1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
2 // vim: ts=8 sw=2 smarttab
4 * Ceph - scalable distributed file system
6 * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
7 * Copyright (C) 2013,2014 Cloudwatt <libre.licensing@cloudwatt.com>
9 * Author: Loic Dachary <loic@dachary.org>
11 * This is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public
13 * License version 2.1, as published by the Free Software
14 * Foundation. See file COPYING.
18 #include <boost/algorithm/string.hpp>
22 #include "common/config.h"
23 #include "common/Formatter.h"
24 #include "common/TextTable.h"
25 #include "include/ceph_features.h"
26 #include "include/str_map.h"
28 #include "common/code_environment.h"
29 #include "mon/health_check.h"
31 #include "crush/CrushTreeDumper.h"
32 #include "common/Clock.h"
33 #include "mon/PGStatService.h"
35 #define dout_subsys ceph_subsys_osd
37 MEMPOOL_DEFINE_OBJECT_FACTORY(OSDMap
, osdmap
, osdmap
);
38 MEMPOOL_DEFINE_OBJECT_FACTORY(OSDMap::Incremental
, osdmap_inc
, osdmap
);
41 // ----------------------------------
44 void osd_info_t::dump(Formatter
*f
) const
46 f
->dump_int("last_clean_begin", last_clean_begin
);
47 f
->dump_int("last_clean_end", last_clean_end
);
48 f
->dump_int("up_from", up_from
);
49 f
->dump_int("up_thru", up_thru
);
50 f
->dump_int("down_at", down_at
);
51 f
->dump_int("lost_at", lost_at
);
54 void osd_info_t::encode(bufferlist
& bl
) const
57 ::encode(struct_v
, bl
);
58 ::encode(last_clean_begin
, bl
);
59 ::encode(last_clean_end
, bl
);
60 ::encode(up_from
, bl
);
61 ::encode(up_thru
, bl
);
62 ::encode(down_at
, bl
);
63 ::encode(lost_at
, bl
);
66 void osd_info_t::decode(bufferlist::iterator
& bl
)
69 ::decode(struct_v
, bl
);
70 ::decode(last_clean_begin
, bl
);
71 ::decode(last_clean_end
, bl
);
72 ::decode(up_from
, bl
);
73 ::decode(up_thru
, bl
);
74 ::decode(down_at
, bl
);
75 ::decode(lost_at
, bl
);
78 void osd_info_t::generate_test_instances(list
<osd_info_t
*>& o
)
80 o
.push_back(new osd_info_t
);
81 o
.push_back(new osd_info_t
);
82 o
.back()->last_clean_begin
= 1;
83 o
.back()->last_clean_end
= 2;
84 o
.back()->up_from
= 30;
85 o
.back()->up_thru
= 40;
86 o
.back()->down_at
= 5;
87 o
.back()->lost_at
= 6;
90 ostream
& operator<<(ostream
& out
, const osd_info_t
& info
)
92 out
<< "up_from " << info
.up_from
93 << " up_thru " << info
.up_thru
94 << " down_at " << info
.down_at
95 << " last_clean_interval [" << info
.last_clean_begin
<< "," << info
.last_clean_end
<< ")";
97 out
<< " lost_at " << info
.lost_at
;
101 // ----------------------------------
104 void osd_xinfo_t::dump(Formatter
*f
) const
106 f
->dump_stream("down_stamp") << down_stamp
;
107 f
->dump_float("laggy_probability", laggy_probability
);
108 f
->dump_int("laggy_interval", laggy_interval
);
109 f
->dump_int("features", features
);
110 f
->dump_unsigned("old_weight", old_weight
);
113 void osd_xinfo_t::encode(bufferlist
& bl
) const
115 ENCODE_START(3, 1, bl
);
116 ::encode(down_stamp
, bl
);
117 __u32 lp
= laggy_probability
* 0xfffffffful
;
119 ::encode(laggy_interval
, bl
);
120 ::encode(features
, bl
);
121 ::encode(old_weight
, bl
);
125 void osd_xinfo_t::decode(bufferlist::iterator
& bl
)
128 ::decode(down_stamp
, bl
);
131 laggy_probability
= (float)lp
/ (float)0xffffffff;
132 ::decode(laggy_interval
, bl
);
134 ::decode(features
, bl
);
138 ::decode(old_weight
, bl
);
144 void osd_xinfo_t::generate_test_instances(list
<osd_xinfo_t
*>& o
)
146 o
.push_back(new osd_xinfo_t
);
147 o
.push_back(new osd_xinfo_t
);
148 o
.back()->down_stamp
= utime_t(2, 3);
149 o
.back()->laggy_probability
= .123;
150 o
.back()->laggy_interval
= 123456;
151 o
.back()->old_weight
= 0x7fff;
154 ostream
& operator<<(ostream
& out
, const osd_xinfo_t
& xi
)
156 return out
<< "down_stamp " << xi
.down_stamp
157 << " laggy_probability " << xi
.laggy_probability
158 << " laggy_interval " << xi
.laggy_interval
159 << " old_weight " << xi
.old_weight
;
162 // ----------------------------------
163 // OSDMap::Incremental
165 int OSDMap::Incremental::get_net_marked_out(const OSDMap
*previous
) const
168 for (auto &weight
: new_weight
) {
169 if (weight
.second
== CEPH_OSD_OUT
&& !previous
->is_out(weight
.first
))
171 else if (weight
.second
!= CEPH_OSD_OUT
&& previous
->is_out(weight
.first
))
177 int OSDMap::Incremental::get_net_marked_down(const OSDMap
*previous
) const
180 for (auto &state
: new_state
) { //
181 if (state
.second
& CEPH_OSD_UP
) {
182 if (previous
->is_up(state
.first
))
191 int OSDMap::Incremental::identify_osd(uuid_d u
) const
193 for (auto &uuid
: new_uuid
)
194 if (uuid
.second
== u
)
199 int OSDMap::Incremental::propagate_snaps_to_tiers(CephContext
*cct
,
200 const OSDMap
& osdmap
)
202 assert(epoch
== osdmap
.get_epoch() + 1);
204 for (auto &new_pool
: new_pools
) {
205 if (!new_pool
.second
.tiers
.empty()) {
206 pg_pool_t
& base
= new_pool
.second
;
208 for (const auto &tier_pool
: base
.tiers
) {
209 const auto &r
= new_pools
.find(tier_pool
);
211 if (r
== new_pools
.end()) {
212 const pg_pool_t
*orig
= osdmap
.get_pg_pool(tier_pool
);
214 lderr(cct
) << __func__
<< " no pool " << tier_pool
<< dendl
;
217 tier
= get_new_pool(tier_pool
, orig
);
221 if (tier
->tier_of
!= new_pool
.first
) {
222 lderr(cct
) << __func__
<< " " << r
->first
<< " tier_of != " << new_pool
.first
<< dendl
;
226 ldout(cct
, 10) << __func__
<< " from " << new_pool
.first
<< " to "
227 << tier_pool
<< dendl
;
228 tier
->snap_seq
= base
.snap_seq
;
229 tier
->snap_epoch
= base
.snap_epoch
;
230 tier
->snaps
= base
.snaps
;
231 tier
->removed_snaps
= base
.removed_snaps
;
239 bool OSDMap::subtree_is_down(int id
, set
<int> *down_cache
) const
245 down_cache
->count(id
)) {
250 crush
->get_children(id
, &children
);
251 for (const auto &child
: children
) {
252 if (!subtree_is_down(child
, down_cache
)) {
257 down_cache
->insert(id
);
262 bool OSDMap::containing_subtree_is_down(CephContext
*cct
, int id
, int subtree_type
, set
<int> *down_cache
) const
264 // use a stack-local down_cache if we didn't get one from the
265 // caller. then at least this particular call will avoid duplicated
267 set
<int> local_down_cache
;
269 down_cache
= &local_down_cache
;
278 type
= crush
->get_bucket_type(current
);
282 if (!subtree_is_down(current
, down_cache
)) {
283 ldout(cct
, 30) << "containing_subtree_is_down(" << id
<< ") = false" << dendl
;
287 // is this a big enough subtree to be marked as down?
288 if (type
>= subtree_type
) {
289 ldout(cct
, 30) << "containing_subtree_is_down(" << id
<< ") = true ... " << type
<< " >= " << subtree_type
<< dendl
;
293 int r
= crush
->get_immediate_parent_id(current
, ¤t
);
300 bool OSDMap::subtree_type_is_down(
304 set
<int> *down_in_osds
,
305 set
<int> *up_in_osds
,
306 set
<int> *subtree_up
,
307 unordered_map
<int, set
<int> > *subtree_type_down
) const
310 bool is_down_ret
= is_down(id
);
313 down_in_osds
->insert(id
);
315 up_in_osds
->insert(id
);
321 if (subtree_type_down
&&
322 (*subtree_type_down
)[subtree_type
].count(id
)) {
327 crush
->get_children(id
, &children
);
328 for (const auto &child
: children
) {
329 if (!subtree_type_is_down(
330 cct
, child
, crush
->get_bucket_type(child
),
331 down_in_osds
, up_in_osds
, subtree_up
, subtree_type_down
)) {
332 subtree_up
->insert(id
);
336 if (subtree_type_down
) {
337 (*subtree_type_down
)[subtree_type
].insert(id
);
342 void OSDMap::Incremental::encode_client_old(bufferlist
& bl
) const
348 ::encode(modified
, bl
);
349 int32_t new_t
= new_pool_max
;
351 ::encode(new_flags
, bl
);
352 ::encode(fullmap
, bl
);
355 ::encode(new_max_osd
, bl
);
356 // for ::encode(new_pools, bl);
357 __u32 n
= new_pools
.size();
359 for (const auto &new_pool
: new_pools
) {
362 ::encode(new_pool
.second
, bl
, 0);
364 // for ::encode(new_pool_names, bl);
365 n
= new_pool_names
.size();
368 for (const auto &new_pool_name
: new_pool_names
) {
369 n
= new_pool_name
.first
;
371 ::encode(new_pool_name
.second
, bl
);
373 // for ::encode(old_pools, bl);
374 n
= old_pools
.size();
376 for (auto &old_pool
: old_pools
) {
380 ::encode(new_up_client
, bl
, 0);
382 // legacy is map<int32_t,uint8_t>
383 uint32_t n
= new_state
.size();
385 for (auto p
: new_state
) {
386 ::encode(p
.first
, bl
);
387 ::encode((uint8_t)p
.second
, bl
);
390 ::encode(new_weight
, bl
);
391 // for ::encode(new_pg_temp, bl);
392 n
= new_pg_temp
.size();
395 for (const auto &pg_temp
: new_pg_temp
) {
396 old_pg_t opg
= pg_temp
.first
.get_old_pg();
398 ::encode(pg_temp
.second
, bl
);
402 void OSDMap::Incremental::encode_classic(bufferlist
& bl
, uint64_t features
) const
404 if ((features
& CEPH_FEATURE_PGID64
) == 0) {
405 encode_client_old(bl
);
414 ::encode(modified
, bl
);
415 ::encode(new_pool_max
, bl
);
416 ::encode(new_flags
, bl
);
417 ::encode(fullmap
, bl
);
420 ::encode(new_max_osd
, bl
);
421 ::encode(new_pools
, bl
, features
);
422 ::encode(new_pool_names
, bl
);
423 ::encode(old_pools
, bl
);
424 ::encode(new_up_client
, bl
, features
);
426 uint32_t n
= new_state
.size();
428 for (auto p
: new_state
) {
429 ::encode(p
.first
, bl
);
430 ::encode((uint8_t)p
.second
, bl
);
433 ::encode(new_weight
, bl
);
434 ::encode(new_pg_temp
, bl
);
439 ::encode(new_hb_back_up
, bl
, features
);
440 ::encode(new_up_thru
, bl
);
441 ::encode(new_last_clean_interval
, bl
);
442 ::encode(new_lost
, bl
);
443 ::encode(new_blacklist
, bl
, features
);
444 ::encode(old_blacklist
, bl
, features
);
445 ::encode(new_up_cluster
, bl
, features
);
446 ::encode(cluster_snapshot
, bl
);
447 ::encode(new_uuid
, bl
);
448 ::encode(new_xinfo
, bl
);
449 ::encode(new_hb_front_up
, bl
, features
);
452 void OSDMap::Incremental::encode(bufferlist
& bl
, uint64_t features
) const
454 if ((features
& CEPH_FEATURE_OSDMAP_ENC
) == 0) {
455 encode_classic(bl
, features
);
459 // only a select set of callers should *ever* be encoding new
460 // OSDMaps. others should be passing around the canonical encoded
461 // buffers from on high. select out those callers by passing in an
462 // "impossible" feature bit.
463 assert(features
& CEPH_FEATURE_RESERVED
);
464 features
&= ~CEPH_FEATURE_RESERVED
;
466 size_t start_offset
= bl
.length();
468 buffer::list::iterator crc_it
;
470 // meta-encoding: how we include client-used and osd-specific data
471 ENCODE_START(8, 7, bl
);
475 if (!HAVE_FEATURE(features
, SERVER_LUMINOUS
)) {
478 ENCODE_START(v
, 1, bl
); // client-usable data
481 ::encode(modified
, bl
);
482 ::encode(new_pool_max
, bl
);
483 ::encode(new_flags
, bl
);
484 ::encode(fullmap
, bl
);
487 ::encode(new_max_osd
, bl
);
488 ::encode(new_pools
, bl
, features
);
489 ::encode(new_pool_names
, bl
);
490 ::encode(old_pools
, bl
);
491 ::encode(new_up_client
, bl
, features
);
493 ::encode(new_state
, bl
);
495 uint32_t n
= new_state
.size();
497 for (auto p
: new_state
) {
498 ::encode(p
.first
, bl
);
499 ::encode((uint8_t)p
.second
, bl
);
502 ::encode(new_weight
, bl
);
503 ::encode(new_pg_temp
, bl
);
504 ::encode(new_primary_temp
, bl
);
505 ::encode(new_primary_affinity
, bl
);
506 ::encode(new_erasure_code_profiles
, bl
);
507 ::encode(old_erasure_code_profiles
, bl
);
509 ::encode(new_pg_upmap
, bl
);
510 ::encode(old_pg_upmap
, bl
);
511 ::encode(new_pg_upmap_items
, bl
);
512 ::encode(old_pg_upmap_items
, bl
);
514 ENCODE_FINISH(bl
); // client-usable data
518 uint8_t target_v
= 6;
519 if (!HAVE_FEATURE(features
, SERVER_LUMINOUS
)) {
522 ENCODE_START(target_v
, 1, bl
); // extended, osd-only data
523 ::encode(new_hb_back_up
, bl
, features
);
524 ::encode(new_up_thru
, bl
);
525 ::encode(new_last_clean_interval
, bl
);
526 ::encode(new_lost
, bl
);
527 ::encode(new_blacklist
, bl
, features
);
528 ::encode(old_blacklist
, bl
, features
);
529 ::encode(new_up_cluster
, bl
, features
);
530 ::encode(cluster_snapshot
, bl
);
531 ::encode(new_uuid
, bl
);
532 ::encode(new_xinfo
, bl
);
533 ::encode(new_hb_front_up
, bl
, features
);
534 ::encode(features
, bl
); // NOTE: features arg, not the member
536 ::encode(new_nearfull_ratio
, bl
);
537 ::encode(new_full_ratio
, bl
);
538 ::encode(new_backfillfull_ratio
, bl
);
540 // 5 was string-based new_require_min_compat_client
542 ::encode(new_require_min_compat_client
, bl
);
543 ::encode(new_require_osd_release
, bl
);
545 ENCODE_FINISH(bl
); // osd-only data
548 ::encode((uint32_t)0, bl
); // dummy inc_crc
551 tail_offset
= bl
.length();
553 ::encode(full_crc
, bl
);
555 ENCODE_FINISH(bl
); // meta-encoding wrapper
559 front
.substr_of(bl
, start_offset
, crc_it
.get_off() - start_offset
);
560 inc_crc
= front
.crc32c(-1);
562 tail
.substr_of(bl
, tail_offset
, bl
.length() - tail_offset
);
563 inc_crc
= tail
.crc32c(inc_crc
);
566 crc_it
.copy_in(4, (char*)&crc_le
);
570 void OSDMap::Incremental::decode_classic(bufferlist::iterator
&p
)
578 ::decode(modified
, p
);
579 if (v
== 4 || v
== 5) {
583 ::decode(new_pool_max
, p
);
584 ::decode(new_flags
, p
);
585 ::decode(fullmap
, p
);
588 ::decode(new_max_osd
, p
);
594 ::decode(new_pools
[t
], p
);
597 ::decode(new_pools
, p
);
600 new_pool_names
.clear();
604 ::decode(new_pool_names
[t
], p
);
607 ::decode(new_pool_names
, p
);
617 ::decode(old_pools
, p
);
619 ::decode(new_up_client
, p
);
621 map
<int32_t,uint8_t> ns
;
624 new_state
[q
.first
] = q
.second
;
627 ::decode(new_weight
, p
);
634 ::decode_raw(opg
, p
);
635 ::decode(new_pg_temp
[pg_t(opg
)], p
);
638 ::decode(new_pg_temp
, p
);
641 // decode short map, too.
642 if (v
== 5 && p
.end())
649 ::decode(new_hb_back_up
, p
);
651 ::decode(new_pool_names
, p
);
652 ::decode(new_up_thru
, p
);
653 ::decode(new_last_clean_interval
, p
);
654 ::decode(new_lost
, p
);
655 ::decode(new_blacklist
, p
);
656 ::decode(old_blacklist
, p
);
658 ::decode(new_up_cluster
, p
);
660 ::decode(cluster_snapshot
, p
);
662 ::decode(new_uuid
, p
);
664 ::decode(new_xinfo
, p
);
666 ::decode(new_hb_front_up
, p
);
669 void OSDMap::Incremental::decode(bufferlist::iterator
& bl
)
672 * Older encodings of the Incremental had a single struct_v which
673 * covered the whole encoding, and was prior to our modern
674 * stuff which includes a compatv and a size. So if we see
675 * a struct_v < 7, we must rewind to the beginning and use our
678 size_t start_offset
= bl
.get_off();
679 size_t tail_offset
= 0;
680 bufferlist crc_front
, crc_tail
;
682 DECODE_START_LEGACY_COMPAT_LEN(8, 7, 7, bl
); // wrapper
684 int struct_v_size
= sizeof(struct_v
);
685 bl
.advance(-struct_v_size
);
689 encode_features
= CEPH_FEATURE_PGID64
;
695 DECODE_START(5, bl
); // client-usable data
698 ::decode(modified
, bl
);
699 ::decode(new_pool_max
, bl
);
700 ::decode(new_flags
, bl
);
701 ::decode(fullmap
, bl
);
704 ::decode(new_max_osd
, bl
);
705 ::decode(new_pools
, bl
);
706 ::decode(new_pool_names
, bl
);
707 ::decode(old_pools
, bl
);
708 ::decode(new_up_client
, bl
);
710 ::decode(new_state
, bl
);
712 map
<int32_t,uint8_t> ns
;
715 new_state
[q
.first
] = q
.second
;
718 ::decode(new_weight
, bl
);
719 ::decode(new_pg_temp
, bl
);
720 ::decode(new_primary_temp
, bl
);
722 ::decode(new_primary_affinity
, bl
);
724 new_primary_affinity
.clear();
726 ::decode(new_erasure_code_profiles
, bl
);
727 ::decode(old_erasure_code_profiles
, bl
);
729 new_erasure_code_profiles
.clear();
730 old_erasure_code_profiles
.clear();
733 ::decode(new_pg_upmap
, bl
);
734 ::decode(old_pg_upmap
, bl
);
735 ::decode(new_pg_upmap_items
, bl
);
736 ::decode(old_pg_upmap_items
, bl
);
738 DECODE_FINISH(bl
); // client-usable data
742 DECODE_START(6, bl
); // extended, osd-only data
743 ::decode(new_hb_back_up
, bl
);
744 ::decode(new_up_thru
, bl
);
745 ::decode(new_last_clean_interval
, bl
);
746 ::decode(new_lost
, bl
);
747 ::decode(new_blacklist
, bl
);
748 ::decode(old_blacklist
, bl
);
749 ::decode(new_up_cluster
, bl
);
750 ::decode(cluster_snapshot
, bl
);
751 ::decode(new_uuid
, bl
);
752 ::decode(new_xinfo
, bl
);
753 ::decode(new_hb_front_up
, bl
);
755 ::decode(encode_features
, bl
);
757 encode_features
= CEPH_FEATURE_PGID64
| CEPH_FEATURE_OSDMAP_ENC
;
759 ::decode(new_nearfull_ratio
, bl
);
760 ::decode(new_full_ratio
, bl
);
762 new_nearfull_ratio
= -1;
766 ::decode(new_backfillfull_ratio
, bl
);
768 new_backfillfull_ratio
= -1;
774 new_require_min_compat_client
= ceph_release_from_name(r
.c_str());
778 ::decode(new_require_min_compat_client
, bl
);
779 ::decode(new_require_osd_release
, bl
);
781 if (new_flags
>= 0 && (new_flags
& CEPH_OSDMAP_REQUIRE_LUMINOUS
)) {
782 // only for compat with post-kraken pre-luminous test clusters
783 new_require_osd_release
= CEPH_RELEASE_LUMINOUS
;
784 new_flags
&= ~(CEPH_OSDMAP_LEGACY_REQUIRE_FLAGS
);
785 } else if (new_flags
>= 0 && (new_flags
& CEPH_OSDMAP_REQUIRE_KRAKEN
)) {
786 new_require_osd_release
= CEPH_RELEASE_KRAKEN
;
787 } else if (new_flags
>= 0 && (new_flags
& CEPH_OSDMAP_REQUIRE_JEWEL
)) {
788 new_require_osd_release
= CEPH_RELEASE_JEWEL
;
790 new_require_osd_release
= -1;
793 DECODE_FINISH(bl
); // osd-only data
798 crc_front
.substr_of(bl
.get_bl(), start_offset
, bl
.get_off() - start_offset
);
799 ::decode(inc_crc
, bl
);
800 tail_offset
= bl
.get_off();
801 ::decode(full_crc
, bl
);
808 DECODE_FINISH(bl
); // wrapper
812 uint32_t actual
= crc_front
.crc32c(-1);
813 if (tail_offset
< bl
.get_off()) {
815 tail
.substr_of(bl
.get_bl(), tail_offset
, bl
.get_off() - tail_offset
);
816 actual
= tail
.crc32c(actual
);
818 if (inc_crc
!= actual
) {
820 ss
<< "bad crc, actual " << actual
<< " != expected " << inc_crc
;
822 throw buffer::malformed_input(s
.c_str());
827 void OSDMap::Incremental::dump(Formatter
*f
) const
829 f
->dump_int("epoch", epoch
);
830 f
->dump_stream("fsid") << fsid
;
831 f
->dump_stream("modified") << modified
;
832 f
->dump_int("new_pool_max", new_pool_max
);
833 f
->dump_int("new_flags", new_flags
);
834 f
->dump_float("new_full_ratio", new_full_ratio
);
835 f
->dump_float("new_nearfull_ratio", new_nearfull_ratio
);
836 f
->dump_float("new_backfillfull_ratio", new_backfillfull_ratio
);
837 f
->dump_int("new_require_min_compat_client", new_require_min_compat_client
);
838 f
->dump_int("new_require_osd_release", new_require_osd_release
);
840 if (fullmap
.length()) {
841 f
->open_object_section("full_map");
843 bufferlist fbl
= fullmap
; // kludge around constness.
844 auto p
= fbl
.begin();
849 if (crush
.length()) {
850 f
->open_object_section("crush");
852 bufferlist tbl
= crush
; // kludge around constness.
853 auto p
= tbl
.begin();
859 f
->dump_int("new_max_osd", new_max_osd
);
861 f
->open_array_section("new_pools");
863 for (const auto &new_pool
: new_pools
) {
864 f
->open_object_section("pool");
865 f
->dump_int("pool", new_pool
.first
);
866 new_pool
.second
.dump(f
);
870 f
->open_array_section("new_pool_names");
872 for (const auto &new_pool_name
: new_pool_names
) {
873 f
->open_object_section("pool_name");
874 f
->dump_int("pool", new_pool_name
.first
);
875 f
->dump_string("name", new_pool_name
.second
);
879 f
->open_array_section("old_pools");
881 for (const auto &old_pool
: old_pools
)
882 f
->dump_int("pool", old_pool
);
885 f
->open_array_section("new_up_osds");
887 for (const auto &upclient
: new_up_client
) {
888 f
->open_object_section("osd");
889 f
->dump_int("osd", upclient
.first
);
890 f
->dump_stream("public_addr") << upclient
.second
;
891 f
->dump_stream("cluster_addr") << new_up_cluster
.find(upclient
.first
)->second
;
892 f
->dump_stream("heartbeat_back_addr") << new_hb_back_up
.find(upclient
.first
)->second
;
893 map
<int32_t, entity_addr_t
>::const_iterator q
;
894 if ((q
= new_hb_front_up
.find(upclient
.first
)) != new_hb_front_up
.end())
895 f
->dump_stream("heartbeat_front_addr") << q
->second
;
900 f
->open_array_section("new_weight");
902 for (const auto &weight
: new_weight
) {
903 f
->open_object_section("osd");
904 f
->dump_int("osd", weight
.first
);
905 f
->dump_int("weight", weight
.second
);
910 f
->open_array_section("osd_state_xor");
911 for (const auto &ns
: new_state
) {
912 f
->open_object_section("osd");
913 f
->dump_int("osd", ns
.first
);
915 calc_state_set(new_state
.find(ns
.first
)->second
, st
);
916 f
->open_array_section("state_xor");
917 for (auto &state
: st
)
918 f
->dump_string("state", state
);
924 f
->open_array_section("new_pg_temp");
926 for (const auto &pg_temp
: new_pg_temp
) {
927 f
->open_object_section("pg");
928 f
->dump_stream("pgid") << pg_temp
.first
;
929 f
->open_array_section("osds");
931 for (const auto &osd
: pg_temp
.second
)
932 f
->dump_int("osd", osd
);
938 f
->open_array_section("primary_temp");
940 for (const auto &primary_temp
: new_primary_temp
) {
941 f
->dump_stream("pgid") << primary_temp
.first
;
942 f
->dump_int("osd", primary_temp
.second
);
944 f
->close_section(); // primary_temp
946 f
->open_array_section("new_pg_upmap");
947 for (auto& i
: new_pg_upmap
) {
948 f
->open_object_section("mapping");
949 f
->dump_stream("pgid") << i
.first
;
950 f
->open_array_section("osds");
951 for (auto osd
: i
.second
) {
952 f
->dump_int("osd", osd
);
958 f
->open_array_section("old_pg_upmap");
959 for (auto& i
: old_pg_upmap
) {
960 f
->dump_stream("pgid") << i
;
964 f
->open_array_section("new_pg_upmap_items");
965 for (auto& i
: new_pg_upmap_items
) {
966 f
->open_object_section("mapping");
967 f
->dump_stream("pgid") << i
.first
;
968 f
->open_array_section("mappings");
969 for (auto& p
: i
.second
) {
970 f
->open_object_section("mapping");
971 f
->dump_int("from", p
.first
);
972 f
->dump_int("to", p
.second
);
979 f
->open_array_section("old_pg_upmap_items");
980 for (auto& i
: old_pg_upmap_items
) {
981 f
->dump_stream("pgid") << i
;
985 f
->open_array_section("new_up_thru");
987 for (const auto &up_thru
: new_up_thru
) {
988 f
->open_object_section("osd");
989 f
->dump_int("osd", up_thru
.first
);
990 f
->dump_int("up_thru", up_thru
.second
);
995 f
->open_array_section("new_lost");
997 for (const auto &lost
: new_lost
) {
998 f
->open_object_section("osd");
999 f
->dump_int("osd", lost
.first
);
1000 f
->dump_int("epoch_lost", lost
.second
);
1005 f
->open_array_section("new_last_clean_interval");
1007 for (const auto &last_clean_interval
: new_last_clean_interval
) {
1008 f
->open_object_section("osd");
1009 f
->dump_int("osd", last_clean_interval
.first
);
1010 f
->dump_int("first", last_clean_interval
.second
.first
);
1011 f
->dump_int("last", last_clean_interval
.second
.second
);
1016 f
->open_array_section("new_blacklist");
1017 for (const auto &blist
: new_blacklist
) {
1020 f
->dump_stream(ss
.str().c_str()) << blist
.second
;
1023 f
->open_array_section("old_blacklist");
1024 for (const auto &blist
: old_blacklist
)
1025 f
->dump_stream("addr") << blist
;
1028 f
->open_array_section("new_xinfo");
1029 for (const auto &xinfo
: new_xinfo
) {
1030 f
->open_object_section("xinfo");
1031 f
->dump_int("osd", xinfo
.first
);
1032 xinfo
.second
.dump(f
);
1037 if (cluster_snapshot
.size())
1038 f
->dump_string("cluster_snapshot", cluster_snapshot
);
1040 f
->open_array_section("new_uuid");
1041 for (const auto &uuid
: new_uuid
) {
1042 f
->open_object_section("osd");
1043 f
->dump_int("osd", uuid
.first
);
1044 f
->dump_stream("uuid") << uuid
.second
;
1049 OSDMap::dump_erasure_code_profiles(new_erasure_code_profiles
, f
);
1050 f
->open_array_section("old_erasure_code_profiles");
1051 for (const auto &erasure_code_profile
: old_erasure_code_profiles
) {
1052 f
->dump_string("old", erasure_code_profile
.c_str());
1057 void OSDMap::Incremental::generate_test_instances(list
<Incremental
*>& o
)
1059 o
.push_back(new Incremental
);
1062 // ----------------------------------
1065 void OSDMap::set_epoch(epoch_t e
)
1068 for (auto &pool
: pools
)
1069 pool
.second
.last_change
= e
;
1072 bool OSDMap::is_blacklisted(const entity_addr_t
& a
) const
1074 if (blacklist
.empty())
1077 // this specific instance?
1078 if (blacklist
.count(a
))
1081 // is entire ip blacklisted?
1083 entity_addr_t b
= a
;
1086 if (blacklist
.count(b
)) {
1094 void OSDMap::get_blacklist(list
<pair
<entity_addr_t
,utime_t
> > *bl
) const
1096 std::copy(blacklist
.begin(), blacklist
.end(), std::back_inserter(*bl
));
1099 void OSDMap::get_blacklist(std::set
<entity_addr_t
> *bl
) const
1101 for (const auto &i
: blacklist
) {
1102 bl
->insert(i
.first
);
1106 void OSDMap::set_max_osd(int m
)
1110 osd_state
.resize(m
);
1111 osd_weight
.resize(m
);
1112 for (; o
<max_osd
; o
++) {
1114 osd_weight
[o
] = CEPH_OSD_OUT
;
1117 osd_xinfo
.resize(m
);
1118 osd_addrs
->client_addr
.resize(m
);
1119 osd_addrs
->cluster_addr
.resize(m
);
1120 osd_addrs
->hb_back_addr
.resize(m
);
1121 osd_addrs
->hb_front_addr
.resize(m
);
1122 osd_uuid
->resize(m
);
1123 if (osd_primary_affinity
)
1124 osd_primary_affinity
->resize(m
, CEPH_OSD_DEFAULT_PRIMARY_AFFINITY
);
1129 int OSDMap::calc_num_osds()
1134 for (int i
=0; i
<max_osd
; i
++) {
1135 if (osd_state
[i
] & CEPH_OSD_EXISTS
) {
1137 if (osd_state
[i
] & CEPH_OSD_UP
) {
1140 if (get_weight(i
) != CEPH_OSD_OUT
) {
1148 void OSDMap::count_full_nearfull_osds(int *full
, int *backfill
, int *nearfull
) const
1153 for (int i
= 0; i
< max_osd
; ++i
) {
1154 if (exists(i
) && is_up(i
) && is_in(i
)) {
1155 if (osd_state
[i
] & CEPH_OSD_FULL
)
1157 else if (osd_state
[i
] & CEPH_OSD_BACKFILLFULL
)
1159 else if (osd_state
[i
] & CEPH_OSD_NEARFULL
)
1165 static bool get_osd_utilization(
1166 const mempool::pgmap::unordered_map
<int32_t,osd_stat_t
> &osd_stat
,
1167 int id
, int64_t* kb
, int64_t* kb_used
, int64_t* kb_avail
)
1169 auto p
= osd_stat
.find(id
);
1170 if (p
== osd_stat
.end())
1173 *kb_used
= p
->second
.kb_used
;
1174 *kb_avail
= p
->second
.kb_avail
;
1178 void OSDMap::get_full_osd_util(
1179 const mempool::pgmap::unordered_map
<int32_t,osd_stat_t
> &osd_stat
,
1180 map
<int, float> *full
, map
<int, float> *backfill
, map
<int, float> *nearfull
) const
1185 for (int i
= 0; i
< max_osd
; ++i
) {
1186 if (exists(i
) && is_up(i
) && is_in(i
)) {
1187 int64_t kb
, kb_used
, kb_avail
;
1188 if (osd_state
[i
] & CEPH_OSD_FULL
) {
1189 if (get_osd_utilization(osd_stat
, i
, &kb
, &kb_used
, &kb_avail
))
1190 full
->emplace(i
, (float)kb_used
/ (float)kb
);
1191 } else if (osd_state
[i
] & CEPH_OSD_BACKFILLFULL
) {
1192 if (get_osd_utilization(osd_stat
, i
, &kb
, &kb_used
, &kb_avail
))
1193 backfill
->emplace(i
, (float)kb_used
/ (float)kb
);
1194 } else if (osd_state
[i
] & CEPH_OSD_NEARFULL
) {
1195 if (get_osd_utilization(osd_stat
, i
, &kb
, &kb_used
, &kb_avail
))
1196 nearfull
->emplace(i
, (float)kb_used
/ (float)kb
);
1202 void OSDMap::get_full_osd_counts(set
<int> *full
, set
<int> *backfill
,
1203 set
<int> *nearfull
) const
1208 for (int i
= 0; i
< max_osd
; ++i
) {
1209 if (exists(i
) && is_up(i
) && is_in(i
)) {
1210 if (osd_state
[i
] & CEPH_OSD_FULL
)
1212 else if (osd_state
[i
] & CEPH_OSD_BACKFILLFULL
)
1213 backfill
->emplace(i
);
1214 else if (osd_state
[i
] & CEPH_OSD_NEARFULL
)
1215 nearfull
->emplace(i
);
1220 void OSDMap::get_all_osds(set
<int32_t>& ls
) const
1222 for (int i
=0; i
<max_osd
; i
++)
1227 void OSDMap::get_up_osds(set
<int32_t>& ls
) const
1229 for (int i
= 0; i
< max_osd
; i
++) {
1235 void OSDMap::get_out_osds(set
<int32_t>& ls
) const
1237 for (int i
= 0; i
< max_osd
; i
++) {
1243 void OSDMap::calc_state_set(int state
, set
<string
>& st
)
1246 for (unsigned s
= 1; t
; s
<<= 1) {
1249 st
.insert(ceph_osd_state_name(s
));
1254 void OSDMap::adjust_osd_weights(const map
<int,double>& weights
, Incremental
& inc
) const
1257 for (const auto &weight
: weights
) {
1258 if (weight
.second
> max
)
1259 max
= weight
.second
;
1262 for (const auto &weight
: weights
) {
1263 inc
.new_weight
[weight
.first
] = (unsigned)((weight
.second
/ max
) * CEPH_OSD_IN
);
1267 int OSDMap::identify_osd(const entity_addr_t
& addr
) const
1269 for (int i
=0; i
<max_osd
; i
++)
1270 if (exists(i
) && (get_addr(i
) == addr
|| get_cluster_addr(i
) == addr
))
1275 int OSDMap::identify_osd(const uuid_d
& u
) const
1277 for (int i
=0; i
<max_osd
; i
++)
1278 if (exists(i
) && get_uuid(i
) == u
)
1283 int OSDMap::identify_osd_on_all_channels(const entity_addr_t
& addr
) const
1285 for (int i
=0; i
<max_osd
; i
++)
1286 if (exists(i
) && (get_addr(i
) == addr
|| get_cluster_addr(i
) == addr
||
1287 get_hb_back_addr(i
) == addr
|| get_hb_front_addr(i
) == addr
))
1292 int OSDMap::find_osd_on_ip(const entity_addr_t
& ip
) const
1294 for (int i
=0; i
<max_osd
; i
++)
1295 if (exists(i
) && (get_addr(i
).is_same_host(ip
) || get_cluster_addr(i
).is_same_host(ip
)))
1301 uint64_t OSDMap::get_features(int entity_type
, uint64_t *pmask
) const
1303 uint64_t features
= 0; // things we actually have
1304 uint64_t mask
= 0; // things we could have
1306 if (crush
->has_nondefault_tunables())
1307 features
|= CEPH_FEATURE_CRUSH_TUNABLES
;
1308 if (crush
->has_nondefault_tunables2())
1309 features
|= CEPH_FEATURE_CRUSH_TUNABLES2
;
1310 if (crush
->has_nondefault_tunables3())
1311 features
|= CEPH_FEATURE_CRUSH_TUNABLES3
;
1312 if (crush
->has_v4_buckets())
1313 features
|= CEPH_FEATURE_CRUSH_V4
;
1314 if (crush
->has_nondefault_tunables5())
1315 features
|= CEPH_FEATURE_CRUSH_TUNABLES5
;
1316 if (crush
->has_incompat_choose_args()) {
1317 features
|= CEPH_FEATUREMASK_CRUSH_CHOOSE_ARGS
;
1319 mask
|= CEPH_FEATURES_CRUSH
;
1321 if (!pg_upmap
.empty() || !pg_upmap_items
.empty())
1322 features
|= CEPH_FEATUREMASK_OSDMAP_PG_UPMAP
;
1323 mask
|= CEPH_FEATUREMASK_OSDMAP_PG_UPMAP
;
1325 for (auto &pool
: pools
) {
1326 if (pool
.second
.has_flag(pg_pool_t::FLAG_HASHPSPOOL
)) {
1327 features
|= CEPH_FEATURE_OSDHASHPSPOOL
;
1329 if (pool
.second
.is_erasure() &&
1330 entity_type
!= CEPH_ENTITY_TYPE_CLIENT
) { // not for clients
1331 features
|= CEPH_FEATURE_OSD_ERASURE_CODES
;
1333 if (!pool
.second
.tiers
.empty() ||
1334 pool
.second
.is_tier()) {
1335 features
|= CEPH_FEATURE_OSD_CACHEPOOL
;
1337 int ruleid
= crush
->find_rule(pool
.second
.get_crush_rule(),
1338 pool
.second
.get_type(),
1339 pool
.second
.get_size());
1341 if (crush
->is_v2_rule(ruleid
))
1342 features
|= CEPH_FEATURE_CRUSH_V2
;
1343 if (crush
->is_v3_rule(ruleid
))
1344 features
|= CEPH_FEATURE_CRUSH_TUNABLES3
;
1345 if (crush
->is_v5_rule(ruleid
))
1346 features
|= CEPH_FEATURE_CRUSH_TUNABLES5
;
1349 if (entity_type
== CEPH_ENTITY_TYPE_OSD
) {
1350 for (auto &erasure_code_profile
: erasure_code_profiles
) {
1351 auto& profile
= erasure_code_profile
.second
;
1352 const auto& plugin
= profile
.find("plugin");
1353 if (plugin
!= profile
.end()) {
1354 if (plugin
->second
== "isa" || plugin
->second
== "lrc")
1355 features
|= CEPH_FEATURE_ERASURE_CODE_PLUGINS_V2
;
1356 if (plugin
->second
== "shec")
1357 features
|= CEPH_FEATURE_ERASURE_CODE_PLUGINS_V3
;
1361 mask
|= CEPH_FEATURE_OSDHASHPSPOOL
| CEPH_FEATURE_OSD_CACHEPOOL
;
1362 if (entity_type
!= CEPH_ENTITY_TYPE_CLIENT
)
1363 mask
|= CEPH_FEATURE_OSD_ERASURE_CODES
;
1365 if (osd_primary_affinity
) {
1366 for (int i
= 0; i
< max_osd
; ++i
) {
1367 if ((*osd_primary_affinity
)[i
] != CEPH_OSD_DEFAULT_PRIMARY_AFFINITY
) {
1368 features
|= CEPH_FEATURE_OSD_PRIMARY_AFFINITY
;
1373 mask
|= CEPH_FEATURE_OSD_PRIMARY_AFFINITY
;
1375 if (entity_type
== CEPH_ENTITY_TYPE_OSD
) {
1376 const uint64_t jewel_features
= CEPH_FEATURE_SERVER_JEWEL
;
1377 if (require_osd_release
>= CEPH_RELEASE_JEWEL
) {
1378 features
|= jewel_features
;
1380 mask
|= jewel_features
;
1382 const uint64_t kraken_features
= CEPH_FEATUREMASK_SERVER_KRAKEN
1383 | CEPH_FEATURE_MSG_ADDR2
;
1384 if (require_osd_release
>= CEPH_RELEASE_KRAKEN
) {
1385 features
|= kraken_features
;
1387 mask
|= kraken_features
;
1395 uint8_t OSDMap::get_min_compat_client() const
1397 uint64_t f
= get_features(CEPH_ENTITY_TYPE_CLIENT
, nullptr);
1399 if (HAVE_FEATURE(f
, OSDMAP_PG_UPMAP
) || // v12.0.0-1733-g27d6f43
1400 HAVE_FEATURE(f
, CRUSH_CHOOSE_ARGS
)) { // v12.0.1-2172-gef1ef28
1401 return CEPH_RELEASE_LUMINOUS
; // v12.2.0
1403 if (HAVE_FEATURE(f
, CRUSH_TUNABLES5
)) { // v10.0.0-612-g043a737
1404 return CEPH_RELEASE_JEWEL
; // v10.2.0
1406 if (HAVE_FEATURE(f
, CRUSH_V4
)) { // v0.91-678-g325fc56
1407 return CEPH_RELEASE_HAMMER
; // v0.94.0
1409 if (HAVE_FEATURE(f
, OSD_PRIMARY_AFFINITY
) || // v0.76-553-gf825624
1410 HAVE_FEATURE(f
, CRUSH_TUNABLES3
) || // v0.76-395-ge20a55d
1411 HAVE_FEATURE(f
, OSD_ERASURE_CODES
) || // v0.73-498-gbfc86a8
1412 HAVE_FEATURE(f
, OSD_CACHEPOOL
)) { // v0.67-401-gb91c1c5
1413 return CEPH_RELEASE_FIREFLY
; // v0.80.0
1415 if (HAVE_FEATURE(f
, CRUSH_TUNABLES2
) || // v0.54-684-g0cc47ff
1416 HAVE_FEATURE(f
, OSDHASHPSPOOL
)) { // v0.57-398-g8cc2b0f
1417 return CEPH_RELEASE_DUMPLING
; // v0.67.0
1419 if (HAVE_FEATURE(f
, CRUSH_TUNABLES
)) { // v0.48argonaut-206-g6f381af
1420 return CEPH_RELEASE_ARGONAUT
; // v0.48argonaut-206-g6f381af
1422 return CEPH_RELEASE_ARGONAUT
; // v0.48argonaut-206-g6f381af
1425 void OSDMap::_calc_up_osd_features()
1428 cached_up_osd_features
= 0;
1429 for (int osd
= 0; osd
< max_osd
; ++osd
) {
1432 const osd_xinfo_t
&xi
= get_xinfo(osd
);
1434 cached_up_osd_features
= xi
.features
;
1437 cached_up_osd_features
&= xi
.features
;
1442 uint64_t OSDMap::get_up_osd_features() const
1444 return cached_up_osd_features
;
1447 void OSDMap::dedup(const OSDMap
*o
, OSDMap
*n
)
1449 if (o
->epoch
== n
->epoch
)
1455 if (o
->max_osd
!= n
->max_osd
)
1457 for (int i
= 0; i
< o
->max_osd
&& i
< n
->max_osd
; i
++) {
1458 if ( n
->osd_addrs
->client_addr
[i
] && o
->osd_addrs
->client_addr
[i
] &&
1459 *n
->osd_addrs
->client_addr
[i
] == *o
->osd_addrs
->client_addr
[i
])
1460 n
->osd_addrs
->client_addr
[i
] = o
->osd_addrs
->client_addr
[i
];
1463 if ( n
->osd_addrs
->cluster_addr
[i
] && o
->osd_addrs
->cluster_addr
[i
] &&
1464 *n
->osd_addrs
->cluster_addr
[i
] == *o
->osd_addrs
->cluster_addr
[i
])
1465 n
->osd_addrs
->cluster_addr
[i
] = o
->osd_addrs
->cluster_addr
[i
];
1468 if ( n
->osd_addrs
->hb_back_addr
[i
] && o
->osd_addrs
->hb_back_addr
[i
] &&
1469 *n
->osd_addrs
->hb_back_addr
[i
] == *o
->osd_addrs
->hb_back_addr
[i
])
1470 n
->osd_addrs
->hb_back_addr
[i
] = o
->osd_addrs
->hb_back_addr
[i
];
1473 if ( n
->osd_addrs
->hb_front_addr
[i
] && o
->osd_addrs
->hb_front_addr
[i
] &&
1474 *n
->osd_addrs
->hb_front_addr
[i
] == *o
->osd_addrs
->hb_front_addr
[i
])
1475 n
->osd_addrs
->hb_front_addr
[i
] = o
->osd_addrs
->hb_front_addr
[i
];
1480 // zoinks, no differences at all!
1481 n
->osd_addrs
= o
->osd_addrs
;
1484 // does crush match?
1486 ::encode(*o
->crush
, oc
, CEPH_FEATURES_SUPPORTED_DEFAULT
);
1487 ::encode(*n
->crush
, nc
, CEPH_FEATURES_SUPPORTED_DEFAULT
);
1488 if (oc
.contents_equal(nc
)) {
1489 n
->crush
= o
->crush
;
1492 // does pg_temp match?
1493 if (*o
->pg_temp
== *n
->pg_temp
)
1494 n
->pg_temp
= o
->pg_temp
;
1496 // does primary_temp match?
1497 if (o
->primary_temp
->size() == n
->primary_temp
->size()) {
1498 if (*o
->primary_temp
== *n
->primary_temp
)
1499 n
->primary_temp
= o
->primary_temp
;
1503 if (o
->osd_uuid
->size() == n
->osd_uuid
->size() &&
1504 *o
->osd_uuid
== *n
->osd_uuid
)
1505 n
->osd_uuid
= o
->osd_uuid
;
1508 void OSDMap::clean_temps(CephContext
*cct
,
1509 const OSDMap
& osdmap
, Incremental
*pending_inc
)
1511 ldout(cct
, 10) << __func__
<< dendl
;
1513 tmpmap
.deepish_copy_from(osdmap
);
1514 tmpmap
.apply_incremental(*pending_inc
);
1516 for (auto pg
: *tmpmap
.pg_temp
) {
1517 // if pool does not exist, remove any existing pg_temps associated with
1518 // it. we don't care about pg_temps on the pending_inc either; if there
1519 // are new_pg_temp entries on the pending, clear them out just as well.
1520 if (!osdmap
.have_pg_pool(pg
.first
.pool())) {
1521 ldout(cct
, 10) << __func__
<< " removing pg_temp " << pg
.first
1522 << " for nonexistent pool " << pg
.first
.pool() << dendl
;
1523 pending_inc
->new_pg_temp
[pg
.first
].clear();
1527 unsigned num_up
= 0;
1528 for (auto o
: pg
.second
) {
1529 if (!tmpmap
.is_down(o
)) {
1535 ldout(cct
, 10) << __func__
<< " removing pg_temp " << pg
.first
1536 << " with all down osds" << pg
.second
<< dendl
;
1537 pending_inc
->new_pg_temp
[pg
.first
].clear();
1540 // redundant pg_temp?
1543 tmpmap
.pg_to_raw_up(pg
.first
, &raw_up
, &primary
);
1544 if (vectors_equal(raw_up
, pg
.second
)) {
1545 ldout(cct
, 10) << __func__
<< " removing pg_temp " << pg
.first
<< " "
1546 << pg
.second
<< " that matches raw_up mapping" << dendl
;
1547 if (osdmap
.pg_temp
->count(pg
.first
))
1548 pending_inc
->new_pg_temp
[pg
.first
].clear();
1550 pending_inc
->new_pg_temp
.erase(pg
.first
);
1554 for (auto &pg
: *tmpmap
.primary_temp
) {
1556 if (tmpmap
.is_down(pg
.second
)) {
1557 ldout(cct
, 10) << __func__
<< " removing primary_temp " << pg
.first
1558 << " to down " << pg
.second
<< dendl
;
1559 pending_inc
->new_primary_temp
[pg
.first
] = -1;
1562 // redundant primary_temp?
1563 vector
<int> real_up
, templess_up
;
1564 int real_primary
, templess_primary
;
1565 pg_t pgid
= pg
.first
;
1566 tmpmap
.pg_to_acting_osds(pgid
, &real_up
, &real_primary
);
1567 tmpmap
.pg_to_raw_up(pgid
, &templess_up
, &templess_primary
);
1568 if (real_primary
== templess_primary
){
1569 ldout(cct
, 10) << __func__
<< " removing primary_temp "
1570 << pgid
<< " -> " << real_primary
1571 << " (unnecessary/redundant)" << dendl
;
1572 if (osdmap
.primary_temp
->count(pgid
))
1573 pending_inc
->new_primary_temp
[pgid
] = -1;
1575 pending_inc
->new_primary_temp
.erase(pgid
);
1580 int OSDMap::apply_incremental(const Incremental
&inc
)
1582 new_blacklist_entries
= false;
1585 else if (inc
.fsid
!= fsid
)
1588 assert(inc
.epoch
== epoch
+1);
1591 modified
= inc
.modified
;
1594 if (inc
.fullmap
.length()) {
1595 bufferlist
bl(inc
.fullmap
);
1600 // nope, incremental.
1601 if (inc
.new_flags
>= 0) {
1602 flags
= inc
.new_flags
;
1603 // the below is just to cover a newly-upgraded luminous mon
1604 // cluster that has to set require_jewel_osds or
1605 // require_kraken_osds before the osds can be upgraded to
1607 if (flags
& CEPH_OSDMAP_REQUIRE_KRAKEN
) {
1608 if (require_osd_release
< CEPH_RELEASE_KRAKEN
) {
1609 require_osd_release
= CEPH_RELEASE_KRAKEN
;
1611 } else if (flags
& CEPH_OSDMAP_REQUIRE_JEWEL
) {
1612 if (require_osd_release
< CEPH_RELEASE_JEWEL
) {
1613 require_osd_release
= CEPH_RELEASE_JEWEL
;
1618 if (inc
.new_max_osd
>= 0)
1619 set_max_osd(inc
.new_max_osd
);
1621 if (inc
.new_pool_max
!= -1)
1622 pool_max
= inc
.new_pool_max
;
1624 for (const auto &pool
: inc
.new_pools
) {
1625 pools
[pool
.first
] = pool
.second
;
1626 pools
[pool
.first
].last_change
= epoch
;
1629 for (const auto &pname
: inc
.new_pool_names
) {
1630 auto pool_name_entry
= pool_name
.find(pname
.first
);
1631 if (pool_name_entry
!= pool_name
.end()) {
1632 name_pool
.erase(pool_name_entry
->second
);
1633 pool_name_entry
->second
= pname
.second
;
1635 pool_name
[pname
.first
] = pname
.second
;
1637 name_pool
[pname
.second
] = pname
.first
;
1640 for (const auto &pool
: inc
.old_pools
) {
1642 name_pool
.erase(pool_name
[pool
]);
1643 pool_name
.erase(pool
);
1646 for (const auto &weight
: inc
.new_weight
) {
1647 set_weight(weight
.first
, weight
.second
);
1649 // if we are marking in, clear the AUTOOUT and NEW bits, and clear
1650 // xinfo old_weight.
1651 if (weight
.second
) {
1652 osd_state
[weight
.first
] &= ~(CEPH_OSD_AUTOOUT
| CEPH_OSD_NEW
);
1653 osd_xinfo
[weight
.first
].old_weight
= 0;
1657 for (const auto &primary_affinity
: inc
.new_primary_affinity
) {
1658 set_primary_affinity(primary_affinity
.first
, primary_affinity
.second
);
1661 // erasure_code_profiles
1662 for (const auto &profile
: inc
.old_erasure_code_profiles
)
1663 erasure_code_profiles
.erase(profile
);
1665 for (const auto &profile
: inc
.new_erasure_code_profiles
) {
1666 set_erasure_code_profile(profile
.first
, profile
.second
);
1670 for (const auto &state
: inc
.new_state
) {
1671 const auto osd
= state
.first
;
1672 int s
= state
.second
? state
.second
: CEPH_OSD_UP
;
1673 if ((osd_state
[osd
] & CEPH_OSD_UP
) &&
1674 (s
& CEPH_OSD_UP
)) {
1675 osd_info
[osd
].down_at
= epoch
;
1676 osd_xinfo
[osd
].down_stamp
= modified
;
1678 if ((osd_state
[osd
] & CEPH_OSD_EXISTS
) &&
1679 (s
& CEPH_OSD_EXISTS
)) {
1680 // osd is destroyed; clear out anything interesting.
1681 (*osd_uuid
)[osd
] = uuid_d();
1682 osd_info
[osd
] = osd_info_t();
1683 osd_xinfo
[osd
] = osd_xinfo_t();
1684 set_primary_affinity(osd
, CEPH_OSD_DEFAULT_PRIMARY_AFFINITY
);
1685 osd_addrs
->client_addr
[osd
].reset(new entity_addr_t());
1686 osd_addrs
->cluster_addr
[osd
].reset(new entity_addr_t());
1687 osd_addrs
->hb_front_addr
[osd
].reset(new entity_addr_t());
1688 osd_addrs
->hb_back_addr
[osd
].reset(new entity_addr_t());
1691 osd_state
[osd
] ^= s
;
1695 for (const auto &client
: inc
.new_up_client
) {
1696 osd_state
[client
.first
] |= CEPH_OSD_EXISTS
| CEPH_OSD_UP
;
1697 osd_addrs
->client_addr
[client
.first
].reset(new entity_addr_t(client
.second
));
1698 if (inc
.new_hb_back_up
.empty())
1699 osd_addrs
->hb_back_addr
[client
.first
].reset(new entity_addr_t(client
.second
)); //this is a backward-compatibility hack
1701 osd_addrs
->hb_back_addr
[client
.first
].reset(
1702 new entity_addr_t(inc
.new_hb_back_up
.find(client
.first
)->second
));
1703 const auto j
= inc
.new_hb_front_up
.find(client
.first
);
1704 if (j
!= inc
.new_hb_front_up
.end())
1705 osd_addrs
->hb_front_addr
[client
.first
].reset(new entity_addr_t(j
->second
));
1707 osd_addrs
->hb_front_addr
[client
.first
].reset();
1709 osd_info
[client
.first
].up_from
= epoch
;
1712 for (const auto &cluster
: inc
.new_up_cluster
)
1713 osd_addrs
->cluster_addr
[cluster
.first
].reset(new entity_addr_t(cluster
.second
));
1716 for (const auto &thru
: inc
.new_up_thru
)
1717 osd_info
[thru
.first
].up_thru
= thru
.second
;
1719 for (const auto &interval
: inc
.new_last_clean_interval
) {
1720 osd_info
[interval
.first
].last_clean_begin
= interval
.second
.first
;
1721 osd_info
[interval
.first
].last_clean_end
= interval
.second
.second
;
1724 for (const auto &lost
: inc
.new_lost
)
1725 osd_info
[lost
.first
].lost_at
= lost
.second
;
1728 for (const auto &xinfo
: inc
.new_xinfo
)
1729 osd_xinfo
[xinfo
.first
] = xinfo
.second
;
1732 for (const auto &uuid
: inc
.new_uuid
)
1733 (*osd_uuid
)[uuid
.first
] = uuid
.second
;
1736 for (const auto &pg
: inc
.new_pg_temp
) {
1737 if (pg
.second
.empty())
1738 pg_temp
->erase(pg
.first
);
1740 pg_temp
->set(pg
.first
, pg
.second
);
1742 if (!inc
.new_pg_temp
.empty()) {
1743 // make sure pg_temp is efficiently stored
1747 for (const auto &pg
: inc
.new_primary_temp
) {
1748 if (pg
.second
== -1)
1749 primary_temp
->erase(pg
.first
);
1751 (*primary_temp
)[pg
.first
] = pg
.second
;
1754 for (auto& p
: inc
.new_pg_upmap
) {
1755 pg_upmap
[p
.first
] = p
.second
;
1757 for (auto& pg
: inc
.old_pg_upmap
) {
1760 for (auto& p
: inc
.new_pg_upmap_items
) {
1761 pg_upmap_items
[p
.first
] = p
.second
;
1763 for (auto& pg
: inc
.old_pg_upmap_items
) {
1764 pg_upmap_items
.erase(pg
);
1768 if (!inc
.new_blacklist
.empty()) {
1769 blacklist
.insert(inc
.new_blacklist
.begin(),inc
.new_blacklist
.end());
1770 new_blacklist_entries
= true;
1772 for (const auto &addr
: inc
.old_blacklist
)
1773 blacklist
.erase(addr
);
1775 // cluster snapshot?
1776 if (inc
.cluster_snapshot
.length()) {
1777 cluster_snapshot
= inc
.cluster_snapshot
;
1778 cluster_snapshot_epoch
= inc
.epoch
;
1780 cluster_snapshot
.clear();
1781 cluster_snapshot_epoch
= 0;
1784 if (inc
.new_nearfull_ratio
>= 0) {
1785 nearfull_ratio
= inc
.new_nearfull_ratio
;
1787 if (inc
.new_backfillfull_ratio
>= 0) {
1788 backfillfull_ratio
= inc
.new_backfillfull_ratio
;
1790 if (inc
.new_full_ratio
>= 0) {
1791 full_ratio
= inc
.new_full_ratio
;
1793 if (inc
.new_require_min_compat_client
> 0) {
1794 require_min_compat_client
= inc
.new_require_min_compat_client
;
1796 if (inc
.new_require_osd_release
>= 0) {
1797 require_osd_release
= inc
.new_require_osd_release
;
1798 if (require_osd_release
>= CEPH_RELEASE_LUMINOUS
) {
1799 flags
&= ~(CEPH_OSDMAP_LEGACY_REQUIRE_FLAGS
);
1800 flags
|= CEPH_OSDMAP_RECOVERY_DELETES
;
1804 // do new crush map last (after up/down stuff)
1805 if (inc
.crush
.length()) {
1806 bufferlist
bl(inc
.crush
);
1807 auto blp
= bl
.begin();
1808 crush
.reset(new CrushWrapper
);
1810 if (require_osd_release
>= CEPH_RELEASE_LUMINOUS
) {
1811 // only increment if this is a luminous-encoded osdmap, lest
1812 // the mon's crush_version diverge from what the osds or others
1813 // are decoding and applying on their end. if we won't encode
1814 // it in the canonical version, don't change it.
1820 _calc_up_osd_features();
1825 int OSDMap::map_to_pg(
1829 const string
& nspace
,
1832 // calculate ps (placement seed)
1833 const pg_pool_t
*pool
= get_pg_pool(poolid
);
1838 ps
= pool
->hash_key(key
, nspace
);
1840 ps
= pool
->hash_key(name
, nspace
);
1841 *pg
= pg_t(ps
, poolid
);
1845 int OSDMap::object_locator_to_pg(
1846 const object_t
& oid
, const object_locator_t
& loc
, pg_t
&pg
) const
1848 if (loc
.hash
>= 0) {
1849 if (!get_pg_pool(loc
.get_pool())) {
1852 pg
= pg_t(loc
.hash
, loc
.get_pool());
1855 return map_to_pg(loc
.get_pool(), oid
.name
, loc
.key
, loc
.nspace
, &pg
);
1858 ceph_object_layout
OSDMap::make_object_layout(
1859 object_t oid
, int pg_pool
, string nspace
) const
1861 object_locator_t
loc(pg_pool
, nspace
);
1863 ceph_object_layout ol
;
1864 pg_t pgid
= object_locator_to_pg(oid
, loc
);
1865 ol
.ol_pgid
= pgid
.get_old_pg().v
;
1866 ol
.ol_stripe_unit
= 0;
1870 void OSDMap::_remove_nonexistent_osds(const pg_pool_t
& pool
,
1871 vector
<int>& osds
) const
1873 if (pool
.can_shift_osds()) {
1874 unsigned removed
= 0;
1875 for (unsigned i
= 0; i
< osds
.size(); i
++) {
1876 if (!exists(osds
[i
])) {
1881 osds
[i
- removed
] = osds
[i
];
1885 osds
.resize(osds
.size() - removed
);
1887 for (auto& osd
: osds
) {
1889 osd
= CRUSH_ITEM_NONE
;
1894 void OSDMap::_pg_to_raw_osds(
1895 const pg_pool_t
& pool
, pg_t pg
,
1900 ps_t pps
= pool
.raw_pg_to_pps(pg
); // placement ps
1901 unsigned size
= pool
.get_size();
1904 int ruleno
= crush
->find_rule(pool
.get_crush_rule(), pool
.get_type(), size
);
1906 crush
->do_rule(ruleno
, pps
, *osds
, size
, osd_weight
, pg
.pool());
1908 _remove_nonexistent_osds(pool
, *osds
);
1914 int OSDMap::_pick_primary(const vector
<int>& osds
) const
1916 for (auto osd
: osds
) {
1917 if (osd
!= CRUSH_ITEM_NONE
) {
1924 void OSDMap::_apply_upmap(const pg_pool_t
& pi
, pg_t raw_pg
, vector
<int> *raw
) const
1926 pg_t pg
= pi
.raw_pg_to_pg(raw_pg
);
1927 auto p
= pg_upmap
.find(pg
);
1928 if (p
!= pg_upmap
.end()) {
1929 // make sure targets aren't marked out
1930 for (auto osd
: p
->second
) {
1931 if (osd
!= CRUSH_ITEM_NONE
&& osd
< max_osd
&& osd_weight
[osd
] == 0) {
1932 // reject/ignore the explicit mapping
1936 *raw
= vector
<int>(p
->second
.begin(), p
->second
.end());
1937 // continue to check and apply pg_upmap_items if any
1940 auto q
= pg_upmap_items
.find(pg
);
1941 if (q
!= pg_upmap_items
.end()) {
1942 for (auto& i
: *raw
) {
1943 for (auto& r
: q
->second
) {
1947 if (!(r
.second
!= CRUSH_ITEM_NONE
&&
1948 r
.second
< max_osd
&&
1949 osd_weight
[r
.second
] == 0)) {
1958 // pg -> (up osd list)
1959 void OSDMap::_raw_to_up_osds(const pg_pool_t
& pool
, const vector
<int>& raw
,
1960 vector
<int> *up
) const
1962 if (pool
.can_shift_osds()) {
1965 up
->reserve(raw
.size());
1966 for (unsigned i
=0; i
<raw
.size(); i
++) {
1967 if (!exists(raw
[i
]) || is_down(raw
[i
]))
1969 up
->push_back(raw
[i
]);
1972 // set down/dne devices to NONE
1973 up
->resize(raw
.size());
1974 for (int i
= raw
.size() - 1; i
>= 0; --i
) {
1975 if (!exists(raw
[i
]) || is_down(raw
[i
])) {
1976 (*up
)[i
] = CRUSH_ITEM_NONE
;
1984 void OSDMap::_apply_primary_affinity(ps_t seed
,
1985 const pg_pool_t
& pool
,
1989 // do we have any non-default primary_affinity values for these osds?
1990 if (!osd_primary_affinity
)
1994 for (const auto osd
: *osds
) {
1995 if (osd
!= CRUSH_ITEM_NONE
&&
1996 (*osd_primary_affinity
)[osd
] != CEPH_OSD_DEFAULT_PRIMARY_AFFINITY
) {
2004 // pick the primary. feed both the seed (for the pg) and the osd
2005 // into the hash/rng so that a proportional fraction of an osd's pgs
2006 // get rejected as primary.
2008 for (unsigned i
= 0; i
< osds
->size(); ++i
) {
2010 if (o
== CRUSH_ITEM_NONE
)
2012 unsigned a
= (*osd_primary_affinity
)[o
];
2013 if (a
< CEPH_OSD_MAX_PRIMARY_AFFINITY
&&
2014 (crush_hash32_2(CRUSH_HASH_RJENKINS1
,
2015 seed
, o
) >> 16) >= a
) {
2016 // we chose not to use this primary. note it anyway as a
2017 // fallback in case we don't pick anyone else, but keep looking.
2028 *primary
= (*osds
)[pos
];
2030 if (pool
.can_shift_osds() && pos
> 0) {
2031 // move the new primary to the front.
2032 for (int i
= pos
; i
> 0; --i
) {
2033 (*osds
)[i
] = (*osds
)[i
-1];
2035 (*osds
)[0] = *primary
;
2039 void OSDMap::_get_temp_osds(const pg_pool_t
& pool
, pg_t pg
,
2040 vector
<int> *temp_pg
, int *temp_primary
) const
2042 pg
= pool
.raw_pg_to_pg(pg
);
2043 const auto p
= pg_temp
->find(pg
);
2045 if (p
!= pg_temp
->end()) {
2046 for (unsigned i
=0; i
<p
->second
.size(); i
++) {
2047 if (!exists(p
->second
[i
]) || is_down(p
->second
[i
])) {
2048 if (pool
.can_shift_osds()) {
2051 temp_pg
->push_back(CRUSH_ITEM_NONE
);
2054 temp_pg
->push_back(p
->second
[i
]);
2058 const auto &pp
= primary_temp
->find(pg
);
2060 if (pp
!= primary_temp
->end()) {
2061 *temp_primary
= pp
->second
;
2062 } else if (!temp_pg
->empty()) { // apply pg_temp's primary
2063 for (unsigned i
= 0; i
< temp_pg
->size(); ++i
) {
2064 if ((*temp_pg
)[i
] != CRUSH_ITEM_NONE
) {
2065 *temp_primary
= (*temp_pg
)[i
];
2072 void OSDMap::pg_to_raw_osds(pg_t pg
, vector
<int> *raw
, int *primary
) const
2076 const pg_pool_t
*pool
= get_pg_pool(pg
.pool());
2079 _pg_to_raw_osds(*pool
, pg
, raw
, NULL
);
2081 *primary
= _pick_primary(*raw
);
2084 void OSDMap::pg_to_raw_up(pg_t pg
, vector
<int> *up
, int *primary
) const
2086 const pg_pool_t
*pool
= get_pg_pool(pg
.pool());
2096 _pg_to_raw_osds(*pool
, pg
, &raw
, &pps
);
2097 _apply_upmap(*pool
, pg
, &raw
);
2098 _raw_to_up_osds(*pool
, raw
, up
);
2099 *primary
= _pick_primary(raw
);
2100 _apply_primary_affinity(pps
, *pool
, up
, primary
);
2103 void OSDMap::_pg_to_up_acting_osds(
2104 const pg_t
& pg
, vector
<int> *up
, int *up_primary
,
2105 vector
<int> *acting
, int *acting_primary
,
2106 bool raw_pg_to_pg
) const
2108 const pg_pool_t
*pool
= get_pg_pool(pg
.pool());
2110 (!raw_pg_to_pg
&& pg
.ps() >= pool
->get_pg_num())) {
2118 *acting_primary
= -1;
2123 vector
<int> _acting
;
2125 int _acting_primary
;
2127 _get_temp_osds(*pool
, pg
, &_acting
, &_acting_primary
);
2128 if (_acting
.empty() || up
|| up_primary
) {
2129 _pg_to_raw_osds(*pool
, pg
, &raw
, &pps
);
2130 _apply_upmap(*pool
, pg
, &raw
);
2131 _raw_to_up_osds(*pool
, raw
, &_up
);
2132 _up_primary
= _pick_primary(_up
);
2133 _apply_primary_affinity(pps
, *pool
, &_up
, &_up_primary
);
2134 if (_acting
.empty()) {
2136 if (_acting_primary
== -1) {
2137 _acting_primary
= _up_primary
;
2144 *up_primary
= _up_primary
;
2148 acting
->swap(_acting
);
2150 *acting_primary
= _acting_primary
;
2153 int OSDMap::calc_pg_rank(int osd
, const vector
<int>& acting
, int nrep
)
2156 nrep
= acting
.size();
2157 for (int i
=0; i
<nrep
; i
++)
2158 if (acting
[i
] == osd
)
2163 int OSDMap::calc_pg_role(int osd
, const vector
<int>& acting
, int nrep
)
2165 return calc_pg_rank(osd
, acting
, nrep
);
2168 bool OSDMap::primary_changed(
2170 const vector
<int> &oldacting
,
2172 const vector
<int> &newacting
)
2174 if (oldacting
.empty() && newacting
.empty())
2175 return false; // both still empty
2176 if (oldacting
.empty() ^ newacting
.empty())
2177 return true; // was empty, now not, or vice versa
2178 if (oldprimary
!= newprimary
)
2179 return true; // primary changed
2180 if (calc_pg_rank(oldprimary
, oldacting
) !=
2181 calc_pg_rank(newprimary
, newacting
))
2183 return false; // same primary (tho replicas may have changed)
2187 // serialize, unserialize
2188 void OSDMap::encode_client_old(bufferlist
& bl
) const
2195 ::encode(epoch
, bl
);
2196 ::encode(created
, bl
);
2197 ::encode(modified
, bl
);
2199 // for ::encode(pools, bl);
2200 __u32 n
= pools
.size();
2203 for (const auto &pool
: pools
) {
2206 ::encode(pool
.second
, bl
, 0);
2208 // for ::encode(pool_name, bl);
2209 n
= pool_name
.size();
2211 for (const auto &pname
: pool_name
) {
2214 ::encode(pname
.second
, bl
);
2216 // for ::encode(pool_max, bl);
2220 ::encode(flags
, bl
);
2222 ::encode(max_osd
, bl
);
2224 uint32_t n
= osd_state
.size();
2226 for (auto s
: osd_state
) {
2227 ::encode((uint8_t)s
, bl
);
2230 ::encode(osd_weight
, bl
);
2231 ::encode(osd_addrs
->client_addr
, bl
, 0);
2233 // for ::encode(pg_temp, bl);
2234 n
= pg_temp
->size();
2236 for (const auto pg
: *pg_temp
) {
2237 old_pg_t opg
= pg
.first
.get_old_pg();
2239 ::encode(pg
.second
, bl
);
2244 crush
->encode(cbl
, 0 /* legacy (no) features */);
2248 void OSDMap::encode_classic(bufferlist
& bl
, uint64_t features
) const
2250 if ((features
& CEPH_FEATURE_PGID64
) == 0) {
2251 encode_client_old(bl
);
2260 ::encode(epoch
, bl
);
2261 ::encode(created
, bl
);
2262 ::encode(modified
, bl
);
2264 ::encode(pools
, bl
, features
);
2265 ::encode(pool_name
, bl
);
2266 ::encode(pool_max
, bl
);
2268 ::encode(flags
, bl
);
2270 ::encode(max_osd
, bl
);
2272 uint32_t n
= osd_state
.size();
2274 for (auto s
: osd_state
) {
2275 ::encode((uint8_t)s
, bl
);
2278 ::encode(osd_weight
, bl
);
2279 ::encode(osd_addrs
->client_addr
, bl
, features
);
2281 ::encode(*pg_temp
, bl
);
2285 crush
->encode(cbl
, 0 /* legacy (no) features */);
2291 ::encode(osd_addrs
->hb_back_addr
, bl
, features
);
2292 ::encode(osd_info
, bl
);
2293 ::encode(blacklist
, bl
, features
);
2294 ::encode(osd_addrs
->cluster_addr
, bl
, features
);
2295 ::encode(cluster_snapshot_epoch
, bl
);
2296 ::encode(cluster_snapshot
, bl
);
2297 ::encode(*osd_uuid
, bl
);
2298 ::encode(osd_xinfo
, bl
);
2299 ::encode(osd_addrs
->hb_front_addr
, bl
, features
);
2302 void OSDMap::encode(bufferlist
& bl
, uint64_t features
) const
2304 if ((features
& CEPH_FEATURE_OSDMAP_ENC
) == 0) {
2305 encode_classic(bl
, features
);
2309 // only a select set of callers should *ever* be encoding new
2310 // OSDMaps. others should be passing around the canonical encoded
2311 // buffers from on high. select out those callers by passing in an
2312 // "impossible" feature bit.
2313 assert(features
& CEPH_FEATURE_RESERVED
);
2314 features
&= ~CEPH_FEATURE_RESERVED
;
2316 size_t start_offset
= bl
.length();
2318 buffer::list::iterator crc_it
;
2320 // meta-encoding: how we include client-used and osd-specific data
2321 ENCODE_START(8, 7, bl
);
2325 if (!HAVE_FEATURE(features
, SERVER_LUMINOUS
)) {
2328 ENCODE_START(v
, 1, bl
); // client-usable data
2331 ::encode(epoch
, bl
);
2332 ::encode(created
, bl
);
2333 ::encode(modified
, bl
);
2335 ::encode(pools
, bl
, features
);
2336 ::encode(pool_name
, bl
);
2337 ::encode(pool_max
, bl
);
2340 decltype(flags
) f
= flags
;
2341 if (require_osd_release
>= CEPH_RELEASE_LUMINOUS
)
2342 f
|= CEPH_OSDMAP_REQUIRE_LUMINOUS
| CEPH_OSDMAP_RECOVERY_DELETES
;
2343 else if (require_osd_release
== CEPH_RELEASE_KRAKEN
)
2344 f
|= CEPH_OSDMAP_REQUIRE_KRAKEN
;
2345 else if (require_osd_release
== CEPH_RELEASE_JEWEL
)
2346 f
|= CEPH_OSDMAP_REQUIRE_JEWEL
;
2349 ::encode(flags
, bl
);
2352 ::encode(max_osd
, bl
);
2354 ::encode(osd_state
, bl
);
2356 uint32_t n
= osd_state
.size();
2358 for (auto s
: osd_state
) {
2359 ::encode((uint8_t)s
, bl
);
2362 ::encode(osd_weight
, bl
);
2363 ::encode(osd_addrs
->client_addr
, bl
, features
);
2365 ::encode(*pg_temp
, bl
);
2366 ::encode(*primary_temp
, bl
);
2367 if (osd_primary_affinity
) {
2368 ::encode(*osd_primary_affinity
, bl
);
2376 crush
->encode(cbl
, features
);
2378 ::encode(erasure_code_profiles
, bl
);
2381 ::encode(pg_upmap
, bl
);
2382 ::encode(pg_upmap_items
, bl
);
2384 assert(pg_upmap
.empty());
2385 assert(pg_upmap_items
.empty());
2388 ::encode(crush_version
, bl
);
2390 ENCODE_FINISH(bl
); // client-usable data
2394 uint8_t target_v
= 5;
2395 if (!HAVE_FEATURE(features
, SERVER_LUMINOUS
)) {
2398 ENCODE_START(target_v
, 1, bl
); // extended, osd-only data
2399 ::encode(osd_addrs
->hb_back_addr
, bl
, features
);
2400 ::encode(osd_info
, bl
);
2402 // put this in a sorted, ordered map<> so that we encode in a
2403 // deterministic order.
2404 map
<entity_addr_t
,utime_t
> blacklist_map
;
2405 for (const auto &addr
: blacklist
)
2406 blacklist_map
.insert(make_pair(addr
.first
, addr
.second
));
2407 ::encode(blacklist_map
, bl
, features
);
2409 ::encode(osd_addrs
->cluster_addr
, bl
, features
);
2410 ::encode(cluster_snapshot_epoch
, bl
);
2411 ::encode(cluster_snapshot
, bl
);
2412 ::encode(*osd_uuid
, bl
);
2413 ::encode(osd_xinfo
, bl
);
2414 ::encode(osd_addrs
->hb_front_addr
, bl
, features
);
2415 if (target_v
>= 2) {
2416 ::encode(nearfull_ratio
, bl
);
2417 ::encode(full_ratio
, bl
);
2418 ::encode(backfillfull_ratio
, bl
);
2420 // 4 was string-based new_require_min_compat_client
2421 if (target_v
>= 5) {
2422 ::encode(require_min_compat_client
, bl
);
2423 ::encode(require_osd_release
, bl
);
2425 ENCODE_FINISH(bl
); // osd-only data
2428 ::encode((uint32_t)0, bl
); // dummy crc
2431 tail_offset
= bl
.length();
2433 ENCODE_FINISH(bl
); // meta-encoding wrapper
2437 front
.substr_of(bl
, start_offset
, crc_it
.get_off() - start_offset
);
2438 crc
= front
.crc32c(-1);
2439 if (tail_offset
< bl
.length()) {
2441 tail
.substr_of(bl
, tail_offset
, bl
.length() - tail_offset
);
2442 crc
= tail
.crc32c(crc
);
2446 crc_it
.copy_in(4, (char*)&crc_le
);
2450 void OSDMap::decode(bufferlist
& bl
)
2452 auto p
= bl
.begin();
2456 void OSDMap::decode_classic(bufferlist::iterator
& p
)
2465 ::decode(created
, p
);
2466 ::decode(modified
, p
);
2470 int32_t max_pools
= 0;
2471 ::decode(max_pools
, p
);
2472 pool_max
= max_pools
;
2478 ::decode(pools
[t
], p
);
2483 } else if (v
== 5) {
2488 ::decode(pool_name
[t
], p
);
2495 ::decode(pool_name
, p
);
2496 ::decode(pool_max
, p
);
2498 // kludge around some old bug that zeroed out pool_max (#2307)
2499 if (pools
.size() && pool_max
< pools
.rbegin()->first
) {
2500 pool_max
= pools
.rbegin()->first
;
2505 ::decode(max_osd
, p
);
2509 osd_state
.resize(os
.size());
2510 for (unsigned i
= 0; i
< os
.size(); ++i
) {
2511 osd_state
[i
] = os
[i
];
2514 ::decode(osd_weight
, p
);
2515 ::decode(osd_addrs
->client_addr
, p
);
2521 ::decode_raw(opg
, p
);
2522 mempool::osdmap::vector
<int32_t> v
;
2524 pg_temp
->set(pg_t(opg
), v
);
2527 ::decode(*pg_temp
, p
);
2533 auto cblp
= cbl
.begin();
2534 crush
->decode(cblp
);
2540 ::decode(osd_addrs
->hb_back_addr
, p
);
2541 ::decode(osd_info
, p
);
2543 ::decode(pool_name
, p
);
2545 ::decode(blacklist
, p
);
2547 ::decode(osd_addrs
->cluster_addr
, p
);
2549 osd_addrs
->cluster_addr
.resize(osd_addrs
->client_addr
.size());
2552 ::decode(cluster_snapshot_epoch
, p
);
2553 ::decode(cluster_snapshot
, p
);
2557 ::decode(*osd_uuid
, p
);
2559 osd_uuid
->resize(max_osd
);
2562 ::decode(osd_xinfo
, p
);
2564 osd_xinfo
.resize(max_osd
);
2567 ::decode(osd_addrs
->hb_front_addr
, p
);
2569 osd_addrs
->hb_front_addr
.resize(osd_addrs
->hb_back_addr
.size());
2571 osd_primary_affinity
.reset();
2576 void OSDMap::decode(bufferlist::iterator
& bl
)
2579 * Older encodings of the OSDMap had a single struct_v which
2580 * covered the whole encoding, and was prior to our modern
2581 * stuff which includes a compatv and a size. So if we see
2582 * a struct_v < 7, we must rewind to the beginning and use our
2585 size_t start_offset
= bl
.get_off();
2586 size_t tail_offset
= 0;
2587 bufferlist crc_front
, crc_tail
;
2589 DECODE_START_LEGACY_COMPAT_LEN(8, 7, 7, bl
); // wrapper
2591 int struct_v_size
= sizeof(struct_v
);
2592 bl
.advance(-struct_v_size
);
2597 * Since we made it past that hurdle, we can use our normal paths.
2600 DECODE_START(6, bl
); // client-usable data
2603 ::decode(epoch
, bl
);
2604 ::decode(created
, bl
);
2605 ::decode(modified
, bl
);
2607 ::decode(pools
, bl
);
2608 ::decode(pool_name
, bl
);
2609 ::decode(pool_max
, bl
);
2611 ::decode(flags
, bl
);
2613 ::decode(max_osd
, bl
);
2614 if (struct_v
>= 5) {
2615 ::decode(osd_state
, bl
);
2619 osd_state
.resize(os
.size());
2620 for (unsigned i
= 0; i
< os
.size(); ++i
) {
2621 osd_state
[i
] = os
[i
];
2624 ::decode(osd_weight
, bl
);
2625 ::decode(osd_addrs
->client_addr
, bl
);
2627 ::decode(*pg_temp
, bl
);
2628 ::decode(*primary_temp
, bl
);
2629 if (struct_v
>= 2) {
2630 osd_primary_affinity
.reset(new mempool::osdmap::vector
<__u32
>);
2631 ::decode(*osd_primary_affinity
, bl
);
2632 if (osd_primary_affinity
->empty())
2633 osd_primary_affinity
.reset();
2635 osd_primary_affinity
.reset();
2641 auto cblp
= cbl
.begin();
2642 crush
->decode(cblp
);
2643 if (struct_v
>= 3) {
2644 ::decode(erasure_code_profiles
, bl
);
2646 erasure_code_profiles
.clear();
2648 if (struct_v
>= 4) {
2649 ::decode(pg_upmap
, bl
);
2650 ::decode(pg_upmap_items
, bl
);
2653 pg_upmap_items
.clear();
2655 if (struct_v
>= 6) {
2656 ::decode(crush_version
, bl
);
2658 DECODE_FINISH(bl
); // client-usable data
2662 DECODE_START(5, bl
); // extended, osd-only data
2663 ::decode(osd_addrs
->hb_back_addr
, bl
);
2664 ::decode(osd_info
, bl
);
2665 ::decode(blacklist
, bl
);
2666 ::decode(osd_addrs
->cluster_addr
, bl
);
2667 ::decode(cluster_snapshot_epoch
, bl
);
2668 ::decode(cluster_snapshot
, bl
);
2669 ::decode(*osd_uuid
, bl
);
2670 ::decode(osd_xinfo
, bl
);
2671 ::decode(osd_addrs
->hb_front_addr
, bl
);
2672 if (struct_v
>= 2) {
2673 ::decode(nearfull_ratio
, bl
);
2674 ::decode(full_ratio
, bl
);
2679 if (struct_v
>= 3) {
2680 ::decode(backfillfull_ratio
, bl
);
2682 backfillfull_ratio
= 0;
2684 if (struct_v
== 4) {
2688 require_min_compat_client
= ceph_release_from_name(r
.c_str());
2690 if (struct_v
>= 5) {
2691 ::decode(require_min_compat_client
, bl
);
2692 ::decode(require_osd_release
, bl
);
2693 if (require_osd_release
>= CEPH_RELEASE_LUMINOUS
) {
2694 flags
&= ~(CEPH_OSDMAP_LEGACY_REQUIRE_FLAGS
);
2695 flags
|= CEPH_OSDMAP_RECOVERY_DELETES
;
2698 if (flags
& CEPH_OSDMAP_REQUIRE_LUMINOUS
) {
2699 // only for compat with post-kraken pre-luminous test clusters
2700 require_osd_release
= CEPH_RELEASE_LUMINOUS
;
2701 flags
&= ~(CEPH_OSDMAP_LEGACY_REQUIRE_FLAGS
);
2702 flags
|= CEPH_OSDMAP_RECOVERY_DELETES
;
2703 } else if (flags
& CEPH_OSDMAP_REQUIRE_KRAKEN
) {
2704 require_osd_release
= CEPH_RELEASE_KRAKEN
;
2705 } else if (flags
& CEPH_OSDMAP_REQUIRE_JEWEL
) {
2706 require_osd_release
= CEPH_RELEASE_JEWEL
;
2708 require_osd_release
= 0;
2711 DECODE_FINISH(bl
); // osd-only data
2714 if (struct_v
>= 8) {
2715 crc_front
.substr_of(bl
.get_bl(), start_offset
, bl
.get_off() - start_offset
);
2717 tail_offset
= bl
.get_off();
2720 crc_defined
= false;
2724 DECODE_FINISH(bl
); // wrapper
2728 uint32_t actual
= crc_front
.crc32c(-1);
2729 if (tail_offset
< bl
.get_off()) {
2731 tail
.substr_of(bl
.get_bl(), tail_offset
, bl
.get_off() - tail_offset
);
2732 actual
= tail
.crc32c(actual
);
2734 if (crc
!= actual
) {
2736 ss
<< "bad crc, actual " << actual
<< " != expected " << crc
;
2737 string s
= ss
.str();
2738 throw buffer::malformed_input(s
.c_str());
2745 void OSDMap::post_decode()
2749 for (const auto &pname
: pool_name
) {
2750 name_pool
[pname
.second
] = pname
.first
;
2754 _calc_up_osd_features();
2757 void OSDMap::dump_erasure_code_profiles(
2758 const mempool::osdmap::map
<string
,map
<string
,string
>>& profiles
,
2761 f
->open_object_section("erasure_code_profiles");
2762 for (const auto &profile
: profiles
) {
2763 f
->open_object_section(profile
.first
.c_str());
2764 for (const auto &profm
: profile
.second
) {
2765 f
->dump_string(profm
.first
.c_str(), profm
.second
.c_str());
2772 void OSDMap::dump(Formatter
*f
) const
2774 f
->dump_int("epoch", get_epoch());
2775 f
->dump_stream("fsid") << get_fsid();
2776 f
->dump_stream("created") << get_created();
2777 f
->dump_stream("modified") << get_modified();
2778 f
->dump_string("flags", get_flag_string());
2779 f
->dump_unsigned("crush_version", get_crush_version());
2780 f
->dump_float("full_ratio", full_ratio
);
2781 f
->dump_float("backfillfull_ratio", backfillfull_ratio
);
2782 f
->dump_float("nearfull_ratio", nearfull_ratio
);
2783 f
->dump_string("cluster_snapshot", get_cluster_snapshot());
2784 f
->dump_int("pool_max", get_pool_max());
2785 f
->dump_int("max_osd", get_max_osd());
2786 f
->dump_string("require_min_compat_client",
2787 ceph_release_name(require_min_compat_client
));
2788 f
->dump_string("min_compat_client",
2789 ceph_release_name(get_min_compat_client()));
2790 f
->dump_string("require_osd_release",
2791 ceph_release_name(require_osd_release
));
2793 f
->open_array_section("pools");
2794 for (const auto &pool
: pools
) {
2795 std::string
name("<unknown>");
2796 const auto &pni
= pool_name
.find(pool
.first
);
2797 if (pni
!= pool_name
.end())
2799 f
->open_object_section("pool");
2800 f
->dump_int("pool", pool
.first
);
2801 f
->dump_string("pool_name", name
);
2802 pool
.second
.dump(f
);
2807 f
->open_array_section("osds");
2808 for (int i
=0; i
<get_max_osd(); i
++)
2810 f
->open_object_section("osd_info");
2811 f
->dump_int("osd", i
);
2812 f
->dump_stream("uuid") << get_uuid(i
);
2813 f
->dump_int("up", is_up(i
));
2814 f
->dump_int("in", is_in(i
));
2815 f
->dump_float("weight", get_weightf(i
));
2816 f
->dump_float("primary_affinity", get_primary_affinityf(i
));
2817 get_info(i
).dump(f
);
2818 f
->dump_stream("public_addr") << get_addr(i
);
2819 f
->dump_stream("cluster_addr") << get_cluster_addr(i
);
2820 f
->dump_stream("heartbeat_back_addr") << get_hb_back_addr(i
);
2821 f
->dump_stream("heartbeat_front_addr") << get_hb_front_addr(i
);
2825 f
->open_array_section("state");
2826 for (const auto &state
: st
)
2827 f
->dump_string("state", state
);
2834 f
->open_array_section("osd_xinfo");
2835 for (int i
=0; i
<get_max_osd(); i
++) {
2837 f
->open_object_section("xinfo");
2838 f
->dump_int("osd", i
);
2839 osd_xinfo
[i
].dump(f
);
2845 f
->open_array_section("pg_upmap");
2846 for (auto& p
: pg_upmap
) {
2847 f
->open_object_section("mapping");
2848 f
->dump_stream("pgid") << p
.first
;
2849 f
->open_array_section("osds");
2850 for (auto q
: p
.second
) {
2851 f
->dump_int("osd", q
);
2857 f
->open_array_section("pg_upmap_items");
2858 for (auto& p
: pg_upmap_items
) {
2859 f
->open_object_section("mapping");
2860 f
->dump_stream("pgid") << p
.first
;
2861 f
->open_array_section("mappings");
2862 for (auto& q
: p
.second
) {
2863 f
->open_object_section("mapping");
2864 f
->dump_int("from", q
.first
);
2865 f
->dump_int("to", q
.second
);
2872 f
->open_array_section("pg_temp");
2876 f
->open_array_section("primary_temp");
2877 for (const auto &pg
: *primary_temp
) {
2878 f
->dump_stream("pgid") << pg
.first
;
2879 f
->dump_int("osd", pg
.second
);
2881 f
->close_section(); // primary_temp
2883 f
->open_object_section("blacklist");
2884 for (const auto &addr
: blacklist
) {
2887 f
->dump_stream(ss
.str().c_str()) << addr
.second
;
2891 dump_erasure_code_profiles(erasure_code_profiles
, f
);
2894 void OSDMap::generate_test_instances(list
<OSDMap
*>& o
)
2896 o
.push_back(new OSDMap
);
2898 CephContext
*cct
= new CephContext(CODE_ENVIRONMENT_UTILITY
);
2899 o
.push_back(new OSDMap
);
2901 o
.back()->build_simple(cct
, 1, fsid
, 16);
2902 o
.back()->created
= o
.back()->modified
= utime_t(1, 2); // fix timestamp
2903 o
.back()->blacklist
[entity_addr_t()] = utime_t(5, 6);
2907 string
OSDMap::get_flag_string(unsigned f
)
2910 if ( f
& CEPH_OSDMAP_NEARFULL
)
2912 if (f
& CEPH_OSDMAP_FULL
)
2914 if (f
& CEPH_OSDMAP_PAUSERD
)
2916 if (f
& CEPH_OSDMAP_PAUSEWR
)
2918 if (f
& CEPH_OSDMAP_PAUSEREC
)
2920 if (f
& CEPH_OSDMAP_NOUP
)
2922 if (f
& CEPH_OSDMAP_NODOWN
)
2924 if (f
& CEPH_OSDMAP_NOOUT
)
2926 if (f
& CEPH_OSDMAP_NOIN
)
2928 if (f
& CEPH_OSDMAP_NOBACKFILL
)
2930 if (f
& CEPH_OSDMAP_NOREBALANCE
)
2931 s
+= ",norebalance";
2932 if (f
& CEPH_OSDMAP_NORECOVER
)
2934 if (f
& CEPH_OSDMAP_NOSCRUB
)
2936 if (f
& CEPH_OSDMAP_NODEEP_SCRUB
)
2937 s
+= ",nodeep-scrub";
2938 if (f
& CEPH_OSDMAP_NOTIERAGENT
)
2939 s
+= ",notieragent";
2940 if (f
& CEPH_OSDMAP_SORTBITWISE
)
2941 s
+= ",sortbitwise";
2942 if (f
& CEPH_OSDMAP_REQUIRE_JEWEL
)
2943 s
+= ",require_jewel_osds";
2944 if (f
& CEPH_OSDMAP_REQUIRE_KRAKEN
)
2945 s
+= ",require_kraken_osds";
2946 if (f
& CEPH_OSDMAP_REQUIRE_LUMINOUS
)
2947 s
+= ",require_luminous_osds";
2948 if (f
& CEPH_OSDMAP_RECOVERY_DELETES
)
2949 s
+= ",recovery_deletes";
2955 string
OSDMap::get_flag_string() const
2957 return get_flag_string(flags
);
2960 void OSDMap::print_pools(ostream
& out
) const
2962 for (const auto &pool
: pools
) {
2963 std::string
name("<unknown>");
2964 const auto &pni
= pool_name
.find(pool
.first
);
2965 if (pni
!= pool_name
.end())
2967 out
<< "pool " << pool
.first
2969 << "' " << pool
.second
<< "\n";
2971 for (const auto &snap
: pool
.second
.snaps
)
2972 out
<< "\tsnap " << snap
.second
.snapid
<< " '" << snap
.second
.name
<< "' " << snap
.second
.stamp
<< "\n";
2974 if (!pool
.second
.removed_snaps
.empty())
2975 out
<< "\tremoved_snaps " << pool
.second
.removed_snaps
<< "\n";
2980 void OSDMap::print(ostream
& out
) const
2982 out
<< "epoch " << get_epoch() << "\n"
2983 << "fsid " << get_fsid() << "\n"
2984 << "created " << get_created() << "\n"
2985 << "modified " << get_modified() << "\n";
2987 out
<< "flags " << get_flag_string() << "\n";
2988 out
<< "crush_version " << get_crush_version() << "\n";
2989 out
<< "full_ratio " << full_ratio
<< "\n";
2990 out
<< "backfillfull_ratio " << backfillfull_ratio
<< "\n";
2991 out
<< "nearfull_ratio " << nearfull_ratio
<< "\n";
2992 if (require_min_compat_client
> 0) {
2993 out
<< "require_min_compat_client "
2994 << ceph_release_name(require_min_compat_client
) << "\n";
2996 out
<< "min_compat_client " << ceph_release_name(get_min_compat_client())
2998 if (require_osd_release
> 0) {
2999 out
<< "require_osd_release " << ceph_release_name(require_osd_release
)
3002 if (get_cluster_snapshot().length())
3003 out
<< "cluster_snapshot " << get_cluster_snapshot() << "\n";
3008 out
<< "max_osd " << get_max_osd() << "\n";
3009 for (int i
=0; i
<get_max_osd(); i
++) {
3012 out
<< (is_up(i
) ? " up ":" down");
3013 out
<< (is_in(i
) ? " in ":" out");
3014 out
<< " weight " << get_weightf(i
);
3015 if (get_primary_affinity(i
) != CEPH_OSD_DEFAULT_PRIMARY_AFFINITY
)
3016 out
<< " primary_affinity " << get_primary_affinityf(i
);
3017 const osd_info_t
& info(get_info(i
));
3019 out
<< " " << get_addr(i
) << " " << get_cluster_addr(i
) << " " << get_hb_back_addr(i
)
3020 << " " << get_hb_front_addr(i
);
3024 if (!get_uuid(i
).is_zero())
3025 out
<< " " << get_uuid(i
);
3031 for (auto& p
: pg_upmap
) {
3032 out
<< "pg_upmap " << p
.first
<< " " << p
.second
<< "\n";
3034 for (auto& p
: pg_upmap_items
) {
3035 out
<< "pg_upmap_items " << p
.first
<< " " << p
.second
<< "\n";
3038 for (const auto pg
: *pg_temp
)
3039 out
<< "pg_temp " << pg
.first
<< " " << pg
.second
<< "\n";
3041 for (const auto pg
: *primary_temp
)
3042 out
<< "primary_temp " << pg
.first
<< " " << pg
.second
<< "\n";
3044 for (const auto &addr
: blacklist
)
3045 out
<< "blacklist " << addr
.first
<< " expires " << addr
.second
<< "\n";
3047 // ignore pg_swap_primary
3050 class OSDTreePlainDumper
: public CrushTreeDumper::Dumper
<TextTable
> {
3052 typedef CrushTreeDumper::Dumper
<TextTable
> Parent
;
3054 OSDTreePlainDumper(const CrushWrapper
*crush
, const OSDMap
*osdmap_
,
3056 : Parent(crush
, osdmap_
->get_pool_names()), osdmap(osdmap_
), filter(f
) { }
3058 bool should_dump_leaf(int i
) const override
{
3060 return true; // normal case
3062 if (((filter
& OSDMap::DUMP_UP
) && osdmap
->is_up(i
)) ||
3063 ((filter
& OSDMap::DUMP_DOWN
) && osdmap
->is_down(i
)) ||
3064 ((filter
& OSDMap::DUMP_IN
) && osdmap
->is_in(i
)) ||
3065 ((filter
& OSDMap::DUMP_OUT
) && osdmap
->is_out(i
)) ||
3066 ((filter
& OSDMap::DUMP_DESTROYED
) && osdmap
->is_destroyed(i
))) {
3072 bool should_dump_empty_bucket() const override
{
3076 void dump(TextTable
*tbl
) {
3077 tbl
->define_column("ID", TextTable::LEFT
, TextTable::RIGHT
);
3078 tbl
->define_column("CLASS", TextTable::LEFT
, TextTable::RIGHT
);
3079 tbl
->define_column("WEIGHT", TextTable::LEFT
, TextTable::RIGHT
);
3080 tbl
->define_column("TYPE NAME", TextTable::LEFT
, TextTable::LEFT
);
3081 tbl
->define_column("STATUS", TextTable::LEFT
, TextTable::RIGHT
);
3082 tbl
->define_column("REWEIGHT", TextTable::LEFT
, TextTable::RIGHT
);
3083 tbl
->define_column("PRI-AFF", TextTable::LEFT
, TextTable::RIGHT
);
3087 for (int i
= 0; i
< osdmap
->get_max_osd(); i
++) {
3088 if (osdmap
->exists(i
) && !is_touched(i
) && should_dump_leaf(i
)) {
3089 dump_item(CrushTreeDumper::Item(i
, 0, 0, 0), tbl
);
3095 void dump_item(const CrushTreeDumper::Item
&qi
, TextTable
*tbl
) override
{
3096 const char *c
= crush
->get_item_class(qi
.id
);
3101 << weightf_t(qi
.weight
);
3104 for (int k
= 0; k
< qi
.depth
; k
++)
3106 if (qi
.is_bucket()) {
3107 name
<< crush
->get_type_name(crush
->get_bucket_type(qi
.id
)) << " "
3108 << crush
->get_item_name(qi
.id
);
3110 name
<< "osd." << qi
.id
;
3114 if (!qi
.is_bucket()) {
3115 if (!osdmap
->exists(qi
.id
)) {
3120 if (osdmap
->is_up(qi
.id
)) {
3122 } else if (osdmap
->is_destroyed(qi
.id
)) {
3128 << weightf_t(osdmap
->get_weightf(qi
.id
))
3129 << weightf_t(osdmap
->get_primary_affinityf(qi
.id
));
3132 *tbl
<< TextTable::endrow
;
3136 const OSDMap
*osdmap
;
3137 const unsigned filter
;
3140 class OSDTreeFormattingDumper
: public CrushTreeDumper::FormattingDumper
{
3142 typedef CrushTreeDumper::FormattingDumper Parent
;
3144 OSDTreeFormattingDumper(const CrushWrapper
*crush
, const OSDMap
*osdmap_
,
3146 : Parent(crush
, osdmap_
->get_pool_names()), osdmap(osdmap_
), filter(f
) { }
3148 bool should_dump_leaf(int i
) const override
{
3150 return true; // normal case
3152 if (((filter
& OSDMap::DUMP_UP
) && osdmap
->is_up(i
)) ||
3153 ((filter
& OSDMap::DUMP_DOWN
) && osdmap
->is_down(i
)) ||
3154 ((filter
& OSDMap::DUMP_IN
) && osdmap
->is_in(i
)) ||
3155 ((filter
& OSDMap::DUMP_OUT
) && osdmap
->is_out(i
)) ||
3156 ((filter
& OSDMap::DUMP_DESTROYED
) && osdmap
->is_destroyed(i
))) {
3162 bool should_dump_empty_bucket() const override
{
3166 void dump(Formatter
*f
) {
3167 f
->open_array_section("nodes");
3170 f
->open_array_section("stray");
3171 for (int i
= 0; i
< osdmap
->get_max_osd(); i
++) {
3172 if (osdmap
->exists(i
) && !is_touched(i
) && should_dump_leaf(i
))
3173 dump_item(CrushTreeDumper::Item(i
, 0, 0, 0), f
);
3179 void dump_item_fields(const CrushTreeDumper::Item
&qi
, Formatter
*f
) override
{
3180 Parent::dump_item_fields(qi
, f
);
3181 if (!qi
.is_bucket())
3184 if (osdmap
->is_up(qi
.id
)) {
3186 } else if (osdmap
->is_destroyed(qi
.id
)) {
3191 f
->dump_unsigned("exists", (int)osdmap
->exists(qi
.id
));
3192 f
->dump_string("status", s
);
3193 f
->dump_float("reweight", osdmap
->get_weightf(qi
.id
));
3194 f
->dump_float("primary_affinity", osdmap
->get_primary_affinityf(qi
.id
));
3199 const OSDMap
*osdmap
;
3200 const unsigned filter
;
3203 void OSDMap::print_tree(Formatter
*f
, ostream
*out
, unsigned filter
) const
3206 OSDTreeFormattingDumper(crush
.get(), this, filter
).dump(f
);
3210 OSDTreePlainDumper(crush
.get(), this, filter
).dump(&tbl
);
3215 void OSDMap::print_summary(Formatter
*f
, ostream
& out
,
3216 const string
& prefix
) const
3219 f
->open_object_section("osdmap");
3220 f
->dump_int("epoch", get_epoch());
3221 f
->dump_int("num_osds", get_num_osds());
3222 f
->dump_int("num_up_osds", get_num_up_osds());
3223 f
->dump_int("num_in_osds", get_num_in_osds());
3224 f
->dump_bool("full", test_flag(CEPH_OSDMAP_FULL
) ? true : false);
3225 f
->dump_bool("nearfull", test_flag(CEPH_OSDMAP_NEARFULL
) ? true : false);
3226 f
->dump_unsigned("num_remapped_pgs", get_num_pg_temp());
3229 out
<< get_num_osds() << " osds: "
3230 << get_num_up_osds() << " up, "
3231 << get_num_in_osds() << " in";
3232 if (get_num_pg_temp())
3233 out
<< "; " << get_num_pg_temp() << " remapped pgs";
3235 uint64_t important_flags
= flags
& ~CEPH_OSDMAP_SEMIHIDDEN_FLAGS
;
3236 if (important_flags
)
3237 out
<< prefix
<< "flags " << get_flag_string(important_flags
) << "\n";
3241 void OSDMap::print_oneline_summary(ostream
& out
) const
3243 out
<< "e" << get_epoch() << ": "
3244 << get_num_osds() << " total, "
3245 << get_num_up_osds() << " up, "
3246 << get_num_in_osds() << " in";
3247 if (test_flag(CEPH_OSDMAP_FULL
))
3249 else if (test_flag(CEPH_OSDMAP_NEARFULL
))
3253 bool OSDMap::crush_ruleset_in_use(int ruleset
) const
3255 for (const auto &pool
: pools
) {
3256 if (pool
.second
.crush_rule
== ruleset
)
3262 int OSDMap::build_simple_optioned(CephContext
*cct
, epoch_t e
, uuid_d
&fsid
,
3263 int nosd
, int pg_bits
, int pgp_bits
,
3266 ldout(cct
, 10) << "build_simple on " << nosd
3267 << " osds" << dendl
;
3270 created
= modified
= ceph_clock_now();
3277 const md_config_t
*conf
= cct
->_conf
;
3278 vector
<string
> sections
;
3279 conf
->get_all_sections(sections
);
3281 for (auto §ion
: sections
) {
3282 if (section
.find("osd.") != 0)
3285 const char *begin
= section
.c_str() + 4;
3286 char *end
= (char*)begin
;
3287 int o
= strtol(begin
, &end
, 10);
3291 if (o
> cct
->_conf
->mon_max_osd
) {
3292 lderr(cct
) << "[osd." << o
<< "] in config has id > mon_max_osd " << cct
->_conf
->mon_max_osd
<< dendl
;
3300 set_max_osd(maxosd
+ 1);
3307 r
= build_simple_crush_map(cct
, *crush
, nosd
, &ss
);
3309 r
= build_simple_crush_map_from_conf(cct
, *crush
, &ss
);
3312 int poolbase
= get_max_osd() ? get_max_osd() : 1;
3314 int const default_replicated_rule
= crush
->get_osd_pool_default_crush_replicated_ruleset(cct
);
3315 assert(default_replicated_rule
>= 0);
3318 // pgp_num <= pg_num
3319 if (pgp_bits
> pg_bits
)
3322 vector
<string
> pool_names
;
3323 pool_names
.push_back("rbd");
3324 for (auto &plname
: pool_names
) {
3325 int64_t pool
= ++pool_max
;
3326 pools
[pool
].type
= pg_pool_t::TYPE_REPLICATED
;
3327 pools
[pool
].flags
= cct
->_conf
->osd_pool_default_flags
;
3328 if (cct
->_conf
->osd_pool_default_flag_hashpspool
)
3329 pools
[pool
].set_flag(pg_pool_t::FLAG_HASHPSPOOL
);
3330 if (cct
->_conf
->osd_pool_default_flag_nodelete
)
3331 pools
[pool
].set_flag(pg_pool_t::FLAG_NODELETE
);
3332 if (cct
->_conf
->osd_pool_default_flag_nopgchange
)
3333 pools
[pool
].set_flag(pg_pool_t::FLAG_NOPGCHANGE
);
3334 if (cct
->_conf
->osd_pool_default_flag_nosizechange
)
3335 pools
[pool
].set_flag(pg_pool_t::FLAG_NOSIZECHANGE
);
3336 pools
[pool
].size
= cct
->_conf
->osd_pool_default_size
;
3337 pools
[pool
].min_size
= cct
->_conf
->get_osd_pool_default_min_size();
3338 pools
[pool
].crush_rule
= default_replicated_rule
;
3339 pools
[pool
].object_hash
= CEPH_STR_HASH_RJENKINS
;
3340 pools
[pool
].set_pg_num(poolbase
<< pg_bits
);
3341 pools
[pool
].set_pgp_num(poolbase
<< pgp_bits
);
3342 pools
[pool
].last_change
= epoch
;
3343 pools
[pool
].application_metadata
.insert(
3344 {pg_pool_t::APPLICATION_NAME_RBD
, {}});
3345 pool_name
[pool
] = plname
;
3346 name_pool
[plname
] = pool
;
3350 for (int i
=0; i
<get_max_osd(); i
++) {
3352 set_weight(i
, CEPH_OSD_OUT
);
3355 map
<string
,string
> profile_map
;
3356 r
= get_erasure_code_profile_default(cct
, profile_map
, &ss
);
3358 lderr(cct
) << ss
.str() << dendl
;
3361 set_erasure_code_profile("default", profile_map
);
3365 int OSDMap::get_erasure_code_profile_default(CephContext
*cct
,
3366 map
<string
,string
> &profile_map
,
3369 int r
= get_json_str_map(cct
->_conf
->osd_pool_default_erasure_code_profile
,
3375 int OSDMap::_build_crush_types(CrushWrapper
& crush
)
3377 crush
.set_type_name(0, "osd");
3378 crush
.set_type_name(1, "host");
3379 crush
.set_type_name(2, "chassis");
3380 crush
.set_type_name(3, "rack");
3381 crush
.set_type_name(4, "row");
3382 crush
.set_type_name(5, "pdu");
3383 crush
.set_type_name(6, "pod");
3384 crush
.set_type_name(7, "room");
3385 crush
.set_type_name(8, "datacenter");
3386 crush
.set_type_name(9, "region");
3387 crush
.set_type_name(10, "root");
3391 int OSDMap::build_simple_crush_map(CephContext
*cct
, CrushWrapper
& crush
,
3392 int nosd
, ostream
*ss
)
3397 int root_type
= _build_crush_types(crush
);
3399 int r
= crush
.add_bucket(0, 0, CRUSH_HASH_DEFAULT
,
3400 root_type
, 0, NULL
, NULL
, &rootid
);
3402 crush
.set_item_name(rootid
, "default");
3404 for (int o
=0; o
<nosd
; o
++) {
3405 map
<string
,string
> loc
;
3406 loc
["host"] = "localhost";
3407 loc
["rack"] = "localrack";
3408 loc
["root"] = "default";
3409 ldout(cct
, 10) << " adding osd." << o
<< " at " << loc
<< dendl
;
3411 snprintf(name
, sizeof(name
), "osd.%d", o
);
3412 crush
.insert_item(cct
, o
, 1.0, name
, loc
);
3415 build_simple_crush_rules(cct
, crush
, "default", ss
);
3422 int OSDMap::build_simple_crush_map_from_conf(CephContext
*cct
,
3423 CrushWrapper
& crush
,
3426 const md_config_t
*conf
= cct
->_conf
;
3431 int root_type
= _build_crush_types(crush
);
3433 int r
= crush
.add_bucket(0, 0,
3435 root_type
, 0, NULL
, NULL
, &rootid
);
3437 crush
.set_item_name(rootid
, "default");
3440 vector
<string
> sections
;
3441 conf
->get_all_sections(sections
);
3443 for (auto §ion
: sections
) {
3444 if (section
.find("osd.") != 0)
3447 const char *begin
= section
.c_str() + 4;
3448 char *end
= (char*)begin
;
3449 int o
= strtol(begin
, &end
, 10);
3453 string host
, rack
, row
, room
, dc
, pool
;
3454 vector
<string
> sectiontmp
;
3455 sectiontmp
.push_back("osd");
3456 sectiontmp
.push_back(section
);
3457 conf
->get_val_from_conf_file(sectiontmp
, "host", host
, false);
3458 conf
->get_val_from_conf_file(sectiontmp
, "rack", rack
, false);
3459 conf
->get_val_from_conf_file(sectiontmp
, "row", row
, false);
3460 conf
->get_val_from_conf_file(sectiontmp
, "room", room
, false);
3461 conf
->get_val_from_conf_file(sectiontmp
, "datacenter", dc
, false);
3462 conf
->get_val_from_conf_file(sectiontmp
, "root", pool
, false);
3464 if (host
.length() == 0)
3465 host
= "unknownhost";
3466 if (rack
.length() == 0)
3467 rack
= "unknownrack";
3469 map
<string
,string
> loc
;
3477 loc
["datacenter"] = dc
;
3478 loc
["root"] = "default";
3480 ldout(cct
, 5) << " adding osd." << o
<< " at " << loc
<< dendl
;
3481 crush
.insert_item(cct
, o
, 1.0, section
, loc
);
3484 build_simple_crush_rules(cct
, crush
, "default", ss
);
3492 int OSDMap::build_simple_crush_rules(
3494 CrushWrapper
& crush
,
3498 int crush_rule
= crush
.get_osd_pool_default_crush_replicated_ruleset(cct
);
3499 string failure_domain
=
3500 crush
.get_type_name(cct
->_conf
->osd_crush_chooseleaf_type
);
3503 r
= crush
.add_simple_rule_at(
3504 "replicated_rule", root
, failure_domain
, "",
3505 "firstn", pg_pool_t::TYPE_REPLICATED
,
3509 // do not add an erasure rule by default or else we will implicitly
3510 // require the crush_v2 feature of clients
3514 int OSDMap::summarize_mapping_stats(
3516 const set
<int64_t> *pools
,
3524 for (auto &p
: get_pools())
3528 unsigned total_pg
= 0;
3529 unsigned moved_pg
= 0;
3530 vector
<unsigned> base_by_osd(get_max_osd(), 0);
3531 vector
<unsigned> new_by_osd(get_max_osd(), 0);
3532 for (int64_t pool_id
: ls
) {
3533 const pg_pool_t
*pi
= get_pg_pool(pool_id
);
3534 vector
<int> up
, up2
;
3536 for (unsigned ps
= 0; ps
< pi
->get_pg_num(); ++ps
) {
3537 pg_t
pgid(ps
, pool_id
, -1);
3538 total_pg
+= pi
->get_size();
3539 pg_to_up_acting_osds(pgid
, &up
, &up_primary
, nullptr, nullptr);
3540 for (int osd
: up
) {
3541 if (osd
>= 0 && osd
< get_max_osd())
3545 newmap
->pg_to_up_acting_osds(pgid
, &up2
, &up_primary
, nullptr, nullptr);
3546 for (int osd
: up2
) {
3547 if (osd
>= 0 && osd
< get_max_osd())
3550 if (pi
->type
== pg_pool_t::TYPE_ERASURE
) {
3551 for (unsigned i
=0; i
<up
.size(); ++i
) {
3552 if (up
[i
] != up2
[i
]) {
3556 } else if (pi
->type
== pg_pool_t::TYPE_REPLICATED
) {
3557 for (int osd
: up
) {
3558 if (std::find(up2
.begin(), up2
.end(), osd
) == up2
.end()) {
3563 assert(0 == "unhandled pool type");
3569 unsigned num_up_in
= 0;
3570 for (int osd
= 0; osd
< get_max_osd(); ++osd
) {
3571 if (is_up(osd
) && is_in(osd
))
3578 float avg_pg
= (float)total_pg
/ (float)num_up_in
;
3579 float base_stddev
= 0, new_stddev
= 0;
3580 int min
= -1, max
= -1;
3581 unsigned min_base_pg
= 0, max_base_pg
= 0;
3582 unsigned min_new_pg
= 0, max_new_pg
= 0;
3583 for (int osd
= 0; osd
< get_max_osd(); ++osd
) {
3584 if (is_up(osd
) && is_in(osd
)) {
3585 float base_diff
= (float)base_by_osd
[osd
] - avg_pg
;
3586 base_stddev
+= base_diff
* base_diff
;
3587 float new_diff
= (float)new_by_osd
[osd
] - avg_pg
;
3588 new_stddev
+= new_diff
* new_diff
;
3589 if (min
< 0 || base_by_osd
[osd
] < min_base_pg
) {
3591 min_base_pg
= base_by_osd
[osd
];
3592 min_new_pg
= new_by_osd
[osd
];
3594 if (max
< 0 || base_by_osd
[osd
] > max_base_pg
) {
3596 max_base_pg
= base_by_osd
[osd
];
3597 max_new_pg
= new_by_osd
[osd
];
3601 base_stddev
= sqrt(base_stddev
/ num_up_in
);
3602 new_stddev
= sqrt(new_stddev
/ num_up_in
);
3604 float edev
= sqrt(avg_pg
* (1.0 - (1.0 / (double)num_up_in
)));
3608 f
->open_object_section("utilization");
3611 f
->dump_unsigned("moved_pgs", moved_pg
);
3612 f
->dump_unsigned("total_pgs", total_pg
);
3616 percent
= (float)moved_pg
* 100.0 / (float)total_pg
;
3617 ss
<< "moved " << moved_pg
<< " / " << total_pg
3618 << " (" << percent
<< "%)\n";
3622 f
->dump_float("avg_pgs", avg_pg
);
3623 f
->dump_float("std_dev", base_stddev
);
3624 f
->dump_float("expected_baseline_std_dev", edev
);
3626 f
->dump_float("new_std_dev", new_stddev
);
3628 ss
<< "avg " << avg_pg
<< "\n";
3629 ss
<< "stddev " << base_stddev
;
3631 ss
<< " -> " << new_stddev
;
3632 ss
<< " (expected baseline " << edev
<< ")\n";
3636 f
->dump_unsigned("min_osd", min
);
3637 f
->dump_unsigned("min_osd_pgs", min_base_pg
);
3639 f
->dump_unsigned("new_min_osd_pgs", min_new_pg
);
3641 ss
<< "min osd." << min
<< " with " << min_base_pg
;
3643 ss
<< " -> " << min_new_pg
;
3644 ss
<< " pgs (" << (float)min_base_pg
/ avg_pg
;
3646 ss
<< " -> " << (float)min_new_pg
/ avg_pg
;
3652 f
->dump_unsigned("max_osd", max
);
3653 f
->dump_unsigned("max_osd_pgs", max_base_pg
);
3655 f
->dump_unsigned("new_max_osd_pgs", max_new_pg
);
3657 ss
<< "max osd." << max
<< " with " << max_base_pg
;
3659 ss
<< " -> " << max_new_pg
;
3660 ss
<< " pgs (" << (float)max_base_pg
/ avg_pg
;
3662 ss
<< " -> " << (float)max_new_pg
/ avg_pg
;
3674 int OSDMap::clean_pg_upmaps(
3676 Incremental
*pending_inc
)
3678 ldout(cct
, 10) << __func__
<< dendl
;
3680 for (auto& p
: pg_upmap
) {
3683 pg_to_raw_osds(p
.first
, &raw
, &primary
);
3684 if (vectors_equal(raw
, p
.second
)) {
3685 ldout(cct
, 10) << " removing redundant pg_upmap " << p
.first
<< " "
3686 << p
.second
<< dendl
;
3687 pending_inc
->old_pg_upmap
.insert(p
.first
);
3691 for (auto& p
: pg_upmap_items
) {
3694 pg_to_raw_osds(p
.first
, &raw
, &primary
);
3695 mempool::osdmap::vector
<pair
<int,int>> newmap
;
3696 for (auto& q
: p
.second
) {
3697 if (std::find(raw
.begin(), raw
.end(), q
.first
) != raw
.end()) {
3698 newmap
.push_back(q
);
3701 if (newmap
.empty()) {
3702 ldout(cct
, 10) << " removing no-op pg_upmap_items " << p
.first
<< " "
3703 << p
.second
<< dendl
;
3704 pending_inc
->old_pg_upmap_items
.insert(p
.first
);
3706 } else if (newmap
!= p
.second
) {
3707 ldout(cct
, 10) << " simplifying partially no-op pg_upmap_items "
3708 << p
.first
<< " " << p
.second
<< " -> " << newmap
<< dendl
;
3709 pending_inc
->new_pg_upmap_items
[p
.first
] = newmap
;
3716 bool OSDMap::try_pg_upmap(
3718 pg_t pg
, ///< pg to potentially remap
3719 const set
<int>& overfull
, ///< osds we'd want to evacuate
3720 const vector
<int>& underfull
, ///< osds to move to, in order of preference
3722 vector
<int> *out
) ///< resulting alternative mapping
3724 const pg_pool_t
*pool
= get_pg_pool(pg
.pool());
3727 int rule
= crush
->find_rule(pool
->get_crush_rule(), pool
->get_type(),
3732 // get original mapping
3733 _pg_to_raw_osds(*pool
, pg
, orig
, NULL
);
3735 // make sure there is something there to remap
3737 for (auto osd
: *orig
) {
3738 if (overfull
.count(osd
)) {
3747 int r
= crush
->try_remap_rule(
3751 overfull
, underfull
,
3761 int OSDMap::calc_pg_upmaps(
3763 float max_deviation_ratio
,
3765 const set
<int64_t>& only_pools_orig
,
3766 OSDMap::Incremental
*pending_inc
)
3768 set
<int64_t> only_pools
;
3769 if (only_pools_orig
.empty()) {
3770 for (auto& i
: pools
) {
3771 only_pools
.insert(i
.first
);
3774 only_pools
= only_pools_orig
;
3777 tmp
.deepish_copy_from(*this);
3778 float start_deviation
= 0;
3779 float end_deviation
= 0;
3780 int num_changed
= 0;
3782 map
<int,set
<pg_t
>> pgs_by_osd
;
3784 float osd_weight_total
= 0;
3785 map
<int,float> osd_weight
;
3786 for (auto& i
: pools
) {
3787 if (!only_pools
.empty() && !only_pools
.count(i
.first
))
3789 for (unsigned ps
= 0; ps
< i
.second
.get_pg_num(); ++ps
) {
3790 pg_t
pg(ps
, i
.first
);
3792 tmp
.pg_to_up_acting_osds(pg
, &up
, nullptr, nullptr, nullptr);
3793 for (auto osd
: up
) {
3794 if (osd
!= CRUSH_ITEM_NONE
)
3795 pgs_by_osd
[osd
].insert(pg
);
3798 total_pgs
+= i
.second
.get_size() * i
.second
.get_pg_num();
3800 map
<int,float> pmap
;
3801 int ruleno
= tmp
.crush
->find_rule(i
.second
.get_crush_rule(),
3802 i
.second
.get_type(),
3803 i
.second
.get_size());
3804 tmp
.crush
->get_rule_weight_osd_map(ruleno
, &pmap
);
3805 ldout(cct
,30) << __func__
<< " pool " << i
.first
<< " ruleno " << ruleno
<< dendl
;
3806 for (auto p
: pmap
) {
3807 osd_weight
[p
.first
] += p
.second
;
3808 osd_weight_total
+= p
.second
;
3811 for (auto& i
: osd_weight
) {
3813 auto p
= pgs_by_osd
.find(i
.first
);
3814 if (p
!= pgs_by_osd
.end())
3815 pgs
= p
->second
.size();
3817 pgs_by_osd
.emplace(i
.first
, set
<pg_t
>());
3818 ldout(cct
, 20) << " osd." << i
.first
<< " weight " << i
.second
3819 << " pgs " << pgs
<< dendl
;
3822 if (osd_weight_total
== 0) {
3823 lderr(cct
) << __func__
<< " abort due to osd_weight_total == 0" << dendl
;
3826 float pgs_per_weight
= total_pgs
/ osd_weight_total
;
3827 ldout(cct
, 10) << " osd_weight_total " << osd_weight_total
<< dendl
;
3828 ldout(cct
, 10) << " pgs_per_weight " << pgs_per_weight
<< dendl
;
3831 float total_deviation
= 0;
3832 map
<int,float> osd_deviation
; // osd, deviation(pgs)
3833 multimap
<float,int> deviation_osd
; // deviation(pgs), osd
3835 for (auto& i
: pgs_by_osd
) {
3836 float target
= osd_weight
[i
.first
] * pgs_per_weight
;
3837 float deviation
= (float)i
.second
.size() - target
;
3838 ldout(cct
, 20) << " osd." << i
.first
3839 << "\tpgs " << i
.second
.size()
3840 << "\ttarget " << target
3841 << "\tdeviation " << deviation
3843 osd_deviation
[i
.first
] = deviation
;
3844 deviation_osd
.insert(make_pair(deviation
, i
.first
));
3845 if (deviation
>= 1.0)
3846 overfull
.insert(i
.first
);
3847 total_deviation
+= abs(deviation
);
3849 if (num_changed
== 0) {
3850 start_deviation
= total_deviation
;
3852 end_deviation
= total_deviation
;
3854 // build underfull, sorted from least-full to most-average
3855 vector
<int> underfull
;
3856 for (auto i
= deviation_osd
.begin();
3857 i
!= deviation_osd
.end();
3859 if (i
->first
>= -.999)
3861 underfull
.push_back(i
->second
);
3863 ldout(cct
, 10) << " total_deviation " << total_deviation
3864 << " overfull " << overfull
3865 << " underfull " << underfull
<< dendl
;
3866 if (overfull
.empty() || underfull
.empty())
3870 bool restart
= false;
3871 for (auto p
= deviation_osd
.rbegin(); p
!= deviation_osd
.rend(); ++p
) {
3872 int osd
= p
->second
;
3873 float deviation
= p
->first
;
3874 float target
= osd_weight
[osd
] * pgs_per_weight
;
3876 if (deviation
/target
< max_deviation_ratio
) {
3877 ldout(cct
, 10) << " osd." << osd
3878 << " target " << target
3879 << " deviation " << deviation
3880 << " -> ratio " << deviation
/target
3881 << " < max ratio " << max_deviation_ratio
<< dendl
;
3884 int num_to_move
= deviation
;
3885 ldout(cct
, 10) << " osd." << osd
<< " move " << num_to_move
<< dendl
;
3886 if (num_to_move
< 1)
3889 set
<pg_t
>& pgs
= pgs_by_osd
[osd
];
3891 // look for remaps we can un-remap
3892 for (auto pg
: pgs
) {
3893 auto p
= tmp
.pg_upmap_items
.find(pg
);
3894 if (p
!= tmp
.pg_upmap_items
.end()) {
3895 for (auto q
: p
->second
) {
3896 if (q
.second
== osd
) {
3897 ldout(cct
, 10) << " dropping pg_upmap_items " << pg
3898 << " " << p
->second
<< dendl
;
3899 tmp
.pg_upmap_items
.erase(p
);
3900 pending_inc
->old_pg_upmap_items
.insert(pg
);
3912 for (auto pg
: pgs
) {
3913 if (tmp
.pg_upmap
.count(pg
) ||
3914 tmp
.pg_upmap_items
.count(pg
)) {
3915 ldout(cct
, 20) << " already remapped " << pg
<< dendl
;
3918 ldout(cct
, 10) << " trying " << pg
<< dendl
;
3919 vector
<int> orig
, out
;
3920 if (!try_pg_upmap(cct
, pg
, overfull
, underfull
, &orig
, &out
)) {
3923 ldout(cct
, 10) << " " << pg
<< " " << orig
<< " -> " << out
<< dendl
;
3924 if (orig
.size() != out
.size()) {
3927 assert(orig
!= out
);
3928 auto& rmi
= tmp
.pg_upmap_items
[pg
];
3929 for (unsigned i
= 0; i
< out
.size(); ++i
) {
3930 if (orig
[i
] != out
[i
]) {
3931 rmi
.push_back(make_pair(orig
[i
], out
[i
]));
3934 pending_inc
->new_pg_upmap_items
[pg
] = rmi
;
3935 ldout(cct
, 10) << " " << pg
<< " pg_upmap_items " << rmi
<< dendl
;
3945 ldout(cct
, 10) << " failed to find any changes to make" << dendl
;
3949 ldout(cct
, 10) << " hit max iterations, stopping" << dendl
;
3953 ldout(cct
, 10) << " start deviation " << start_deviation
<< dendl
;
3954 ldout(cct
, 10) << " end deviation " << end_deviation
<< dendl
;
3958 int OSDMap::get_osds_by_bucket_name(const string
&name
, set
<int> *osds
) const
3960 return crush
->get_leaves(name
, osds
);
3963 template <typename F
>
3964 class OSDUtilizationDumper
: public CrushTreeDumper::Dumper
<F
> {
3966 typedef CrushTreeDumper::Dumper
<F
> Parent
;
3968 OSDUtilizationDumper(const CrushWrapper
*crush
, const OSDMap
*osdmap_
,
3969 const PGStatService
*pgs_
, bool tree_
) :
3970 Parent(crush
, osdmap_
->get_pool_names()),
3974 average_util(average_utilization()),
3982 void dump_stray(F
*f
) {
3983 for (int i
= 0; i
< osdmap
->get_max_osd(); i
++) {
3984 if (osdmap
->exists(i
) && !this->is_touched(i
))
3985 dump_item(CrushTreeDumper::Item(i
, 0, 0, 0), f
);
3989 void dump_item(const CrushTreeDumper::Item
&qi
, F
*f
) override
{
3990 if (!tree
&& qi
.is_bucket())
3993 float reweight
= qi
.is_bucket() ? -1 : osdmap
->get_weightf(qi
.id
);
3994 int64_t kb
= 0, kb_used
= 0, kb_avail
= 0;
3996 if (get_bucket_utilization(qi
.id
, &kb
, &kb_used
, &kb_avail
))
3998 util
= 100.0 * (double)kb_used
/ (double)kb
;
4002 var
= util
/ average_util
;
4004 size_t num_pgs
= qi
.is_bucket() ? 0 : pgs
->get_num_pg_by_osd(qi
.id
);
4006 dump_item(qi
, reweight
, kb
, kb_used
, kb_avail
, util
, var
, num_pgs
, f
);
4008 if (!qi
.is_bucket() && reweight
> 0) {
4009 if (min_var
< 0 || var
< min_var
)
4011 if (max_var
< 0 || var
> max_var
)
4014 double dev
= util
- average_util
;
4016 stddev
+= reweight
* dev
;
4021 virtual void dump_item(const CrushTreeDumper::Item
&qi
,
4028 const size_t num_pgs
,
4032 return sum
> 0 ? sqrt(stddev
/ sum
) : 0;
4035 double average_utilization() {
4036 int64_t kb
= 0, kb_used
= 0;
4037 for (int i
= 0; i
< osdmap
->get_max_osd(); i
++) {
4038 if (!osdmap
->exists(i
) || osdmap
->get_weight(i
) == 0)
4040 int64_t kb_i
, kb_used_i
, kb_avail_i
;
4041 if (get_osd_utilization(i
, &kb_i
, &kb_used_i
, &kb_avail_i
)) {
4043 kb_used
+= kb_used_i
;
4046 return kb
> 0 ? 100.0 * (double)kb_used
/ (double)kb
: 0;
4049 bool get_osd_utilization(int id
, int64_t* kb
, int64_t* kb_used
,
4050 int64_t* kb_avail
) const {
4051 const osd_stat_t
*p
= pgs
->get_osd_stat(id
);
4052 if (!p
) return false;
4054 *kb_used
= p
->kb_used
;
4055 *kb_avail
= p
->kb_avail
;
4059 bool get_bucket_utilization(int id
, int64_t* kb
, int64_t* kb_used
,
4060 int64_t* kb_avail
) const {
4062 if (osdmap
->is_out(id
)) {
4068 return get_osd_utilization(id
, kb
, kb_used
, kb_avail
);
4075 for (int k
= osdmap
->crush
->get_bucket_size(id
) - 1; k
>= 0; k
--) {
4076 int item
= osdmap
->crush
->get_bucket_item(id
, k
);
4077 int64_t kb_i
= 0, kb_used_i
= 0, kb_avail_i
= 0;
4078 if (!get_bucket_utilization(item
, &kb_i
, &kb_used_i
, &kb_avail_i
))
4081 *kb_used
+= kb_used_i
;
4082 *kb_avail
+= kb_avail_i
;
4088 const OSDMap
*osdmap
;
4089 const PGStatService
*pgs
;
4091 double average_util
;
4099 class OSDUtilizationPlainDumper
: public OSDUtilizationDumper
<TextTable
> {
4101 typedef OSDUtilizationDumper
<TextTable
> Parent
;
4103 OSDUtilizationPlainDumper(const CrushWrapper
*crush
, const OSDMap
*osdmap
,
4104 const PGStatService
*pgs
, bool tree
) :
4105 Parent(crush
, osdmap
, pgs
, tree
) {}
4107 void dump(TextTable
*tbl
) {
4108 tbl
->define_column("ID", TextTable::LEFT
, TextTable::RIGHT
);
4109 tbl
->define_column("CLASS", TextTable::LEFT
, TextTable::RIGHT
);
4110 tbl
->define_column("WEIGHT", TextTable::LEFT
, TextTable::RIGHT
);
4111 tbl
->define_column("REWEIGHT", TextTable::LEFT
, TextTable::RIGHT
);
4112 tbl
->define_column("SIZE", TextTable::LEFT
, TextTable::RIGHT
);
4113 tbl
->define_column("USE", TextTable::LEFT
, TextTable::RIGHT
);
4114 tbl
->define_column("AVAIL", TextTable::LEFT
, TextTable::RIGHT
);
4115 tbl
->define_column("%USE", TextTable::LEFT
, TextTable::RIGHT
);
4116 tbl
->define_column("VAR", TextTable::LEFT
, TextTable::RIGHT
);
4117 tbl
->define_column("PGS", TextTable::LEFT
, TextTable::RIGHT
);
4119 tbl
->define_column("TYPE NAME", TextTable::LEFT
, TextTable::LEFT
);
4128 << si_t(pgs
->get_osd_sum().kb
<< 10)
4129 << si_t(pgs
->get_osd_sum().kb_used
<< 10)
4130 << si_t(pgs
->get_osd_sum().kb_avail
<< 10)
4131 << lowprecision_t(average_util
)
4133 << TextTable::endrow
;
4137 struct lowprecision_t
{
4139 explicit lowprecision_t(float _v
) : v(_v
) {}
4141 friend std::ostream
&operator<<(ostream
& out
, const lowprecision_t
& v
);
4143 using OSDUtilizationDumper
<TextTable
>::dump_item
;
4144 void dump_item(const CrushTreeDumper::Item
&qi
,
4151 const size_t num_pgs
,
4152 TextTable
*tbl
) override
{
4153 const char *c
= crush
->get_item_class(qi
.id
);
4158 << weightf_t(qi
.weight
)
4159 << weightf_t(reweight
)
4161 << si_t(kb_used
<< 10)
4162 << si_t(kb_avail
<< 10)
4163 << lowprecision_t(util
)
4164 << lowprecision_t(var
);
4166 if (qi
.is_bucket()) {
4174 for (int k
= 0; k
< qi
.depth
; k
++)
4176 if (qi
.is_bucket()) {
4177 int type
= crush
->get_bucket_type(qi
.id
);
4178 name
<< crush
->get_type_name(type
) << " "
4179 << crush
->get_item_name(qi
.id
);
4181 name
<< "osd." << qi
.id
;
4186 *tbl
<< TextTable::endrow
;
4192 out
<< "MIN/MAX VAR: " << lowprecision_t(min_var
)
4193 << "/" << lowprecision_t(max_var
) << " "
4194 << "STDDEV: " << lowprecision_t(dev());
4199 ostream
& operator<<(ostream
& out
,
4200 const OSDUtilizationPlainDumper::lowprecision_t
& v
)
4204 } else if (v
.v
< 0.001) {
4207 std::streamsize p
= out
.precision();
4208 return out
<< std::fixed
<< std::setprecision(2) << v
.v
<< std::setprecision(p
);
4212 class OSDUtilizationFormatDumper
: public OSDUtilizationDumper
<Formatter
> {
4214 typedef OSDUtilizationDumper
<Formatter
> Parent
;
4216 OSDUtilizationFormatDumper(const CrushWrapper
*crush
, const OSDMap
*osdmap
,
4217 const PGStatService
*pgs
, bool tree
) :
4218 Parent(crush
, osdmap
, pgs
, tree
) {}
4220 void dump(Formatter
*f
) {
4221 f
->open_array_section("nodes");
4225 f
->open_array_section("stray");
4231 using OSDUtilizationDumper
<Formatter
>::dump_item
;
4232 void dump_item(const CrushTreeDumper::Item
&qi
,
4239 const size_t num_pgs
,
4240 Formatter
*f
) override
{
4241 f
->open_object_section("item");
4242 CrushTreeDumper::dump_item_fields(crush
, weight_set_names
, qi
, f
);
4243 f
->dump_float("reweight", reweight
);
4244 f
->dump_int("kb", kb
);
4245 f
->dump_int("kb_used", kb_used
);
4246 f
->dump_int("kb_avail", kb_avail
);
4247 f
->dump_float("utilization", util
);
4248 f
->dump_float("var", var
);
4249 f
->dump_unsigned("pgs", num_pgs
);
4250 CrushTreeDumper::dump_bucket_children(crush
, qi
, f
);
4255 void summary(Formatter
*f
) {
4256 f
->open_object_section("summary");
4257 f
->dump_int("total_kb", pgs
->get_osd_sum().kb
);
4258 f
->dump_int("total_kb_used", pgs
->get_osd_sum().kb_used
);
4259 f
->dump_int("total_kb_avail", pgs
->get_osd_sum().kb_avail
);
4260 f
->dump_float("average_utilization", average_util
);
4261 f
->dump_float("min_var", min_var
);
4262 f
->dump_float("max_var", max_var
);
4263 f
->dump_float("dev", dev());
4268 void print_osd_utilization(const OSDMap
& osdmap
,
4269 const PGStatService
*pgstat
,
4274 const CrushWrapper
*crush
= osdmap
.crush
.get();
4276 f
->open_object_section("df");
4277 OSDUtilizationFormatDumper
d(crush
, &osdmap
, pgstat
, tree
);
4283 OSDUtilizationPlainDumper
d(crush
, &osdmap
, pgstat
, tree
);
4286 out
<< tbl
<< d
.summary() << "\n";
4290 void OSDMap::check_health(health_check_map_t
*checks
) const
4292 int num_osds
= get_num_osds();
4295 // OSD_$subtree_DOWN
4297 if (num_osds
>= 0) {
4298 int num_in_osds
= 0;
4299 int num_down_in_osds
= 0;
4301 set
<int> down_in_osds
;
4302 set
<int> up_in_osds
;
4303 set
<int> subtree_up
;
4304 unordered_map
<int, set
<int> > subtree_type_down
;
4305 unordered_map
<int, int> num_osds_subtree
;
4306 int max_type
= crush
->get_max_type_id();
4308 for (int i
= 0; i
< get_max_osd(); i
++) {
4310 if (crush
->item_exists(i
)) {
4318 if (down_in_osds
.count(i
) || up_in_osds
.count(i
))
4321 down_in_osds
.insert(i
);
4324 for (int type
= 0; type
<= max_type
; type
++) {
4325 if (!crush
->get_type_name(type
))
4327 int r
= crush
->get_immediate_parent_id(current
, &parent_id
);
4330 // break early if this parent is already marked as up
4331 if (subtree_up
.count(parent_id
))
4333 type
= crush
->get_bucket_type(parent_id
);
4334 if (!subtree_type_is_down(
4335 g_ceph_context
, parent_id
, type
,
4336 &down_in_osds
, &up_in_osds
, &subtree_up
, &subtree_type_down
))
4338 current
= parent_id
;
4343 // calculate the number of down osds in each down subtree and
4344 // store it in num_osds_subtree
4345 for (int type
= 1; type
<= max_type
; type
++) {
4346 if (!crush
->get_type_name(type
))
4348 for (auto j
= subtree_type_down
[type
].begin();
4349 j
!= subtree_type_down
[type
].end();
4353 int num_children
= crush
->get_children(*j
, &children
);
4354 if (num_children
== 0)
4356 for (auto l
= children
.begin(); l
!= children
.end(); ++l
) {
4359 } else if (num_osds_subtree
[*l
] > 0) {
4360 num
= num
+ num_osds_subtree
[*l
];
4363 num_osds_subtree
[*j
] = num
;
4366 num_down_in_osds
= down_in_osds
.size();
4367 assert(num_down_in_osds
<= num_in_osds
);
4368 if (num_down_in_osds
> 0) {
4369 // summary of down subtree types and osds
4370 for (int type
= max_type
; type
> 0; type
--) {
4371 if (!crush
->get_type_name(type
))
4373 if (subtree_type_down
[type
].size() > 0) {
4375 ss
<< subtree_type_down
[type
].size() << " "
4376 << crush
->get_type_name(type
);
4377 if (subtree_type_down
[type
].size() > 1) {
4380 int sum_down_osds
= 0;
4381 for (auto j
= subtree_type_down
[type
].begin();
4382 j
!= subtree_type_down
[type
].end();
4384 sum_down_osds
= sum_down_osds
+ num_osds_subtree
[*j
];
4386 ss
<< " (" << sum_down_osds
<< " osds) down";
4387 string err
= string("OSD_") +
4388 string(crush
->get_type_name(type
)) + "_DOWN";
4389 boost::to_upper(err
);
4390 auto& d
= checks
->add(err
, HEALTH_WARN
, ss
.str());
4391 for (auto j
= subtree_type_down
[type
].rbegin();
4392 j
!= subtree_type_down
[type
].rend();
4395 ss
<< crush
->get_type_name(type
);
4397 ss
<< crush
->get_item_name(*j
);
4398 // at the top level, do not print location
4399 if (type
!= max_type
) {
4401 ss
<< crush
->get_full_location_ordered_string(*j
);
4404 int num
= num_osds_subtree
[*j
];
4405 ss
<< " (" << num
<< " osds)";
4407 d
.detail
.push_back(ss
.str());
4412 ss
<< down_in_osds
.size() << " osds down";
4413 auto& d
= checks
->add("OSD_DOWN", HEALTH_WARN
, ss
.str());
4414 for (auto it
= down_in_osds
.begin(); it
!= down_in_osds
.end(); ++it
) {
4416 ss
<< "osd." << *it
<< " (";
4417 ss
<< crush
->get_full_location_ordered_string(*it
);
4419 d
.detail
.push_back(ss
.str());
4423 if (!osds
.empty()) {
4425 ss
<< osds
.size() << " osds exist in the crush map but not in the osdmap";
4426 auto& d
= checks
->add("OSD_ORPHAN", HEALTH_WARN
, ss
.str());
4427 for (auto osd
: osds
) {
4429 ss
<< "osd." << osd
<< " exists in crush map but not in osdmap";
4430 d
.detail
.push_back(ss
.str());
4435 // OSD_OUT_OF_ORDER_FULL
4437 // An osd could configure failsafe ratio, to something different
4438 // but for now assume it is the same here.
4439 float fsr
= g_conf
->osd_failsafe_full_ratio
;
4440 if (fsr
> 1.0) fsr
/= 100;
4441 float fr
= get_full_ratio();
4442 float br
= get_backfillfull_ratio();
4443 float nr
= get_nearfull_ratio();
4445 list
<string
> detail
;
4446 // These checks correspond to how OSDService::check_full_status() in an OSD
4447 // handles the improper setting of these values.
4450 ss
<< "backfillfull_ratio (" << br
4451 << ") < nearfull_ratio (" << nr
<< "), increased";
4452 detail
.push_back(ss
.str());
4457 ss
<< "full_ratio (" << fr
<< ") < backfillfull_ratio (" << br
4459 detail
.push_back(ss
.str());
4464 ss
<< "osd_failsafe_full_ratio (" << fsr
<< ") < full_ratio (" << fr
4466 detail
.push_back(ss
.str());
4468 if (!detail
.empty()) {
4469 auto& d
= checks
->add("OSD_OUT_OF_ORDER_FULL", HEALTH_ERR
,
4470 "full ratio(s) out of order");
4471 d
.detail
.swap(detail
);
4478 // OSD_FAILSAFE_FULL
4480 set
<int> full
, backfillfull
, nearfull
;
4481 get_full_osd_counts(&full
, &backfillfull
, &nearfull
);
4484 ss
<< full
.size() << " full osd(s)";
4485 auto& d
= checks
->add("OSD_FULL", HEALTH_ERR
, ss
.str());
4486 for (auto& i
: full
) {
4488 ss
<< "osd." << i
<< " is full";
4489 d
.detail
.push_back(ss
.str());
4492 if (backfillfull
.size()) {
4494 ss
<< backfillfull
.size() << " backfillfull osd(s)";
4495 auto& d
= checks
->add("OSD_BACKFILLFULL", HEALTH_WARN
, ss
.str());
4496 for (auto& i
: backfillfull
) {
4498 ss
<< "osd." << i
<< " is backfill full";
4499 d
.detail
.push_back(ss
.str());
4502 if (nearfull
.size()) {
4504 ss
<< nearfull
.size() << " nearfull osd(s)";
4505 auto& d
= checks
->add("OSD_NEARFULL", HEALTH_WARN
, ss
.str());
4506 for (auto& i
: nearfull
) {
4508 ss
<< "osd." << i
<< " is near full";
4509 d
.detail
.push_back(ss
.str());
4517 uint64_t warn_flags
=
4519 CEPH_OSDMAP_PAUSERD
|
4520 CEPH_OSDMAP_PAUSEWR
|
4521 CEPH_OSDMAP_PAUSEREC
|
4523 CEPH_OSDMAP_NODOWN
|
4526 CEPH_OSDMAP_NOBACKFILL
|
4527 CEPH_OSDMAP_NORECOVER
|
4528 CEPH_OSDMAP_NOSCRUB
|
4529 CEPH_OSDMAP_NODEEP_SCRUB
|
4530 CEPH_OSDMAP_NOTIERAGENT
|
4531 CEPH_OSDMAP_NOREBALANCE
;
4532 if (test_flag(warn_flags
)) {
4534 ss
<< get_flag_string(get_flags() & warn_flags
)
4536 checks
->add("OSDMAP_FLAGS", HEALTH_WARN
, ss
.str());
4542 list
<string
> detail
;
4543 const unsigned flags
=
4548 for (int i
= 0; i
< max_osd
; ++i
) {
4549 if (osd_state
[i
] & flags
) {
4552 OSDMap::calc_state_set(osd_state
[i
] & flags
, states
);
4553 ss
<< "osd." << i
<< " has flags " << states
;
4554 detail
.push_back(ss
.str());
4557 if (!detail
.empty()) {
4559 ss
<< detail
.size() << " osd(s) have {NOUP,NODOWN,NOIN,NOOUT} flags set";
4560 auto& d
= checks
->add("OSD_FLAGS", HEALTH_WARN
, ss
.str());
4561 d
.detail
.swap(detail
);
4565 // OLD_CRUSH_TUNABLES
4566 if (g_conf
->mon_warn_on_legacy_crush_tunables
) {
4567 string min
= crush
->get_min_required_version();
4568 if (min
< g_conf
->mon_crush_min_required_version
) {
4570 ss
<< "crush map has legacy tunables (require " << min
4571 << ", min is " << g_conf
->mon_crush_min_required_version
<< ")";
4572 auto& d
= checks
->add("OLD_CRUSH_TUNABLES", HEALTH_WARN
, ss
.str());
4573 d
.detail
.push_back("see http://docs.ceph.com/docs/master/rados/operations/crush-map/#tunables");
4577 // OLD_CRUSH_STRAW_CALC_VERSION
4578 if (g_conf
->mon_warn_on_crush_straw_calc_version_zero
) {
4579 if (crush
->get_straw_calc_version() == 0) {
4581 ss
<< "crush map has straw_calc_version=0";
4582 auto& d
= checks
->add("OLD_CRUSH_STRAW_CALC_VERSION", HEALTH_WARN
, ss
.str());
4584 "see http://docs.ceph.com/docs/master/rados/operations/crush-map/#tunables");
4588 // CACHE_POOL_NO_HIT_SET
4589 if (g_conf
->mon_warn_on_cache_pools_without_hit_sets
) {
4590 list
<string
> detail
;
4591 for (map
<int64_t, pg_pool_t
>::const_iterator p
= pools
.begin();
4594 const pg_pool_t
& info
= p
->second
;
4595 if (info
.cache_mode_requires_hit_set() &&
4596 info
.hit_set_params
.get_type() == HitSet::TYPE_NONE
) {
4598 ss
<< "pool '" << get_pool_name(p
->first
)
4599 << "' with cache_mode " << info
.get_cache_mode_name()
4600 << " needs hit_set_type to be set but it is not";
4601 detail
.push_back(ss
.str());
4604 if (!detail
.empty()) {
4606 ss
<< detail
.size() << " cache pools are missing hit_sets";
4607 auto& d
= checks
->add("CACHE_POOL_NO_HIT_SET", HEALTH_WARN
, ss
.str());
4608 d
.detail
.swap(detail
);
4612 // OSD_NO_SORTBITWISE
4613 if (!test_flag(CEPH_OSDMAP_SORTBITWISE
) &&
4614 (get_up_osd_features() &
4615 CEPH_FEATURE_OSD_BITWISE_HOBJ_SORT
)) {
4617 ss
<< "no legacy OSD present but 'sortbitwise' flag is not set";
4618 checks
->add("OSD_NO_SORTBITWISE", HEALTH_WARN
, ss
.str());
4621 // OSD_UPGRADE_FINISHED
4622 // none of these (yet) since we don't run until luminous upgrade is done.
4626 list
<string
> detail
;
4627 for (auto it
: get_pools()) {
4628 const pg_pool_t
&pool
= it
.second
;
4629 if (pool
.has_flag(pg_pool_t::FLAG_FULL
)) {
4630 const string
& pool_name
= get_pool_name(it
.first
);
4632 ss
<< "pool '" << pool_name
<< "' is full";
4633 detail
.push_back(ss
.str());
4636 if (!detail
.empty()) {
4638 ss
<< detail
.size() << " pool(s) full";
4639 auto& d
= checks
->add("POOL_FULL", HEALTH_WARN
, ss
.str());
4640 d
.detail
.swap(detail
);