY_TOOFULL
, {RECOVERY_FULL
, {}} },
2457 { PG_STATE_DEGRADED
, {DEGRADED
, {}} },
2458 { PG_STATE_DOWN
, {UNAVAILABLE
, {}} },
2459 // Delayed (wait until stuck) reports
2460 { PG_STATE_PEERING
, {UNAVAILABLE
, [](const pg_stat_t
&p
){return p
.last_peered
;} } },
2461 { PG_STATE_UNDERSIZED
, {DEGRADED
, [](const pg_stat_t
&p
){return p
.last_fullsized
;} } },
2462 { PG_STATE_STALE
, {UNAVAILABLE
, [](const pg_stat_t
&p
){return p
.last_unstale
;} } },
2463 // Delayed and inverted reports
2464 { PG_STATE_ACTIVE
, {UNAVAILABLE
, [](const pg_stat_t
&p
){return p
.last_active
;}, true} }
2467 // Specialized state printer that takes account of inversion of
2468 // ACTIVE, CLEAN checks.
2469 auto state_name
= [](const uint64_t &state
) {
2470 // Special cases for the states that are inverted checks
2471 if (state
== PG_STATE_CLEAN
) {
2472 return std::string("unclean");
2473 } else if (state
== PG_STATE_ACTIVE
) {
2474 return std::string("inactive");
2476 return pg_state_string(state
);
2480 // Map of what is wrong to information about why, implicitly also stores
2481 // the list of what is wrong.
2482 std::map
<pg_consequence_t
, PgCauses
> detected
;
2484 // Optimisation: trim down the number of checks to apply based on
2485 // the summary counters
2486 std::map
<unsigned, PgStateResponse
> possible_responses
;
2487 for (const auto &i
: num_pg_by_state
) {
2488 for (const auto &j
: state_to_response
) {
2489 if (!j
.second
.invert
) {
2490 // Check for normal tests by seeing if any pgs have the flag
2491 if (i
.first
& j
.first
) {
2492 possible_responses
.insert(j
);
2498 for (const auto &j
: state_to_response
) {
2499 if (j
.second
.invert
) {
2500 // Check for inverted tests by seeing if not-all pgs have the flag
2501 const auto &found
= num_pg_by_state
.find(j
.first
);
2502 if (found
== num_pg_by_state
.end() || found
->second
!= num_pg
) {
2503 possible_responses
.insert(j
);
2508 utime_t cutoff
= now
- utime_t(cct
->_conf
.get_val
<int64_t>("mon_pg_stuck_threshold"), 0);
2509 // Loop over all PGs, if there are any possibly-unhealthy states in there
2510 if (!possible_responses
.empty()) {
2511 for (const auto& i
: pg_stat
) {
2512 const auto &pg_id
= i
.first
;
2513 const auto &pg_info
= i
.second
;
2515 for (const auto &j
: state_to_response
) {
2516 const auto &pg_response_state
= j
.first
;
2517 const auto &pg_response
= j
.second
;
2519 // Apply the state test
2520 if (!(bool(pg_info
.state
& pg_response_state
) != pg_response
.invert
)) {
2524 // Apply stuckness test if needed
2525 if (pg_response
.stuck_since
) {
2526 // Delayed response, check for stuckness
2527 utime_t last_whatever
= pg_response
.stuck_since(pg_info
);
2528 if (last_whatever
>= cutoff
) {
2529 // Not stuck enough, ignore.
2536 auto &causes
= detected
[pg_response
.consequence
];
2537 causes
.states
[pg_response_state
]++;
2538 causes
.pgs
.insert(pg_id
);
2540 // Don't bother composing detail string if we have already recorded
2542 if (causes
.pg_messages
.size() > max
) {
2546 std::ostringstream ss
;
2547 if (pg_response
.stuck_since
) {
2548 utime_t since
= pg_response
.stuck_since(pg_info
);
2549 ss
<< "pg " << pg_id
<< " is stuck " << state_name(pg_response_state
);
2550 if (since
== utime_t()) {
2551 ss
<< " since forever";
2553 utime_t dur
= now
- since
;
2554 ss
<< " for " << utimespan_str(dur
);
2556 ss
<< ", current state " << pg_state_string(pg_info
.state
)
2557 << ", last acting " << pg_info
.acting
;
2559 ss
<< "pg " << pg_id
<< " is "
2560 << pg_state_string(pg_info
.state
);
2561 ss
<< ", acting " << pg_info
.acting
;
2562 if (pg_info
.stats
.sum
.num_objects_unfound
) {
2563 ss
<< ", " << pg_info
.stats
.sum
.num_objects_unfound
2568 if (pg_info
.state
& PG_STATE_INCOMPLETE
) {
2569 const pg_pool_t
*pi
= osdmap
.get_pg_pool(pg_id
.pool());
2570 if (pi
&& pi
->min_size
> 1) {
2571 ss
<< " (reducing pool "
2572 << osdmap
.get_pool_name(pg_id
.pool())
2573 << " min_size from " << (int)pi
->min_size
2574 << " may help; search ceph.com/docs for 'incomplete')";
2578 causes
.pg_messages
[pg_id
] = ss
.str();
2582 dout(10) << __func__
<< " skipping loop over PGs: counters look OK" << dendl
;
2585 for (const auto &i
: detected
) {
2586 std::string health_code
;
2587 health_status_t sev
;
2588 std::string summary
;
2591 health_code
= "PG_AVAILABILITY";
2593 summary
= "Reduced data availability: ";
2596 health_code
= "PG_DEGRADED";
2597 summary
= "Degraded data redundancy: ";
2601 health_code
= "PG_BACKFILL_FULL";
2602 summary
= "Low space hindering backfill (add storage if this doesn't resolve itself): ";
2606 health_code
= "PG_DAMAGED";
2607 summary
= "Possible data damage: ";
2611 health_code
= "PG_RECOVERY_FULL";
2612 summary
= "Full OSDs blocking recovery: ";
2619 if (i
.first
== DEGRADED
) {
2620 if (pg_sum
.stats
.sum
.num_objects_degraded
&&
2621 pg_sum
.stats
.sum
.num_object_copies
> 0) {
2622 double pc
= (double)pg_sum
.stats
.sum
.num_objects_degraded
/
2623 (double)pg_sum
.stats
.sum
.num_object_copies
* (double)100.0;
2625 snprintf(b
, sizeof(b
), "%.3lf", pc
);
2627 ss
<< pg_sum
.stats
.sum
.num_objects_degraded
2628 << "/" << pg_sum
.stats
.sum
.num_object_copies
<< " objects degraded ("
2631 // Throw in a comma for the benefit of the following PG counts
2632 summary
+= ss
.str() + ", ";
2636 // Compose summary message saying how many PGs in what states led
2637 // to this health check failing
2638 std::vector
<std::string
> pg_msgs
;
2640 for (const auto &j
: i
.second
.states
) {
2641 std::ostringstream msg
;
2642 msg
<< j
.second
<< (j
.second
> 1 ? " pgs " : " pg ") << state_name(j
.first
);
2643 pg_msgs
.push_back(msg
.str());
2646 summary
+= joinify(pg_msgs
.begin(), pg_msgs
.end(), std::string(", "));
2648 health_check_t
*check
= &checks
->add(
2654 // Compose list of PGs contributing to this health check failing
2655 for (const auto &j
: i
.second
.pg_messages
) {
2656 check
->detail
.push_back(j
.second
);
2661 if (pg_sum
.stats
.sum
.num_scrub_errors
) {
2663 ss
<< pg_sum
.stats
.sum
.num_scrub_errors
<< " scrub errors";
2664 checks
->add("OSD_SCRUB_ERRORS", HEALTH_ERR
, ss
.str(),
2665 pg_sum
.stats
.sum
.num_scrub_errors
);
2668 // LARGE_OMAP_OBJECTS
2669 if (pg_sum
.stats
.sum
.num_large_omap_objects
) {
2670 list
<string
> detail
;
2671 for (auto &pool
: pools
) {
2672 const string
& pool_name
= osdmap
.get_pool_name(pool
.first
);
2673 auto it2
= pg_pool_sum
.find(pool
.first
);
2674 if (it2
== pg_pool_sum
.end()) {
2677 const pool_stat_t
*pstat
= &it2
->second
;
2678 if (pstat
== nullptr) {
2681 const object_stat_sum_t
& sum
= pstat
->stats
.sum
;
2682 if (sum
.num_large_omap_objects
) {
2684 ss
<< sum
.num_large_omap_objects
<< " large objects found in pool "
2685 << "'" << pool_name
<< "'";
2686 detail
.push_back(ss
.str());
2689 if (!detail
.empty()) {
2691 ss
<< pg_sum
.stats
.sum
.num_large_omap_objects
<< " large omap objects";
2692 auto& d
= checks
->add("LARGE_OMAP_OBJECTS", HEALTH_WARN
, ss
.str(),
2693 pg_sum
.stats
.sum
.num_large_omap_objects
);
2695 tip
<< "Search the cluster log for 'Large omap object found' for more "
2697 detail
.push_back(tip
.str());
2698 d
.detail
.swap(detail
);
2702 // CACHE_POOL_NEAR_FULL
2704 list
<string
> detail
;
2705 unsigned num_pools
= 0;
2706 for (auto& p
: pools
) {
2707 if ((!p
.second
.target_max_objects
&& !p
.second
.target_max_bytes
) ||
2708 !pg_pool_sum
.count(p
.first
)) {
2711 bool nearfull
= false;
2712 const string
& name
= osdmap
.get_pool_name(p
.first
);
2713 const pool_stat_t
& st
= get_pg_pool_sum_stat(p
.first
);
2714 uint64_t ratio
= p
.second
.cache_target_full_ratio_micro
+
2715 ((1000000 - p
.second
.cache_target_full_ratio_micro
) *
2716 cct
->_conf
->mon_cache_target_full_warn_ratio
);
2717 if (p
.second
.target_max_objects
&&
2718 (uint64_t)(st
.stats
.sum
.num_objects
-
2719 st
.stats
.sum
.num_objects_hit_set_archive
) >
2720 p
.second
.target_max_objects
* (ratio
/ 1000000.0)) {
2722 ss
<< "cache pool '" << name
<< "' with "
2723 << si_u_t(st
.stats
.sum
.num_objects
)
2724 << " objects at/near target max "
2725 << si_u_t(p
.second
.target_max_objects
) << " objects";
2726 detail
.push_back(ss
.str());
2729 if (p
.second
.target_max_bytes
&&
2730 (uint64_t)(st
.stats
.sum
.num_bytes
-
2731 st
.stats
.sum
.num_bytes_hit_set_archive
) >
2732 p
.second
.target_max_bytes
* (ratio
/ 1000000.0)) {
2734 ss
<< "cache pool '" << name
2735 << "' with " << byte_u_t(st
.stats
.sum
.num_bytes
)
2736 << " at/near target max "
2737 << byte_u_t(p
.second
.target_max_bytes
);
2738 detail
.push_back(ss
.str());
2745 if (!detail
.empty()) {
2747 ss
<< num_pools
<< " cache pools at or near target size";
2748 auto& d
= checks
->add("CACHE_POOL_NEAR_FULL", HEALTH_WARN
, ss
.str(),
2750 d
.detail
.swap(detail
);
2755 unsigned num_in
= osdmap
.get_num_in_osds();
2756 auto sum_pg_up
= std::max(static_cast<size_t>(pg_sum
.up
), pg_stat
.size());
2757 const auto min_pg_per_osd
=
2758 cct
->_conf
.get_val
<uint64_t>("mon_pg_warn_min_per_osd");
2759 if (num_in
&& min_pg_per_osd
> 0 && osdmap
.get_pools().size() > 0) {
2760 auto per
= sum_pg_up
/ num_in
;
2761 if (per
< min_pg_per_osd
&& per
) {
2763 ss
<< "too few PGs per OSD (" << per
2764 << " < min " << min_pg_per_osd
<< ")";
2765 checks
->add("TOO_FEW_PGS", HEALTH_WARN
, ss
.str(),
2766 min_pg_per_osd
- per
);
2771 auto max_pg_per_osd
= cct
->_conf
.get_val
<uint64_t>("mon_max_pg_per_osd");
2772 if (num_in
&& max_pg_per_osd
> 0) {
2773 auto per
= sum_pg_up
/ num_in
;
2774 if (per
> max_pg_per_osd
) {
2776 ss
<< "too many PGs per OSD (" << per
2777 << " > max " << max_pg_per_osd
<< ")";
2778 checks
->add("TOO_MANY_PGS", HEALTH_WARN
, ss
.str(),
2779 per
- max_pg_per_osd
);
2784 auto warn_too_few_osds
= cct
->_conf
.get_val
<bool>("mon_warn_on_too_few_osds");
2785 auto osd_pool_default_size
= cct
->_conf
.get_val
<uint64_t>("osd_pool_default_size");
2786 if (warn_too_few_osds
&& osdmap
.get_num_osds() < osd_pool_default_size
) {
2788 ss
<< "OSD count " << osdmap
.get_num_osds()
2789 << " < osd_pool_default_size " << osd_pool_default_size
;
2790 checks
->add("TOO_FEW_OSDS", HEALTH_WARN
, ss
.str(),
2791 osd_pool_default_size
- osdmap
.get_num_osds());
2795 // Convert milliseconds to microseconds
2796 auto warn_slow_ping_time
= cct
->_conf
.get_val
<double>("mon_warn_on_slow_ping_time") * 1000;
2797 auto grace
= cct
->_conf
.get_val
<int64_t>("osd_heartbeat_grace");
2798 if (warn_slow_ping_time
== 0) {
2799 double ratio
= cct
->_conf
.get_val
<double>("mon_warn_on_slow_ping_ratio");
2800 warn_slow_ping_time
= grace
;
2801 warn_slow_ping_time
*= 1000000 * ratio
; // Seconds of grace to microseconds at ratio
2803 if (warn_slow_ping_time
> 0) {
2805 struct mon_ping_item_t
{
2811 bool operator<(const mon_ping_item_t
& rhs
) const {
2812 if (pingtime
< rhs
.pingtime
)
2814 if (pingtime
> rhs
.pingtime
)
2816 if (from
< rhs
.from
)
2818 if (from
> rhs
.from
)
2824 list
<string
> detail_back
;
2825 list
<string
> detail_front
;
2826 set
<mon_ping_item_t
> back_sorted
, front_sorted
;
2827 for (auto i
: osd_stat
) {
2828 for (auto j
: i
.second
.hb_pingtime
) {
2830 // Maybe source info is old
2831 if (now
.sec() - j
.second
.last_update
> grace
* 60)
2834 mon_ping_item_t back
;
2835 back
.pingtime
= std::max(j
.second
.back_pingtime
[0], j
.second
.back_pingtime
[1]);
2836 back
.pingtime
= std::max(back
.pingtime
, j
.second
.back_pingtime
[2]);
2837 back
.from
= i
.first
;
2839 if (back
.pingtime
> warn_slow_ping_time
) {
2840 back
.improving
= (j
.second
.back_pingtime
[0] < j
.second
.back_pingtime
[1]
2841 && j
.second
.back_pingtime
[1] < j
.second
.back_pingtime
[2]);
2842 back_sorted
.emplace(back
);
2845 mon_ping_item_t front
;
2846 front
.pingtime
= std::max(j
.second
.front_pingtime
[0], j
.second
.front_pingtime
[1]);
2847 front
.pingtime
= std::max(front
.pingtime
, j
.second
.front_pingtime
[2]);
2848 front
.from
= i
.first
;
2850 if (front
.pingtime
> warn_slow_ping_time
) {
2851 front
.improving
= (j
.second
.front_pingtime
[0] < j
.second
.front_pingtime
[1]
2852 && j
.second
.front_pingtime
[1] < j
.second
.back_pingtime
[2]);
2853 front_sorted
.emplace(front
);
2857 int max_detail
= 10;
2858 for (auto &sback
: boost::adaptors::reverse(back_sorted
)) {
2860 if (max_detail
== 0) {
2861 ss
<< "Truncated long network list. Use ceph daemon mgr.# dump_osd_network for more information";
2862 detail_back
.push_back(ss
.str());
2866 ss
<< "Slow OSD heartbeats on back from osd." << sback
.from
2867 << " [" << osd_parentage(osdmap
, sback
.from
) << "]"
2868 << (osdmap
.is_down(sback
.from
) ? " (down)" : "")
2869 << " to osd." << sback
.to
2870 << " [" << osd_parentage(osdmap
, sback
.to
) << "]"
2871 << (osdmap
.is_down(sback
.to
) ? " (down)" : "")
2872 << " " << fixed_u_to_string(sback
.pingtime
, 3) << " msec"
2873 << (sback
.improving
? " possibly improving" : "");
2874 detail_back
.push_back(ss
.str());
2877 for (auto &sfront
: boost::adaptors::reverse(front_sorted
)) {
2879 if (max_detail
== 0) {
2880 ss
<< "Truncated long network list. Use ceph daemon mgr.# dump_osd_network for more information";
2881 detail_front
.push_back(ss
.str());
2885 // Get crush parentage for each osd
2886 ss
<< "Slow OSD heartbeats on front from osd." << sfront
.from
2887 << " [" << osd_parentage(osdmap
, sfront
.from
) << "]"
2888 << (osdmap
.is_down(sfront
.from
) ? " (down)" : "")
2889 << " to osd." << sfront
.to
2890 << " [" << osd_parentage(osdmap
, sfront
.to
) << "]"
2891 << (osdmap
.is_down(sfront
.to
) ? " (down)" : "")
2892 << " " << fixed_u_to_string(sfront
.pingtime
, 3) << " msec"
2893 << (sfront
.improving
? " possibly improving" : "");
2894 detail_front
.push_back(ss
.str());
2896 if (detail_back
.size() != 0) {
2898 ss
<< "Slow OSD heartbeats on back (longest "
2899 << fixed_u_to_string(back_sorted
.rbegin()->pingtime
, 3) << "ms)";
2900 auto& d
= checks
->add("OSD_SLOW_PING_TIME_BACK", HEALTH_WARN
, ss
.str(),
2901 back_sorted
.size());
2902 d
.detail
.swap(detail_back
);
2904 if (detail_front
.size() != 0) {
2906 ss
<< "Slow OSD heartbeats on front (longest "
2907 << fixed_u_to_string(front_sorted
.rbegin()->pingtime
, 3) << "ms)";
2908 auto& d
= checks
->add("OSD_SLOW_PING_TIME_FRONT", HEALTH_WARN
, ss
.str(),
2909 front_sorted
.size());
2910 d
.detail
.swap(detail_front
);
2915 // MANY_OBJECTS_PER_PG
2916 if (!pg_stat
.empty()) {
2917 list
<string
> pgp_detail
, many_detail
;
2918 const auto mon_pg_warn_min_objects
=
2919 cct
->_conf
.get_val
<int64_t>("mon_pg_warn_min_objects");
2920 const auto mon_pg_warn_min_pool_objects
=
2921 cct
->_conf
.get_val
<int64_t>("mon_pg_warn_min_pool_objects");
2922 const auto mon_pg_warn_max_object_skew
=
2923 cct
->_conf
.get_val
<double>("mon_pg_warn_max_object_skew");
2924 for (auto p
= pg_pool_sum
.begin();
2925 p
!= pg_pool_sum
.end();
2927 const pg_pool_t
*pi
= osdmap
.get_pg_pool(p
->first
);
2929 continue; // in case osdmap changes haven't propagated to PGMap yet
2930 const string
& name
= osdmap
.get_pool_name(p
->first
);
2931 // NOTE: we use pg_num_target and pgp_num_target for the purposes of
2932 // the warnings. If the cluster is failing to converge on the target
2933 // values that is a separate issue!
2934 if (pi
->get_pg_num_target() > pi
->get_pgp_num_target() &&
2935 !(name
.find(".DELETED") != string::npos
&&
2936 cct
->_conf
->mon_fake_pool_delete
)) {
2938 ss
<< "pool " << name
<< " pg_num "
2939 << pi
->get_pg_num_target()
2940 << " > pgp_num " << pi
->get_pgp_num_target();
2941 pgp_detail
.push_back(ss
.str());
2943 int average_objects_per_pg
= pg_sum
.stats
.sum
.num_objects
/ pg_stat
.size();
2944 if (average_objects_per_pg
> 0 &&
2945 pg_sum
.stats
.sum
.num_objects
>= mon_pg_warn_min_objects
&&
2946 p
->second
.stats
.sum
.num_objects
>= mon_pg_warn_min_pool_objects
) {
2947 int objects_per_pg
= p
->second
.stats
.sum
.num_objects
/
2948 pi
->get_pg_num_target();
2949 float ratio
= (float)objects_per_pg
/ (float)average_objects_per_pg
;
2950 if (mon_pg_warn_max_object_skew
> 0 &&
2951 ratio
> mon_pg_warn_max_object_skew
) {
2953 ss
<< "pool " << name
<< " objects per pg ("
2954 << objects_per_pg
<< ") is more than " << ratio
2955 << " times cluster average ("
2956 << average_objects_per_pg
<< ")";
2957 many_detail
.push_back(ss
.str());
2961 if (!pgp_detail
.empty()) {
2963 ss
<< pgp_detail
.size() << " pools have pg_num > pgp_num";
2964 auto& d
= checks
->add("SMALLER_PGP_NUM", HEALTH_WARN
, ss
.str(),
2966 d
.detail
.swap(pgp_detail
);
2968 if (!many_detail
.empty()) {
2970 ss
<< many_detail
.size() << " pools have many more objects per pg than"
2972 auto& d
= checks
->add("MANY_OBJECTS_PER_PG", HEALTH_WARN
, ss
.str(),
2973 many_detail
.size());
2974 d
.detail
.swap(many_detail
);
2981 float warn_threshold
= (float)g_conf().get_val
<int64_t>("mon_pool_quota_warn_threshold")/100;
2982 float crit_threshold
= (float)g_conf().get_val
<int64_t>("mon_pool_quota_crit_threshold")/100;
2983 list
<string
> full_detail
, nearfull_detail
;
2984 unsigned full_pools
= 0, nearfull_pools
= 0;
2985 for (auto it
: pools
) {
2986 auto it2
= pg_pool_sum
.find(it
.first
);
2987 if (it2
== pg_pool_sum
.end()) {
2990 const pool_stat_t
*pstat
= &it2
->second
;
2991 const object_stat_sum_t
& sum
= pstat
->stats
.sum
;
2992 const string
& pool_name
= osdmap
.get_pool_name(it
.first
);
2993 const pg_pool_t
&pool
= it
.second
;
2994 bool full
= false, nearfull
= false;
2995 if (pool
.quota_max_objects
> 0) {
2997 if ((uint64_t)sum
.num_objects
>= pool
.quota_max_objects
) {
2998 } else if (crit_threshold
> 0 &&
2999 sum
.num_objects
>= pool
.quota_max_objects
*crit_threshold
) {
3000 ss
<< "pool '" << pool_name
3001 << "' has " << sum
.num_objects
<< " objects"
3002 << " (max " << pool
.quota_max_objects
<< ")";
3003 full_detail
.push_back(ss
.str());
3005 } else if (warn_threshold
> 0 &&
3006 sum
.num_objects
>= pool
.quota_max_objects
*warn_threshold
) {
3007 ss
<< "pool '" << pool_name
3008 << "' has " << sum
.num_objects
<< " objects"
3009 << " (max " << pool
.quota_max_objects
<< ")";
3010 nearfull_detail
.push_back(ss
.str());
3014 if (pool
.quota_max_bytes
> 0) {
3016 if ((uint64_t)sum
.num_bytes
>= pool
.quota_max_bytes
) {
3017 } else if (crit_threshold
> 0 &&
3018 sum
.num_bytes
>= pool
.quota_max_bytes
*crit_threshold
) {
3019 ss
<< "pool '" << pool_name
3020 << "' has " << byte_u_t(sum
.num_bytes
)
3021 << " (max " << byte_u_t(pool
.quota_max_bytes
) << ")";
3022 full_detail
.push_back(ss
.str());
3024 } else if (warn_threshold
> 0 &&
3025 sum
.num_bytes
>= pool
.quota_max_bytes
*warn_threshold
) {
3026 ss
<< "pool '" << pool_name
3027 << "' has " << byte_u_t(sum
.num_bytes
)
3028 << " (max " << byte_u_t(pool
.quota_max_bytes
) << ")";
3029 nearfull_detail
.push_back(ss
.str());
3042 ss
<< full_pools
<< " pools full";
3043 auto& d
= checks
->add("POOL_FULL", HEALTH_ERR
, ss
.str(), full_pools
);
3044 d
.detail
.swap(full_detail
);
3046 if (nearfull_pools
) {
3048 ss
<< nearfull_pools
<< " pools nearfull";
3049 auto& d
= checks
->add("POOL_NEAR_FULL", HEALTH_WARN
, ss
.str(), nearfull_pools
);
3050 d
.detail
.swap(nearfull_detail
);
3055 if (pg_sum
.stats
.sum
.num_objects_misplaced
&&
3056 pg_sum
.stats
.sum
.num_object_copies
> 0 &&
3057 cct
->_conf
->mon_warn_on_misplaced
) {
3058 double pc
= (double)pg_sum
.stats
.sum
.num_objects_misplaced
/
3059 (double)pg_sum
.stats
.sum
.num_object_copies
* (double)100.0;
3061 snprintf(b
, sizeof(b
), "%.3lf", pc
);
3063 ss
<< pg_sum
.stats
.sum
.num_objects_misplaced
3064 << "/" << pg_sum
.stats
.sum
.num_object_copies
<< " objects misplaced ("
3066 checks
->add("OBJECT_MISPLACED", HEALTH_WARN
, ss
.str(),
3067 pg_sum
.stats
.sum
.num_objects_misplaced
);
3071 if (pg_sum
.stats
.sum
.num_objects_unfound
&&
3072 pg_sum
.stats
.sum
.num_objects
) {
3073 double pc
= (double)pg_sum
.stats
.sum
.num_objects_unfound
/
3074 (double)pg_sum
.stats
.sum
.num_objects
* (double)100.0;
3076 snprintf(b
, sizeof(b
), "%.3lf", pc
);
3078 ss
<< pg_sum
.stats
.sum
.num_objects_unfound
3079 << "/" << pg_sum
.stats
.sum
.num_objects
<< " objects unfound (" << b
<< "%)";
3080 auto& d
= checks
->add("OBJECT_UNFOUND", HEALTH_WARN
, ss
.str(),
3081 pg_sum
.stats
.sum
.num_objects_unfound
);
3083 for (auto& p
: pg_stat
) {
3084 if (p
.second
.stats
.sum
.num_objects_unfound
) {
3086 ss
<< "pg " << p
.first
3087 << " has " << p
.second
.stats
.sum
.num_objects_unfound
3088 << " unfound objects";
3089 d
.detail
.push_back(ss
.str());
3090 if (d
.detail
.size() > max
) {
3091 d
.detail
.push_back("(additional pgs left out for brevity)");
3100 // SLOW_OPS unifies them in mimic.
3101 if (osdmap
.require_osd_release
< ceph_release_t::mimic
&&
3102 cct
->_conf
->mon_osd_warn_op_age
> 0 &&
3103 !osd_sum
.op_queue_age_hist
.h
.empty() &&
3104 osd_sum
.op_queue_age_hist
.upper_bound() / 1000.0 >
3105 cct
->_conf
->mon_osd_warn_op_age
) {
3106 list
<string
> warn_detail
, error_detail
;
3107 unsigned warn
= 0, error
= 0;
3109 cct
->_conf
->mon_osd_warn_op_age
* cct
->_conf
->mon_osd_err_op_age_ratio
;
3110 const pow2_hist_t
& h
= osd_sum
.op_queue_age_hist
;
3111 for (unsigned i
= h
.h
.size() - 1; i
> 0; --i
) {
3112 float ub
= (float)(1 << i
) / 1000.0;
3113 if (ub
< cct
->_conf
->mon_osd_warn_op_age
)
3117 ss
<< h
.h
[i
] << " ops are blocked > " << ub
<< " sec";
3120 error_detail
.push_back(ss
.str());
3123 warn_detail
.push_back(ss
.str());
3128 map
<float,set
<int>> warn_osd_by_max
; // max -> osds
3129 map
<float,set
<int>> error_osd_by_max
; // max -> osds
3130 if (!warn_detail
.empty() || !error_detail
.empty()) {
3131 for (auto& p
: osd_stat
) {
3132 const pow2_hist_t
& h
= p
.second
.op_queue_age_hist
;
3133 for (unsigned i
= h
.h
.size() - 1; i
> 0; --i
) {
3134 float ub
= (float)(1 << i
) / 1000.0;
3135 if (ub
< cct
->_conf
->mon_osd_warn_op_age
)
3139 error_osd_by_max
[ub
].insert(p
.first
);
3141 warn_osd_by_max
[ub
].insert(p
.first
);
3149 if (!warn_detail
.empty()) {
3151 ss
<< warn
<< " slow requests are blocked > "
3152 << cct
->_conf
->mon_osd_warn_op_age
<< " sec";
3153 auto& d
= checks
->add("REQUEST_SLOW", HEALTH_WARN
, ss
.str(), warn
);
3154 d
.detail
.swap(warn_detail
);
3156 for (auto& p
: warn_osd_by_max
) {
3158 if (p
.second
.size() > 1) {
3159 ss
<< "osds " << p
.second
3160 << " have blocked requests > " << p
.first
<< " sec";
3162 ss
<< "osd." << *p
.second
.begin()
3163 << " has blocked requests > " << p
.first
<< " sec";
3165 d
.detail
.push_back(ss
.str());
3171 if (!error_detail
.empty()) {
3173 ss
<< error
<< " stuck requests are blocked > "
3174 << err_age
<< " sec";
3175 auto& d
= checks
->add("REQUEST_STUCK", HEALTH_ERR
, ss
.str(), error
);
3176 d
.detail
.swap(error_detail
);
3178 for (auto& p
: error_osd_by_max
) {
3180 if (p
.second
.size() > 1) {
3181 ss
<< "osds " << p
.second
3182 << " have stuck requests > " << p
.first
<< " sec";
3184 ss
<< "osd." << *p
.second
.begin()
3185 << " has stuck requests > " << p
.first
<< " sec";
3187 d
.detail
.push_back(ss
.str());
3195 // OBJECT_STORE_WARN
3196 if (osd_sum
.os_alerts
.size()) {
3197 map
<string
, pair
<size_t, list
<string
>>> os_alerts_sum
;
3199 for (auto& a
: osd_sum
.os_alerts
) {
3201 string s0
= " osd.";
3202 s0
+= stringify(a
.first
);
3203 for (auto& aa
: a
.second
) {
3207 auto it
= os_alerts_sum
.find(aa
.first
);
3208 if (it
== os_alerts_sum
.end()) {
3211 os_alerts_sum
.emplace(aa
.first
, std::make_pair(1, d
));
3213 auto& p
= it
->second
;
3215 p
.second
.emplace_back(s
);
3223 for (auto& asum
: os_alerts_sum
) {
3224 string summary
= stringify(asum
.second
.first
) + " OSD(s)";
3225 if (asum
.first
== "BLUEFS_SPILLOVER") {
3226 summary
+= " experiencing BlueFS spillover";
3227 } else if (asum
.first
== "BLUESTORE_NO_COMPRESSION") {
3228 summary
+= " have broken BlueStore compression";
3229 } else if (asum
.first
== "BLUESTORE_LEGACY_STATFS") {
3230 summary
+= " reporting legacy (not per-pool) BlueStore stats";
3231 } else if (asum
.first
== "BLUESTORE_DISK_SIZE_MISMATCH") {
3232 summary
+= " have dangerous mismatch between BlueStore block device and free list sizes";
3233 } else if (asum
.first
== "BLUESTORE_NO_PER_POOL_OMAP") {
3234 summary
+= " reporting legacy (not per-pool) BlueStore omap usage stats";
3236 auto& d
= checks
->add(asum
.first
, HEALTH_WARN
, summary
, asum
.second
.first
);
3237 for (auto& s
: asum
.second
.second
) {
3238 d
.detail
.push_back(s
);
3243 // PG_NOT_DEEP_SCRUBBED
3244 if (cct
->_conf
->mon_warn_pg_not_scrubbed_ratio
||
3245 cct
->_conf
->mon_warn_pg_not_deep_scrubbed_ratio
) {
3246 list
<string
> detail
, deep_detail
;
3247 int detail_max
= max
, deep_detail_max
= max
;
3248 int detail_more
= 0, deep_detail_more
= 0;
3249 int detail_total
= 0, deep_detail_total
= 0;
3250 for (auto& p
: pg_stat
) {
3251 int64_t pnum
= p
.first
.pool();
3252 auto pool
= osdmap
.get_pg_pool(pnum
);
3255 if (cct
->_conf
->mon_warn_pg_not_scrubbed_ratio
) {
3256 double scrub_max_interval
= 0;
3257 pool
->opts
.get(pool_opts_t::SCRUB_MAX_INTERVAL
, &scrub_max_interval
);
3258 if (scrub_max_interval
<= 0) {
3259 scrub_max_interval
= cct
->_conf
->osd_scrub_max_interval
;
3261 const double age
= (cct
->_conf
->mon_warn_pg_not_scrubbed_ratio
* scrub_max_interval
) +
3263 utime_t cutoff
= now
;
3265 if (p
.second
.last_scrub_stamp
< cutoff
) {
3266 if (detail_max
> 0) {
3268 ss
<< "pg " << p
.first
<< " not scrubbed since "
3269 << p
.second
.last_scrub_stamp
;
3270 detail
.push_back(ss
.str());
3278 if (cct
->_conf
->mon_warn_pg_not_deep_scrubbed_ratio
) {
3279 double deep_scrub_interval
= 0;
3280 pool
->opts
.get(pool_opts_t::DEEP_SCRUB_INTERVAL
, &deep_scrub_interval
);
3281 if (deep_scrub_interval
<= 0) {
3282 deep_scrub_interval
= cct
->_conf
->osd_deep_scrub_interval
;
3284 double deep_age
= (cct
->_conf
->mon_warn_pg_not_deep_scrubbed_ratio
* deep_scrub_interval
) +
3285 deep_scrub_interval
;
3286 utime_t deep_cutoff
= now
;
3287 deep_cutoff
-= deep_age
;
3288 if (p
.second
.last_deep_scrub_stamp
< deep_cutoff
) {
3289 if (deep_detail_max
> 0) {
3291 ss
<< "pg " << p
.first
<< " not deep-scrubbed since "
3292 << p
.second
.last_deep_scrub_stamp
;
3293 deep_detail
.push_back(ss
.str());
3298 ++deep_detail_total
;
3304 ss
<< detail_total
<< " pgs not scrubbed in time";
3305 auto& d
= checks
->add("PG_NOT_SCRUBBED", HEALTH_WARN
, ss
.str(), detail_total
);
3307 if (!detail
.empty()) {
3308 d
.detail
.swap(detail
);
3312 ss
<< detail_more
<< " more pgs... ";
3313 d
.detail
.push_back(ss
.str());
3317 if (deep_detail_total
) {
3319 ss
<< deep_detail_total
<< " pgs not deep-scrubbed in time";
3320 auto& d
= checks
->add("PG_NOT_DEEP_SCRUBBED", HEALTH_WARN
, ss
.str(),
3323 if (!deep_detail
.empty()) {
3324 d
.detail
.swap(deep_detail
);
3326 if (deep_detail_more
) {
3328 ss
<< deep_detail_more
<< " more pgs... ";
3329 d
.detail
.push_back(ss
.str());
3336 if (g_conf().get_val
<bool>("mon_warn_on_pool_no_app")) {
3337 list
<string
> detail
;
3338 for (auto &it
: pools
) {
3339 const pg_pool_t
&pool
= it
.second
;
3340 const string
& pool_name
= osdmap
.get_pool_name(it
.first
);
3341 auto it2
= pg_pool_sum
.find(it
.first
);
3342 if (it2
== pg_pool_sum
.end()) {
3345 const pool_stat_t
*pstat
= &it2
->second
;
3346 if (pstat
== nullptr) {
3349 const object_stat_sum_t
& sum
= pstat
->stats
.sum
;
3350 // application metadata is not encoded until luminous is minimum
3352 if (sum
.num_objects
> 0 && pool
.application_metadata
.empty() &&
3355 ss
<< "application not enabled on pool '" << pool_name
<< "'";
3356 detail
.push_back(ss
.str());
3359 if (!detail
.empty()) {
3361 ss
<< detail
.size() << " pool(s) do not have an application enabled";
3362 auto& d
= checks
->add("POOL_APP_NOT_ENABLED", HEALTH_WARN
, ss
.str(),
3365 tip
<< "use 'ceph osd pool application enable <pool-name> "
3366 << "<app-name>', where <app-name> is 'cephfs', 'rbd', 'rgw', "
3367 << "or freeform for custom applications.";
3368 detail
.push_back(tip
.str());
3369 d
.detail
.swap(detail
);
3373 // PG_SLOW_SNAP_TRIMMING
3374 if (!pg_stat
.empty() && cct
->_conf
->mon_osd_snap_trim_queue_warn_on
> 0) {
3375 uint32_t snapthreshold
= cct
->_conf
->mon_osd_snap_trim_queue_warn_on
;
3376 uint64_t snaptrimq_exceeded
= 0;
3377 uint32_t longest_queue
= 0;
3378 const pg_t
* longest_q_pg
= nullptr;
3379 list
<string
> detail
;
3381 for (auto& i
: pg_stat
) {
3382 uint32_t current_len
= i
.second
.snaptrimq_len
;
3383 if (current_len
>= snapthreshold
) {
3384 snaptrimq_exceeded
++;
3385 if (longest_queue
<= current_len
) {
3386 longest_q_pg
= &i
.first
;
3387 longest_queue
= current_len
;
3389 if (detail
.size() < max
- 1) {
3391 ss
<< "snap trim queue for pg " << i
.first
<< " at " << current_len
;
3392 detail
.push_back(ss
.str());
3395 if (detail
.size() < max
) {
3396 detail
.push_back("...more pgs affected");
3402 if (snaptrimq_exceeded
) {
3405 ss
<< "longest queue on pg " << *longest_q_pg
<< " at " << longest_queue
;
3406 detail
.push_back(ss
.str());
3410 ss
<< "snap trim queue for " << snaptrimq_exceeded
<< " pg(s) >= " << snapthreshold
<< " (mon_osd_snap_trim_queue_warn_on)";
3411 auto& d
= checks
->add("PG_SLOW_SNAP_TRIMMING", HEALTH_WARN
, ss
.str(),
3412 snaptrimq_exceeded
);
3413 detail
.push_back("try decreasing \"osd snap trim sleep\" and/or increasing \"osd pg max concurrent snap trims\".");
3414 d
.detail
.swap(detail
);
3419 void PGMap::print_summary(ceph::Formatter
*f
, ostream
*out
) const
3422 f
->open_array_section("pgs_by_pool_state");
3423 for (auto& i
: num_pg_by_pool_state
) {
3424 f
->open_object_section("per_pool_pgs_by_state");
3425 f
->dump_int("pool_id", i
.first
);
3426 f
->open_array_section("pg_state_counts");
3427 for (auto& j
: i
.second
) {
3428 f
->open_object_section("pg_state_count");
3429 f
->dump_string("state_name", pg_state_string(j
.first
));
3430 f
->dump_int("count", j
.second
);
3438 PGMapDigest::print_summary(f
, out
);
3441 int process_pg_map_command(
3442 const string
& orig_prefix
,
3443 const cmdmap_t
& orig_cmdmap
,
3444 const PGMap
& pg_map
,
3445 const OSDMap
& osdmap
,
3450 string prefix
= orig_prefix
;
3451 auto cmdmap
= orig_cmdmap
;
3453 string omap_stats_note
=
3454 "\n* NOTE: Omap statistics are gathered during deep scrub and "
3455 "may be inaccurate soon afterwards depending on utilization. See "
3456 "http://docs.ceph.com/docs/master/dev/placement-group/#omap-statistics "
3457 "for further details.\n";
3458 bool omap_stats_note_required
= false;
3460 // perhaps these would be better in the parsing, but it's weird
3461 bool primary
= false;
3462 if (prefix
== "pg dump_json") {
3464 v
.push_back(string("all"));
3465 cmd_putval(g_ceph_context
, cmdmap
, "format", string("json"));
3466 cmd_putval(g_ceph_context
, cmdmap
, "dumpcontents", v
);
3468 } else if (prefix
== "pg dump_pools_json") {
3470 v
.push_back(string("pools"));
3471 cmd_putval(g_ceph_context
, cmdmap
, "format", string("json"));
3472 cmd_putval(g_ceph_context
, cmdmap
, "dumpcontents", v
);
3474 } else if (prefix
== "pg ls-by-primary") {
3477 } else if (prefix
== "pg ls-by-osd") {
3479 } else if (prefix
== "pg ls-by-pool") {
3482 cmd_getval(cmdmap
, "poolstr", poolstr
);
3483 int64_t pool
= osdmap
.lookup_pg_pool_name(poolstr
.c_str());
3485 *ss
<< "pool " << poolstr
<< " does not exist";
3488 cmd_putval(g_ceph_context
, cmdmap
, "pool", pool
);
3492 if (prefix
== "pg stat") {
3494 f
->open_object_section("pg_summary");
3495 pg_map
.print_oneline_summary(f
, NULL
);
3505 if (prefix
== "pg getmap") {
3506 pg_map
.encode(*odata
);
3507 *ss
<< "got pgmap version " << pg_map
.version
;
3511 if (prefix
== "pg dump") {
3513 vector
<string
> dumpcontents
;
3515 if (cmd_getval(cmdmap
, "dumpcontents", dumpcontents
)) {
3516 copy(dumpcontents
.begin(), dumpcontents
.end(),
3517 inserter(what
, what
.end()));
3522 if (what
.count("all")) {
3523 f
->open_object_section("pg_map");
3526 } else if (what
.count("summary") || what
.count("sum")) {
3527 f
->open_object_section("pg_map");
3528 pg_map
.dump_basic(f
);
3531 if (what
.count("pools")) {
3532 pg_map
.dump_pool_stats(f
);
3534 if (what
.count("osds")) {
3535 pg_map
.dump_osd_stats(f
);
3537 if (what
.count("pgs")) {
3538 pg_map
.dump_pg_stats(f
, false);
3540 if (what
.count("pgs_brief")) {
3541 pg_map
.dump_pg_stats(f
, true);
3543 if (what
.count("delta")) {
3544 f
->open_object_section("delta");
3545 pg_map
.dump_delta(f
);
3551 if (what
.count("all")) {
3553 omap_stats_note_required
= true;
3554 } else if (what
.count("summary") || what
.count("sum")) {
3555 pg_map
.dump_basic(ds
);
3556 pg_map
.dump_pg_sum_stats(ds
, true);
3557 pg_map
.dump_osd_sum_stats(ds
);
3558 omap_stats_note_required
= true;
3560 if (what
.count("pgs_brief")) {
3561 pg_map
.dump_pg_stats(ds
, true);
3564 if (what
.count("pgs")) {
3565 pg_map
.dump_pg_stats(ds
, false);
3567 omap_stats_note_required
= true;
3569 if (what
.count("pools")) {
3570 pg_map
.dump_pool_stats(ds
, header
);
3571 omap_stats_note_required
= true;
3573 if (what
.count("osds")) {
3574 pg_map
.dump_osd_stats(ds
);
3578 if (omap_stats_note_required
) {
3579 odata
->append(omap_stats_note
);
3582 *ss
<< "dumped " << what
;
3586 if (prefix
== "pg ls") {
3589 vector
<string
>states
;
3591 cmd_getval(cmdmap
, "pool", pool
);
3592 cmd_getval(cmdmap
, "osd", osd
);
3593 cmd_getval(cmdmap
, "states", states
);
3594 if (pool
>= 0 && !osdmap
.have_pg_pool(pool
)) {
3595 *ss
<< "pool " << pool
<< " does not exist";
3598 if (osd
>= 0 && !osdmap
.is_up(osd
)) {
3599 *ss
<< "osd " << osd
<< " is not up";
3603 states
.push_back("all");
3607 while (!states
.empty()) {
3608 string state_str
= states
.back();
3610 if (state_str
== "all") {
3614 auto filter
= pg_string_state(state_str
);
3616 *ss
<< "'" << state_str
<< "' is not a valid pg state,"
3617 << " available choices: " << pg_state_string(0xFFFFFFFF);
3626 pg_map
.get_filtered_pg_stats(state
, pool
, osd
, primary
, pgs
);
3628 if (f
&& !pgs
.empty()) {
3629 pg_map
.dump_filtered_pg_stats(f
, pgs
);
3631 } else if (!pgs
.empty()) {
3632 pg_map
.dump_filtered_pg_stats(ds
, pgs
);
3634 odata
->append(omap_stats_note
);
3639 if (prefix
== "pg dump_stuck") {
3640 vector
<string
> stuckop_vec
;
3641 cmd_getval(cmdmap
, "stuckops", stuckop_vec
);
3642 if (stuckop_vec
.empty())
3643 stuckop_vec
.push_back("unclean");
3645 cmd_getval(cmdmap
, "threshold", threshold
,
3646 g_conf().get_val
<int64_t>("mon_pg_stuck_threshold"));
3648 if (pg_map
.dump_stuck_pg_stats(ds
, f
, (int)threshold
, stuckop_vec
) < 0) {
3657 if (prefix
== "pg debug") {
3659 cmd_getval(cmdmap
, "debugop", debugop
,
3660 string("unfound_objects_exist"));
3661 if (debugop
== "unfound_objects_exist") {
3662 bool unfound_objects_exist
= false;
3663 for (const auto& p
: pg_map
.pg_stat
) {
3664 if (p
.second
.stats
.sum
.num_objects_unfound
> 0) {
3665 unfound_objects_exist
= true;
3669 if (unfound_objects_exist
)
3676 if (debugop
== "degraded_pgs_exist") {
3677 bool degraded_pgs_exist
= false;
3678 for (const auto& p
: pg_map
.pg_stat
) {
3679 if (p
.second
.stats
.sum
.num_objects_degraded
> 0) {
3680 degraded_pgs_exist
= true;
3684 if (degraded_pgs_exist
)
3693 if (prefix
== "osd perf") {
3695 f
->open_object_section("osdstats");
3696 pg_map
.dump_osd_perf_stats(f
);
3700 pg_map
.print_osd_perf_stats(&ds
);
3706 if (prefix
== "osd blocked-by") {
3708 f
->open_object_section("osd_blocked_by");
3709 pg_map
.dump_osd_blocked_by_stats(f
);
3713 pg_map
.print_osd_blocked_by_stats(&ds
);
3722 void PGMapUpdater::check_osd_map(
3724 const OSDMap
& osdmap
,
3726 PGMap::Incremental
*pending_inc
)
3728 for (auto& p
: pgmap
.osd_stat
) {
3729 if (!osdmap
.exists(p
.first
)) {
3731 pending_inc
->rm_stat(p
.first
);
3732 } else if (osdmap
.is_out(p
.first
)) {
3734 if (p
.second
.statfs
.total
!= 0) {
3735 pending_inc
->stat_osd_out(p
.first
);
3737 } else if (!osdmap
.is_up(p
.first
)) {
3738 // zero the op_queue_age_hist
3739 if (!p
.second
.op_queue_age_hist
.empty()) {
3740 pending_inc
->stat_osd_down_up(p
.first
, pgmap
);
3745 // deleted pgs (pools)?
3746 for (auto& p
: pgmap
.pg_pool_sum
) {
3747 if (!osdmap
.have_pg_pool(p
.first
)) {
3748 ldout(cct
, 10) << __func__
<< " pool " << p
.first
<< " gone, removing pgs"
3750 for (auto& q
: pgmap
.pg_stat
) {
3751 if (q
.first
.pool() == p
.first
) {
3752 pending_inc
->pg_remove
.insert(q
.first
);
3755 auto q
= pending_inc
->pg_stat_updates
.begin();
3756 while (q
!= pending_inc
->pg_stat_updates
.end()) {
3757 if (q
->first
.pool() == p
.first
) {
3758 q
= pending_inc
->pg_stat_updates
.erase(q
);
3766 // new (split or new pool) or merged pgs?
3767 map
<int64_t,unsigned> new_pg_num
;
3768 for (auto& p
: osdmap
.get_pools()) {
3769 int64_t poolid
= p
.first
;
3770 const pg_pool_t
& pi
= p
.second
;
3771 auto q
= pgmap
.num_pg_by_pool
.find(poolid
);
3772 unsigned my_pg_num
= 0;
3773 if (q
!= pgmap
.num_pg_by_pool
.end())
3774 my_pg_num
= q
->second
;
3775 unsigned pg_num
= pi
.get_pg_num();
3776 new_pg_num
[poolid
] = pg_num
;
3777 if (my_pg_num
< pg_num
) {
3778 ldout(cct
,10) << __func__
<< " pool " << poolid
<< " pg_num " << pg_num
3779 << " > my pg_num " << my_pg_num
<< dendl
;
3780 for (unsigned ps
= my_pg_num
; ps
< pg_num
; ++ps
) {
3781 pg_t
pgid(ps
, poolid
);
3782 if (pending_inc
->pg_stat_updates
.count(pgid
) == 0) {
3783 ldout(cct
,20) << __func__
<< " adding " << pgid
<< dendl
;
3784 pg_stat_t
&stats
= pending_inc
->pg_stat_updates
[pgid
];
3785 stats
.last_fresh
= osdmap
.get_modified();
3786 stats
.last_active
= osdmap
.get_modified();
3787 stats
.last_change
= osdmap
.get_modified();
3788 stats
.last_peered
= osdmap
.get_modified();
3789 stats
.last_clean
= osdmap
.get_modified();
3790 stats
.last_unstale
= osdmap
.get_modified();
3791 stats
.last_undegraded
= osdmap
.get_modified();
3792 stats
.last_fullsized
= osdmap
.get_modified();
3793 stats
.last_scrub_stamp
= osdmap
.get_modified();
3794 stats
.last_deep_scrub_stamp
= osdmap
.get_modified();
3795 stats
.last_clean_scrub_stamp
= osdmap
.get_modified();
3798 } else if (my_pg_num
> pg_num
) {
3799 ldout(cct
,10) << __func__
<< " pool " << poolid
<< " pg_num " << pg_num
3800 << " < my pg_num " << my_pg_num
<< dendl
;
3801 for (unsigned i
= pg_num
; i
< my_pg_num
; ++i
) {
3802 pg_t
pgid(i
, poolid
);
3803 ldout(cct
,20) << __func__
<< " removing merged " << pgid
<< dendl
;
3804 if (pgmap
.pg_stat
.count(pgid
)) {
3805 pending_inc
->pg_remove
.insert(pgid
);
3807 pending_inc
->pg_stat_updates
.erase(pgid
);
3811 auto i
= pending_inc
->pg_stat_updates
.begin();
3812 while (i
!= pending_inc
->pg_stat_updates
.end()) {
3813 auto j
= new_pg_num
.find(i
->first
.pool());
3814 if (j
== new_pg_num
.end() ||
3815 i
->first
.ps() >= j
->second
) {
3816 ldout(cct
,20) << __func__
<< " removing pending update to old "
3817 << i
->first
<< dendl
;
3818 i
= pending_inc
->pg_stat_updates
.erase(i
);
3825 static void _try_mark_pg_stale(
3826 const OSDMap
& osdmap
,
3828 const pg_stat_t
& cur
,
3829 PGMap::Incremental
*pending_inc
)
3831 if ((cur
.state
& PG_STATE_STALE
) == 0 &&
3832 cur
.acting_primary
!= -1 &&
3833 osdmap
.is_down(cur
.acting_primary
)) {
3835 auto q
= pending_inc
->pg_stat_updates
.find(pgid
);
3836 if (q
!= pending_inc
->pg_stat_updates
.end()) {
3837 if ((q
->second
.acting_primary
== cur
.acting_primary
) ||
3838 ((q
->second
.state
& PG_STATE_STALE
) == 0 &&
3839 q
->second
.acting_primary
!= -1 &&
3840 osdmap
.is_down(q
->second
.acting_primary
))) {
3841 newstat
= &q
->second
;
3843 // pending update is no longer down or already stale
3847 newstat
= &pending_inc
->pg_stat_updates
[pgid
];
3850 dout(10) << __func__
<< " marking pg " << pgid
3851 << " stale (acting_primary " << newstat
->acting_primary
3853 newstat
->state
|= PG_STATE_STALE
;
3854 newstat
->last_unstale
= ceph_clock_now();
3858 void PGMapUpdater::check_down_pgs(
3859 const OSDMap
&osdmap
,
3860 const PGMap
&pg_map
,
3862 const set
<int>& need_check_down_pg_osds
,
3863 PGMap::Incremental
*pending_inc
)
3865 // if a large number of osds changed state, just iterate over the whole
3867 if (need_check_down_pg_osds
.size() > (unsigned)osdmap
.get_num_osds() *
3868 g_conf().get_val
<double>("mon_pg_check_down_all_threshold")) {
3873 for (const auto& p
: pg_map
.pg_stat
) {
3874 _try_mark_pg_stale(osdmap
, p
.first
, p
.second
, pending_inc
);
3877 for (auto osd
: need_check_down_pg_osds
) {
3878 if (osdmap
.is_down(osd
)) {
3879 auto p
= pg_map
.pg_by_osd
.find(osd
);
3880 if (p
== pg_map
.pg_by_osd
.end()) {
3883 for (auto pgid
: p
->second
) {
3884 const pg_stat_t
&stat
= pg_map
.pg_stat
.at(pgid
);
3885 ceph_assert(stat
.acting_primary
== osd
);
3886 _try_mark_pg_stale(osdmap
, pgid
, stat
, pending_inc
);
3893 int reweight::by_utilization(
3894 const OSDMap
&osdmap
,
3899 bool by_pg
, const set
<int64_t> *pools
,
3901 mempool::osdmap::map
<int32_t, uint32_t>* new_weights
,
3902 std::stringstream
*ss
,
3903 std::string
*out_str
,
3907 *ss
<< "You must give a percentage higher than 100. "
3908 "The reweighting threshold will be calculated as <average-utilization> "
3909 "times <input-percentage>. For example, an argument of 200 would "
3910 "reweight OSDs which are twice as utilized as the average OSD.\n";
3914 vector
<int> pgs_by_osd(osdmap
.get_max_osd());
3916 // Avoid putting a small number (or 0) in the denominator when calculating
3918 double average_util
;
3921 double weight_sum
= 0.0; // sum up the crush weights
3922 unsigned num_pg_copies
= 0;
3924 for (const auto& pg
: pgm
.pg_stat
) {
3925 if (pools
&& pools
->count(pg
.first
.pool()) == 0)
3927 for (const auto acting
: pg
.second
.acting
) {
3928 if (!osdmap
.exists(acting
)) {
3931 if (acting
>= (int)pgs_by_osd
.size())
3932 pgs_by_osd
.resize(acting
);
3933 if (pgs_by_osd
[acting
] == 0) {
3934 if (osdmap
.crush
->get_item_weightf(acting
) <= 0) {
3935 //skip if we currently can not identify item
3938 weight_sum
+= osdmap
.crush
->get_item_weightf(acting
);
3941 ++pgs_by_osd
[acting
];
3946 if (!num_osds
|| (num_pg_copies
/ num_osds
< g_conf()->mon_reweight_min_pgs_per_osd
)) {
3947 *ss
<< "Refusing to reweight: we only have " << num_pg_copies
3948 << " PGs across " << num_osds
<< " osds!\n";
3952 average_util
= (double)num_pg_copies
/ weight_sum
;
3954 // by osd utilization
3955 int num_osd
= std::max
<size_t>(1, pgm
.osd_stat
.size());
3956 if ((uint64_t)pgm
.osd_sum
.statfs
.total
/ num_osd
3957 < g_conf()->mon_reweight_min_bytes_per_osd
) {
3958 *ss
<< "Refusing to reweight: we only have " << pgm
.osd_sum
.statfs
.kb()
3959 << " kb across all osds!\n";
3962 if ((uint64_t)pgm
.osd_sum
.statfs
.get_used_raw() / num_osd
3963 < g_conf()->mon_reweight_min_bytes_per_osd
) {
3964 *ss
<< "Refusing to reweight: we only have "
3965 << pgm
.osd_sum
.statfs
.kb_used_raw()
3966 << " kb used across all osds!\n";
3970 average_util
= (double)pgm
.osd_sum
.statfs
.get_used_raw() /
3971 (double)pgm
.osd_sum
.statfs
.total
;
3974 // adjust down only if we are above the threshold
3975 const double overload_util
= average_util
* (double)oload
/ 100.0;
3977 // but aggressively adjust weights up whenever possible.
3978 const double underload_util
= average_util
;
3980 const unsigned max_change
= (unsigned)(max_changef
* (double)0x10000);
3984 f
->open_object_section("reweight_by_utilization");
3985 f
->dump_int("overload_min", oload
);
3986 f
->dump_float("max_change", max_changef
);
3987 f
->dump_int("max_change_osds", max_osds
);
3988 f
->dump_float("average_utilization", average_util
);
3989 f
->dump_float("overload_utilization", overload_util
);
3991 oss
<< "oload " << oload
<< "\n";
3992 oss
<< "max_change " << max_changef
<< "\n";
3993 oss
<< "max_change_osds " << max_osds
<< "\n";
3995 oss
<< "average_utilization " << std::fixed
<< average_util
<< "\n";
3996 oss
<< "overload_utilization " << overload_util
<< "\n";
3998 int num_changed
= 0;
4000 // precompute util for each OSD
4001 std::vector
<std::pair
<int, float> > util_by_osd
;
4002 for (const auto& p
: pgm
.osd_stat
) {
4003 std::pair
<int, float> osd_util
;
4004 osd_util
.first
= p
.first
;
4006 if (p
.first
>= (int)pgs_by_osd
.size() ||
4007 pgs_by_osd
[p
.first
] == 0) {
4008 // skip if this OSD does not contain any pg
4009 // belonging to the specified pool(s).
4013 if (osdmap
.crush
->get_item_weightf(p
.first
) <= 0) {
4014 // skip if we are unable to locate item.
4019 pgs_by_osd
[p
.first
] / osdmap
.crush
->get_item_weightf(p
.first
);
4022 (double)p
.second
.statfs
.get_used_raw() / (double)p
.second
.statfs
.total
;
4024 util_by_osd
.push_back(osd_util
);
4027 // sort by absolute deviation from the mean utilization,
4028 // in descending order.
4029 std::sort(util_by_osd
.begin(), util_by_osd
.end(),
4030 [average_util
](std::pair
<int, float> l
, std::pair
<int, float> r
) {
4031 return abs(l
.second
- average_util
) > abs(r
.second
- average_util
);
4036 f
->open_array_section("reweights");
4038 for (const auto& p
: util_by_osd
) {
4039 unsigned weight
= osdmap
.get_weight(p
.first
);
4041 // skip if OSD is currently out
4044 float util
= p
.second
;
4046 if (util
>= overload_util
) {
4047 // Assign a lower weight to overloaded OSDs. The current weight
4048 // is a factor to take into account the original weights,
4049 // to represent e.g. differing storage capacities
4050 unsigned new_weight
= (unsigned)((average_util
/ util
) * (float)weight
);
4051 if (weight
> max_change
)
4052 new_weight
= std::max(new_weight
, weight
- max_change
);
4053 new_weights
->insert({p
.first
, new_weight
});
4055 f
->open_object_section("osd");
4056 f
->dump_int("osd", p
.first
);
4057 f
->dump_float("weight", (float)weight
/ (float)0x10000);
4058 f
->dump_float("new_weight", (float)new_weight
/ (float)0x10000);
4061 oss
<< "osd." << p
.first
<< " weight "
4062 << (float)weight
/ (float)0x10000 << " -> "
4063 << (float)new_weight
/ (float)0x10000 << "\n";
4065 if (++num_changed
>= max_osds
)
4068 if (!no_increasing
&& util
<= underload_util
) {
4069 // assign a higher weight.. if we can.
4070 unsigned new_weight
= (unsigned)((average_util
/ util
) * (float)weight
);
4071 new_weight
= std::min(new_weight
, weight
+ max_change
);
4072 if (new_weight
> 0x10000)
4073 new_weight
= 0x10000;
4074 if (new_weight
> weight
) {
4075 new_weights
->insert({p
.first
, new_weight
});
4076 oss
<< "osd." << p
.first
<< " weight "
4077 << (float)weight
/ (float)0x10000 << " -> "
4078 << (float)new_weight
/ (float)0x10000 << "\n";
4079 if (++num_changed
>= max_osds
)
4089 newmap
.deepish_copy_from(osdmap
);
4090 OSDMap::Incremental newinc
;
4091 newinc
.fsid
= newmap
.get_fsid();
4092 newinc
.epoch
= newmap
.get_epoch() + 1;
4093 newinc
.new_weight
= *new_weights
;
4094 newmap
.apply_incremental(newinc
);
4096 osdmap
.summarize_mapping_stats(&newmap
, pools
, out_str
, f
);
4102 *out_str
+= oss
.str();