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
);
923 f
->open_array_section("new_pg_temp");
925 for (const auto &pg_temp
: new_pg_temp
) {
926 f
->open_object_section("pg");
927 f
->dump_stream("pgid") << pg_temp
.first
;
928 f
->open_array_section("osds");
930 for (const auto &osd
: pg_temp
.second
)
931 f
->dump_int("osd", osd
);
937 f
->open_array_section("primary_temp");
939 for (const auto &primary_temp
: new_primary_temp
) {
940 f
->dump_stream("pgid") << primary_temp
.first
;
941 f
->dump_int("osd", primary_temp
.second
);
943 f
->close_section(); // primary_temp
945 f
->open_array_section("new_pg_upmap");
946 for (auto& i
: new_pg_upmap
) {
947 f
->open_object_section("mapping");
948 f
->dump_stream("pgid") << i
.first
;
949 f
->open_array_section("osds");
950 for (auto osd
: i
.second
) {
951 f
->dump_int("osd", osd
);
957 f
->open_array_section("old_pg_upmap");
958 for (auto& i
: old_pg_upmap
) {
959 f
->dump_stream("pgid") << i
;
963 f
->open_array_section("new_pg_upmap_items");
964 for (auto& i
: new_pg_upmap_items
) {
965 f
->open_object_section("mapping");
966 f
->dump_stream("pgid") << i
.first
;
967 f
->open_array_section("mappings");
968 for (auto& p
: i
.second
) {
969 f
->open_object_section("mapping");
970 f
->dump_int("from", p
.first
);
971 f
->dump_int("to", p
.second
);
978 f
->open_array_section("old_pg_upmap_items");
979 for (auto& i
: old_pg_upmap_items
) {
980 f
->dump_stream("pgid") << i
;
984 f
->open_array_section("new_up_thru");
986 for (const auto &up_thru
: new_up_thru
) {
987 f
->open_object_section("osd");
988 f
->dump_int("osd", up_thru
.first
);
989 f
->dump_int("up_thru", up_thru
.second
);
994 f
->open_array_section("new_lost");
996 for (const auto &lost
: new_lost
) {
997 f
->open_object_section("osd");
998 f
->dump_int("osd", lost
.first
);
999 f
->dump_int("epoch_lost", lost
.second
);
1004 f
->open_array_section("new_last_clean_interval");
1006 for (const auto &last_clean_interval
: new_last_clean_interval
) {
1007 f
->open_object_section("osd");
1008 f
->dump_int("osd", last_clean_interval
.first
);
1009 f
->dump_int("first", last_clean_interval
.second
.first
);
1010 f
->dump_int("last", last_clean_interval
.second
.second
);
1015 f
->open_array_section("new_blacklist");
1016 for (const auto &blist
: new_blacklist
) {
1019 f
->dump_stream(ss
.str().c_str()) << blist
.second
;
1022 f
->open_array_section("old_blacklist");
1023 for (const auto &blist
: old_blacklist
)
1024 f
->dump_stream("addr") << blist
;
1027 f
->open_array_section("new_xinfo");
1028 for (const auto &xinfo
: new_xinfo
) {
1029 f
->open_object_section("xinfo");
1030 f
->dump_int("osd", xinfo
.first
);
1031 xinfo
.second
.dump(f
);
1036 if (cluster_snapshot
.size())
1037 f
->dump_string("cluster_snapshot", cluster_snapshot
);
1039 f
->open_array_section("new_uuid");
1040 for (const auto &uuid
: new_uuid
) {
1041 f
->open_object_section("osd");
1042 f
->dump_int("osd", uuid
.first
);
1043 f
->dump_stream("uuid") << uuid
.second
;
1048 OSDMap::dump_erasure_code_profiles(new_erasure_code_profiles
, f
);
1049 f
->open_array_section("old_erasure_code_profiles");
1050 for (const auto &erasure_code_profile
: old_erasure_code_profiles
) {
1051 f
->dump_string("old", erasure_code_profile
.c_str());
1056 void OSDMap::Incremental::generate_test_instances(list
<Incremental
*>& o
)
1058 o
.push_back(new Incremental
);
1061 // ----------------------------------
1064 void OSDMap::set_epoch(epoch_t e
)
1067 for (auto &pool
: pools
)
1068 pool
.second
.last_change
= e
;
1071 bool OSDMap::is_blacklisted(const entity_addr_t
& a
) const
1073 if (blacklist
.empty())
1076 // this specific instance?
1077 if (blacklist
.count(a
))
1080 // is entire ip blacklisted?
1082 entity_addr_t b
= a
;
1085 if (blacklist
.count(b
)) {
1093 void OSDMap::get_blacklist(list
<pair
<entity_addr_t
,utime_t
> > *bl
) const
1095 std::copy(blacklist
.begin(), blacklist
.end(), std::back_inserter(*bl
));
1098 void OSDMap::get_blacklist(std::set
<entity_addr_t
> *bl
) const
1100 for (const auto &i
: blacklist
) {
1101 bl
->insert(i
.first
);
1105 void OSDMap::set_max_osd(int m
)
1109 osd_state
.resize(m
);
1110 osd_weight
.resize(m
);
1111 for (; o
<max_osd
; o
++) {
1113 osd_weight
[o
] = CEPH_OSD_OUT
;
1116 osd_xinfo
.resize(m
);
1117 osd_addrs
->client_addr
.resize(m
);
1118 osd_addrs
->cluster_addr
.resize(m
);
1119 osd_addrs
->hb_back_addr
.resize(m
);
1120 osd_addrs
->hb_front_addr
.resize(m
);
1121 osd_uuid
->resize(m
);
1122 if (osd_primary_affinity
)
1123 osd_primary_affinity
->resize(m
, CEPH_OSD_DEFAULT_PRIMARY_AFFINITY
);
1128 int OSDMap::calc_num_osds()
1133 for (int i
=0; i
<max_osd
; i
++) {
1134 if (osd_state
[i
] & CEPH_OSD_EXISTS
) {
1136 if (osd_state
[i
] & CEPH_OSD_UP
) {
1139 if (get_weight(i
) != CEPH_OSD_OUT
) {
1147 void OSDMap::count_full_nearfull_osds(int *full
, int *backfill
, int *nearfull
) const
1152 for (int i
= 0; i
< max_osd
; ++i
) {
1153 if (exists(i
) && is_up(i
) && is_in(i
)) {
1154 if (osd_state
[i
] & CEPH_OSD_FULL
)
1156 else if (osd_state
[i
] & CEPH_OSD_BACKFILLFULL
)
1158 else if (osd_state
[i
] & CEPH_OSD_NEARFULL
)
1164 static bool get_osd_utilization(
1165 const mempool::pgmap::unordered_map
<int32_t,osd_stat_t
> &osd_stat
,
1166 int id
, int64_t* kb
, int64_t* kb_used
, int64_t* kb_avail
)
1168 auto p
= osd_stat
.find(id
);
1169 if (p
== osd_stat
.end())
1172 *kb_used
= p
->second
.kb_used
;
1173 *kb_avail
= p
->second
.kb_avail
;
1177 void OSDMap::get_full_osd_util(
1178 const mempool::pgmap::unordered_map
<int32_t,osd_stat_t
> &osd_stat
,
1179 map
<int, float> *full
, map
<int, float> *backfill
, map
<int, float> *nearfull
) const
1184 for (int i
= 0; i
< max_osd
; ++i
) {
1185 if (exists(i
) && is_up(i
) && is_in(i
)) {
1186 int64_t kb
, kb_used
, kb_avail
;
1187 if (osd_state
[i
] & CEPH_OSD_FULL
) {
1188 if (get_osd_utilization(osd_stat
, i
, &kb
, &kb_used
, &kb_avail
))
1189 full
->emplace(i
, (float)kb_used
/ (float)kb
);
1190 } else if (osd_state
[i
] & CEPH_OSD_BACKFILLFULL
) {
1191 if (get_osd_utilization(osd_stat
, i
, &kb
, &kb_used
, &kb_avail
))
1192 backfill
->emplace(i
, (float)kb_used
/ (float)kb
);
1193 } else if (osd_state
[i
] & CEPH_OSD_NEARFULL
) {
1194 if (get_osd_utilization(osd_stat
, i
, &kb
, &kb_used
, &kb_avail
))
1195 nearfull
->emplace(i
, (float)kb_used
/ (float)kb
);
1201 void OSDMap::get_full_osd_counts(set
<int> *full
, set
<int> *backfill
,
1202 set
<int> *nearfull
) const
1207 for (int i
= 0; i
< max_osd
; ++i
) {
1208 if (exists(i
) && is_up(i
) && is_in(i
)) {
1209 if (osd_state
[i
] & CEPH_OSD_FULL
)
1211 else if (osd_state
[i
] & CEPH_OSD_BACKFILLFULL
)
1212 backfill
->emplace(i
);
1213 else if (osd_state
[i
] & CEPH_OSD_NEARFULL
)
1214 nearfull
->emplace(i
);
1219 void OSDMap::get_all_osds(set
<int32_t>& ls
) const
1221 for (int i
=0; i
<max_osd
; i
++)
1226 void OSDMap::get_up_osds(set
<int32_t>& ls
) const
1228 for (int i
= 0; i
< max_osd
; i
++) {
1234 void OSDMap::get_out_osds(set
<int32_t>& ls
) const
1236 for (int i
= 0; i
< max_osd
; i
++) {
1242 void OSDMap::calc_state_set(int state
, set
<string
>& st
)
1245 for (unsigned s
= 1; t
; s
<<= 1) {
1248 st
.insert(ceph_osd_state_name(s
));
1253 void OSDMap::adjust_osd_weights(const map
<int,double>& weights
, Incremental
& inc
) const
1256 for (const auto &weight
: weights
) {
1257 if (weight
.second
> max
)
1258 max
= weight
.second
;
1261 for (const auto &weight
: weights
) {
1262 inc
.new_weight
[weight
.first
] = (unsigned)((weight
.second
/ max
) * CEPH_OSD_IN
);
1266 int OSDMap::identify_osd(const entity_addr_t
& addr
) const
1268 for (int i
=0; i
<max_osd
; i
++)
1269 if (exists(i
) && (get_addr(i
) == addr
|| get_cluster_addr(i
) == addr
))
1274 int OSDMap::identify_osd(const uuid_d
& u
) const
1276 for (int i
=0; i
<max_osd
; i
++)
1277 if (exists(i
) && get_uuid(i
) == u
)
1282 int OSDMap::identify_osd_on_all_channels(const entity_addr_t
& addr
) const
1284 for (int i
=0; i
<max_osd
; i
++)
1285 if (exists(i
) && (get_addr(i
) == addr
|| get_cluster_addr(i
) == addr
||
1286 get_hb_back_addr(i
) == addr
|| get_hb_front_addr(i
) == addr
))
1291 int OSDMap::find_osd_on_ip(const entity_addr_t
& ip
) const
1293 for (int i
=0; i
<max_osd
; i
++)
1294 if (exists(i
) && (get_addr(i
).is_same_host(ip
) || get_cluster_addr(i
).is_same_host(ip
)))
1300 uint64_t OSDMap::get_features(int entity_type
, uint64_t *pmask
) const
1302 uint64_t features
= 0; // things we actually have
1303 uint64_t mask
= 0; // things we could have
1305 if (crush
->has_nondefault_tunables())
1306 features
|= CEPH_FEATURE_CRUSH_TUNABLES
;
1307 if (crush
->has_nondefault_tunables2())
1308 features
|= CEPH_FEATURE_CRUSH_TUNABLES2
;
1309 if (crush
->has_nondefault_tunables3())
1310 features
|= CEPH_FEATURE_CRUSH_TUNABLES3
;
1311 if (crush
->has_v4_buckets())
1312 features
|= CEPH_FEATURE_CRUSH_V4
;
1313 if (crush
->has_nondefault_tunables5())
1314 features
|= CEPH_FEATURE_CRUSH_TUNABLES5
;
1315 if (crush
->has_incompat_choose_args())
1316 features
|= CEPH_FEATURE_CRUSH_CHOOSE_ARGS
;
1317 mask
|= CEPH_FEATURES_CRUSH
;
1319 if (!pg_upmap
.empty() || !pg_upmap_items
.empty())
1320 features
|= CEPH_FEATUREMASK_OSDMAP_PG_UPMAP
;
1321 mask
|= CEPH_FEATUREMASK_OSDMAP_PG_UPMAP
;
1323 for (auto &pool
: pools
) {
1324 if (pool
.second
.has_flag(pg_pool_t::FLAG_HASHPSPOOL
)) {
1325 features
|= CEPH_FEATURE_OSDHASHPSPOOL
;
1327 if (pool
.second
.is_erasure() &&
1328 entity_type
!= CEPH_ENTITY_TYPE_CLIENT
) { // not for clients
1329 features
|= CEPH_FEATURE_OSD_ERASURE_CODES
;
1331 if (!pool
.second
.tiers
.empty() ||
1332 pool
.second
.is_tier()) {
1333 features
|= CEPH_FEATURE_OSD_CACHEPOOL
;
1335 int ruleid
= crush
->find_rule(pool
.second
.get_crush_rule(),
1336 pool
.second
.get_type(),
1337 pool
.second
.get_size());
1339 if (crush
->is_v2_rule(ruleid
))
1340 features
|= CEPH_FEATURE_CRUSH_V2
;
1341 if (crush
->is_v3_rule(ruleid
))
1342 features
|= CEPH_FEATURE_CRUSH_TUNABLES3
;
1343 if (crush
->is_v5_rule(ruleid
))
1344 features
|= CEPH_FEATURE_CRUSH_TUNABLES5
;
1347 if (entity_type
== CEPH_ENTITY_TYPE_OSD
) {
1348 for (auto &erasure_code_profile
: erasure_code_profiles
) {
1349 auto& profile
= erasure_code_profile
.second
;
1350 const auto& plugin
= profile
.find("plugin");
1351 if (plugin
!= profile
.end()) {
1352 if (plugin
->second
== "isa" || plugin
->second
== "lrc")
1353 features
|= CEPH_FEATURE_ERASURE_CODE_PLUGINS_V2
;
1354 if (plugin
->second
== "shec")
1355 features
|= CEPH_FEATURE_ERASURE_CODE_PLUGINS_V3
;
1359 mask
|= CEPH_FEATURE_OSDHASHPSPOOL
| CEPH_FEATURE_OSD_CACHEPOOL
;
1360 if (entity_type
!= CEPH_ENTITY_TYPE_CLIENT
)
1361 mask
|= CEPH_FEATURE_OSD_ERASURE_CODES
;
1363 if (osd_primary_affinity
) {
1364 for (int i
= 0; i
< max_osd
; ++i
) {
1365 if ((*osd_primary_affinity
)[i
] != CEPH_OSD_DEFAULT_PRIMARY_AFFINITY
) {
1366 features
|= CEPH_FEATURE_OSD_PRIMARY_AFFINITY
;
1371 mask
|= CEPH_FEATURE_OSD_PRIMARY_AFFINITY
;
1373 if (entity_type
== CEPH_ENTITY_TYPE_OSD
) {
1374 const uint64_t jewel_features
= CEPH_FEATURE_SERVER_JEWEL
;
1375 if (require_osd_release
>= CEPH_RELEASE_JEWEL
) {
1376 features
|= jewel_features
;
1378 mask
|= jewel_features
;
1380 const uint64_t kraken_features
= CEPH_FEATUREMASK_SERVER_KRAKEN
1381 | CEPH_FEATURE_MSG_ADDR2
;
1382 if (require_osd_release
>= CEPH_RELEASE_KRAKEN
) {
1383 features
|= kraken_features
;
1385 mask
|= kraken_features
;
1393 uint8_t OSDMap::get_min_compat_client() const
1395 uint64_t f
= get_features(CEPH_ENTITY_TYPE_CLIENT
, nullptr);
1397 if (HAVE_FEATURE(f
, OSDMAP_PG_UPMAP
) || // v12.0.0-1733-g27d6f43
1398 HAVE_FEATURE(f
, CRUSH_CHOOSE_ARGS
)) { // v12.0.1-2172-gef1ef28
1399 return CEPH_RELEASE_LUMINOUS
; // v12.2.0
1401 if (HAVE_FEATURE(f
, CRUSH_TUNABLES5
)) { // v10.0.0-612-g043a737
1402 return CEPH_RELEASE_JEWEL
; // v10.2.0
1404 if (HAVE_FEATURE(f
, CRUSH_V4
)) { // v0.91-678-g325fc56
1405 return CEPH_RELEASE_HAMMER
; // v0.94.0
1407 if (HAVE_FEATURE(f
, OSD_PRIMARY_AFFINITY
) || // v0.76-553-gf825624
1408 HAVE_FEATURE(f
, CRUSH_TUNABLES3
) || // v0.76-395-ge20a55d
1409 HAVE_FEATURE(f
, OSD_ERASURE_CODES
) || // v0.73-498-gbfc86a8
1410 HAVE_FEATURE(f
, OSD_CACHEPOOL
)) { // v0.67-401-gb91c1c5
1411 return CEPH_RELEASE_FIREFLY
; // v0.80.0
1413 if (HAVE_FEATURE(f
, CRUSH_TUNABLES2
) || // v0.54-684-g0cc47ff
1414 HAVE_FEATURE(f
, OSDHASHPSPOOL
)) { // v0.57-398-g8cc2b0f
1415 return CEPH_RELEASE_DUMPLING
; // v0.67.0
1417 if (HAVE_FEATURE(f
, CRUSH_TUNABLES
)) { // v0.48argonaut-206-g6f381af
1418 return CEPH_RELEASE_ARGONAUT
; // v0.48argonaut-206-g6f381af
1420 return CEPH_RELEASE_ARGONAUT
; // v0.48argonaut-206-g6f381af
1423 void OSDMap::_calc_up_osd_features()
1426 cached_up_osd_features
= 0;
1427 for (int osd
= 0; osd
< max_osd
; ++osd
) {
1430 const osd_xinfo_t
&xi
= get_xinfo(osd
);
1432 cached_up_osd_features
= xi
.features
;
1435 cached_up_osd_features
&= xi
.features
;
1440 uint64_t OSDMap::get_up_osd_features() const
1442 return cached_up_osd_features
;
1445 void OSDMap::dedup(const OSDMap
*o
, OSDMap
*n
)
1447 if (o
->epoch
== n
->epoch
)
1453 if (o
->max_osd
!= n
->max_osd
)
1455 for (int i
= 0; i
< o
->max_osd
&& i
< n
->max_osd
; i
++) {
1456 if ( n
->osd_addrs
->client_addr
[i
] && o
->osd_addrs
->client_addr
[i
] &&
1457 *n
->osd_addrs
->client_addr
[i
] == *o
->osd_addrs
->client_addr
[i
])
1458 n
->osd_addrs
->client_addr
[i
] = o
->osd_addrs
->client_addr
[i
];
1461 if ( n
->osd_addrs
->cluster_addr
[i
] && o
->osd_addrs
->cluster_addr
[i
] &&
1462 *n
->osd_addrs
->cluster_addr
[i
] == *o
->osd_addrs
->cluster_addr
[i
])
1463 n
->osd_addrs
->cluster_addr
[i
] = o
->osd_addrs
->cluster_addr
[i
];
1466 if ( n
->osd_addrs
->hb_back_addr
[i
] && o
->osd_addrs
->hb_back_addr
[i
] &&
1467 *n
->osd_addrs
->hb_back_addr
[i
] == *o
->osd_addrs
->hb_back_addr
[i
])
1468 n
->osd_addrs
->hb_back_addr
[i
] = o
->osd_addrs
->hb_back_addr
[i
];
1471 if ( n
->osd_addrs
->hb_front_addr
[i
] && o
->osd_addrs
->hb_front_addr
[i
] &&
1472 *n
->osd_addrs
->hb_front_addr
[i
] == *o
->osd_addrs
->hb_front_addr
[i
])
1473 n
->osd_addrs
->hb_front_addr
[i
] = o
->osd_addrs
->hb_front_addr
[i
];
1478 // zoinks, no differences at all!
1479 n
->osd_addrs
= o
->osd_addrs
;
1482 // does crush match?
1484 ::encode(*o
->crush
, oc
, CEPH_FEATURES_SUPPORTED_DEFAULT
);
1485 ::encode(*n
->crush
, nc
, CEPH_FEATURES_SUPPORTED_DEFAULT
);
1486 if (oc
.contents_equal(nc
)) {
1487 n
->crush
= o
->crush
;
1490 // does pg_temp match?
1491 if (*o
->pg_temp
== *n
->pg_temp
)
1492 n
->pg_temp
= o
->pg_temp
;
1494 // does primary_temp match?
1495 if (o
->primary_temp
->size() == n
->primary_temp
->size()) {
1496 if (*o
->primary_temp
== *n
->primary_temp
)
1497 n
->primary_temp
= o
->primary_temp
;
1501 if (o
->osd_uuid
->size() == n
->osd_uuid
->size() &&
1502 *o
->osd_uuid
== *n
->osd_uuid
)
1503 n
->osd_uuid
= o
->osd_uuid
;
1506 void OSDMap::clean_temps(CephContext
*cct
,
1507 const OSDMap
& osdmap
, Incremental
*pending_inc
)
1509 ldout(cct
, 10) << __func__
<< dendl
;
1511 tmpmap
.deepish_copy_from(osdmap
);
1512 tmpmap
.apply_incremental(*pending_inc
);
1514 for (auto pg
: *tmpmap
.pg_temp
) {
1515 // if pool does not exist, remove any existing pg_temps associated with
1516 // it. we don't care about pg_temps on the pending_inc either; if there
1517 // are new_pg_temp entries on the pending, clear them out just as well.
1518 if (!osdmap
.have_pg_pool(pg
.first
.pool())) {
1519 ldout(cct
, 10) << __func__
<< " removing pg_temp " << pg
.first
1520 << " for nonexistent pool " << pg
.first
.pool() << dendl
;
1521 pending_inc
->new_pg_temp
[pg
.first
].clear();
1525 unsigned num_up
= 0;
1526 for (auto o
: pg
.second
) {
1527 if (!tmpmap
.is_down(o
)) {
1533 ldout(cct
, 10) << __func__
<< " removing pg_temp " << pg
.first
1534 << " with all down osds" << pg
.second
<< dendl
;
1535 pending_inc
->new_pg_temp
[pg
.first
].clear();
1538 // redundant pg_temp?
1541 tmpmap
.pg_to_raw_up(pg
.first
, &raw_up
, &primary
);
1542 if (vectors_equal(raw_up
, pg
.second
)) {
1543 ldout(cct
, 10) << __func__
<< " removing pg_temp " << pg
.first
<< " "
1544 << pg
.second
<< " that matches raw_up mapping" << dendl
;
1545 if (osdmap
.pg_temp
->count(pg
.first
))
1546 pending_inc
->new_pg_temp
[pg
.first
].clear();
1548 pending_inc
->new_pg_temp
.erase(pg
.first
);
1552 for (auto &pg
: *tmpmap
.primary_temp
) {
1554 if (tmpmap
.is_down(pg
.second
)) {
1555 ldout(cct
, 10) << __func__
<< " removing primary_temp " << pg
.first
1556 << " to down " << pg
.second
<< dendl
;
1557 pending_inc
->new_primary_temp
[pg
.first
] = -1;
1560 // redundant primary_temp?
1561 vector
<int> real_up
, templess_up
;
1562 int real_primary
, templess_primary
;
1563 pg_t pgid
= pg
.first
;
1564 tmpmap
.pg_to_acting_osds(pgid
, &real_up
, &real_primary
);
1565 tmpmap
.pg_to_raw_up(pgid
, &templess_up
, &templess_primary
);
1566 if (real_primary
== templess_primary
){
1567 ldout(cct
, 10) << __func__
<< " removing primary_temp "
1568 << pgid
<< " -> " << real_primary
1569 << " (unnecessary/redundant)" << dendl
;
1570 if (osdmap
.primary_temp
->count(pgid
))
1571 pending_inc
->new_primary_temp
[pgid
] = -1;
1573 pending_inc
->new_primary_temp
.erase(pgid
);
1578 int OSDMap::apply_incremental(const Incremental
&inc
)
1580 new_blacklist_entries
= false;
1583 else if (inc
.fsid
!= fsid
)
1586 assert(inc
.epoch
== epoch
+1);
1589 modified
= inc
.modified
;
1592 if (inc
.fullmap
.length()) {
1593 bufferlist
bl(inc
.fullmap
);
1598 // nope, incremental.
1599 if (inc
.new_flags
>= 0) {
1600 flags
= inc
.new_flags
;
1601 // the below is just to cover a newly-upgraded luminous mon
1602 // cluster that has to set require_jewel_osds or
1603 // require_kraken_osds before the osds can be upgraded to
1605 if (flags
& CEPH_OSDMAP_REQUIRE_KRAKEN
) {
1606 if (require_osd_release
< CEPH_RELEASE_KRAKEN
) {
1607 require_osd_release
= CEPH_RELEASE_KRAKEN
;
1609 } else if (flags
& CEPH_OSDMAP_REQUIRE_JEWEL
) {
1610 if (require_osd_release
< CEPH_RELEASE_JEWEL
) {
1611 require_osd_release
= CEPH_RELEASE_JEWEL
;
1616 if (inc
.new_max_osd
>= 0)
1617 set_max_osd(inc
.new_max_osd
);
1619 if (inc
.new_pool_max
!= -1)
1620 pool_max
= inc
.new_pool_max
;
1622 for (const auto &pool
: inc
.new_pools
) {
1623 pools
[pool
.first
] = pool
.second
;
1624 pools
[pool
.first
].last_change
= epoch
;
1627 for (const auto &pname
: inc
.new_pool_names
) {
1628 auto pool_name_entry
= pool_name
.find(pname
.first
);
1629 if (pool_name_entry
!= pool_name
.end()) {
1630 name_pool
.erase(pool_name_entry
->second
);
1631 pool_name_entry
->second
= pname
.second
;
1633 pool_name
[pname
.first
] = pname
.second
;
1635 name_pool
[pname
.second
] = pname
.first
;
1638 for (const auto &pool
: inc
.old_pools
) {
1640 name_pool
.erase(pool_name
[pool
]);
1641 pool_name
.erase(pool
);
1644 for (const auto &weight
: inc
.new_weight
) {
1645 set_weight(weight
.first
, weight
.second
);
1647 // if we are marking in, clear the AUTOOUT and NEW bits, and clear
1648 // xinfo old_weight.
1649 if (weight
.second
) {
1650 osd_state
[weight
.first
] &= ~(CEPH_OSD_AUTOOUT
| CEPH_OSD_NEW
);
1651 osd_xinfo
[weight
.first
].old_weight
= 0;
1655 for (const auto &primary_affinity
: inc
.new_primary_affinity
) {
1656 set_primary_affinity(primary_affinity
.first
, primary_affinity
.second
);
1659 // erasure_code_profiles
1660 for (const auto &profile
: inc
.old_erasure_code_profiles
)
1661 erasure_code_profiles
.erase(profile
);
1663 for (const auto &profile
: inc
.new_erasure_code_profiles
) {
1664 set_erasure_code_profile(profile
.first
, profile
.second
);
1668 for (const auto &state
: inc
.new_state
) {
1669 const auto osd
= state
.first
;
1670 int s
= state
.second
? state
.second
: CEPH_OSD_UP
;
1671 if ((osd_state
[osd
] & CEPH_OSD_UP
) &&
1672 (s
& CEPH_OSD_UP
)) {
1673 osd_info
[osd
].down_at
= epoch
;
1674 osd_xinfo
[osd
].down_stamp
= modified
;
1676 if ((osd_state
[osd
] & CEPH_OSD_EXISTS
) &&
1677 (s
& CEPH_OSD_EXISTS
)) {
1678 // osd is destroyed; clear out anything interesting.
1679 (*osd_uuid
)[osd
] = uuid_d();
1680 osd_info
[osd
] = osd_info_t();
1681 osd_xinfo
[osd
] = osd_xinfo_t();
1682 set_primary_affinity(osd
, CEPH_OSD_DEFAULT_PRIMARY_AFFINITY
);
1683 osd_addrs
->client_addr
[osd
].reset(new entity_addr_t());
1684 osd_addrs
->cluster_addr
[osd
].reset(new entity_addr_t());
1685 osd_addrs
->hb_front_addr
[osd
].reset(new entity_addr_t());
1686 osd_addrs
->hb_back_addr
[osd
].reset(new entity_addr_t());
1689 osd_state
[osd
] ^= s
;
1693 for (const auto &client
: inc
.new_up_client
) {
1694 osd_state
[client
.first
] |= CEPH_OSD_EXISTS
| CEPH_OSD_UP
;
1695 osd_addrs
->client_addr
[client
.first
].reset(new entity_addr_t(client
.second
));
1696 if (inc
.new_hb_back_up
.empty())
1697 osd_addrs
->hb_back_addr
[client
.first
].reset(new entity_addr_t(client
.second
)); //this is a backward-compatibility hack
1699 osd_addrs
->hb_back_addr
[client
.first
].reset(
1700 new entity_addr_t(inc
.new_hb_back_up
.find(client
.first
)->second
));
1701 const auto j
= inc
.new_hb_front_up
.find(client
.first
);
1702 if (j
!= inc
.new_hb_front_up
.end())
1703 osd_addrs
->hb_front_addr
[client
.first
].reset(new entity_addr_t(j
->second
));
1705 osd_addrs
->hb_front_addr
[client
.first
].reset();
1707 osd_info
[client
.first
].up_from
= epoch
;
1710 for (const auto &cluster
: inc
.new_up_cluster
)
1711 osd_addrs
->cluster_addr
[cluster
.first
].reset(new entity_addr_t(cluster
.second
));
1714 for (const auto &thru
: inc
.new_up_thru
)
1715 osd_info
[thru
.first
].up_thru
= thru
.second
;
1717 for (const auto &interval
: inc
.new_last_clean_interval
) {
1718 osd_info
[interval
.first
].last_clean_begin
= interval
.second
.first
;
1719 osd_info
[interval
.first
].last_clean_end
= interval
.second
.second
;
1722 for (const auto &lost
: inc
.new_lost
)
1723 osd_info
[lost
.first
].lost_at
= lost
.second
;
1726 for (const auto &xinfo
: inc
.new_xinfo
)
1727 osd_xinfo
[xinfo
.first
] = xinfo
.second
;
1730 for (const auto &uuid
: inc
.new_uuid
)
1731 (*osd_uuid
)[uuid
.first
] = uuid
.second
;
1734 for (const auto &pg
: inc
.new_pg_temp
) {
1735 if (pg
.second
.empty())
1736 pg_temp
->erase(pg
.first
);
1738 pg_temp
->set(pg
.first
, pg
.second
);
1740 if (!inc
.new_pg_temp
.empty()) {
1741 // make sure pg_temp is efficiently stored
1745 for (const auto &pg
: inc
.new_primary_temp
) {
1746 if (pg
.second
== -1)
1747 primary_temp
->erase(pg
.first
);
1749 (*primary_temp
)[pg
.first
] = pg
.second
;
1752 for (auto& p
: inc
.new_pg_upmap
) {
1753 pg_upmap
[p
.first
] = p
.second
;
1755 for (auto& pg
: inc
.old_pg_upmap
) {
1758 for (auto& p
: inc
.new_pg_upmap_items
) {
1759 pg_upmap_items
[p
.first
] = p
.second
;
1761 for (auto& pg
: inc
.old_pg_upmap_items
) {
1762 pg_upmap_items
.erase(pg
);
1766 if (!inc
.new_blacklist
.empty()) {
1767 blacklist
.insert(inc
.new_blacklist
.begin(),inc
.new_blacklist
.end());
1768 new_blacklist_entries
= true;
1770 for (const auto &addr
: inc
.old_blacklist
)
1771 blacklist
.erase(addr
);
1773 // cluster snapshot?
1774 if (inc
.cluster_snapshot
.length()) {
1775 cluster_snapshot
= inc
.cluster_snapshot
;
1776 cluster_snapshot_epoch
= inc
.epoch
;
1778 cluster_snapshot
.clear();
1779 cluster_snapshot_epoch
= 0;
1782 if (inc
.new_nearfull_ratio
>= 0) {
1783 nearfull_ratio
= inc
.new_nearfull_ratio
;
1785 if (inc
.new_backfillfull_ratio
>= 0) {
1786 backfillfull_ratio
= inc
.new_backfillfull_ratio
;
1788 if (inc
.new_full_ratio
>= 0) {
1789 full_ratio
= inc
.new_full_ratio
;
1791 if (inc
.new_require_min_compat_client
> 0) {
1792 require_min_compat_client
= inc
.new_require_min_compat_client
;
1794 if (inc
.new_require_osd_release
>= 0) {
1795 require_osd_release
= inc
.new_require_osd_release
;
1796 if (require_osd_release
>= CEPH_RELEASE_LUMINOUS
) {
1797 flags
&= ~(CEPH_OSDMAP_LEGACY_REQUIRE_FLAGS
);
1801 // do new crush map last (after up/down stuff)
1802 if (inc
.crush
.length()) {
1803 bufferlist
bl(inc
.crush
);
1804 auto blp
= bl
.begin();
1805 crush
.reset(new CrushWrapper
);
1807 if (require_osd_release
>= CEPH_RELEASE_LUMINOUS
) {
1808 // only increment if this is a luminous-encoded osdmap, lest
1809 // the mon's crush_version diverge from what the osds or others
1810 // are decoding and applying on their end. if we won't encode
1811 // it in the canonical version, don't change it.
1817 _calc_up_osd_features();
1822 int OSDMap::map_to_pg(
1826 const string
& nspace
,
1829 // calculate ps (placement seed)
1830 const pg_pool_t
*pool
= get_pg_pool(poolid
);
1835 ps
= pool
->hash_key(key
, nspace
);
1837 ps
= pool
->hash_key(name
, nspace
);
1838 *pg
= pg_t(ps
, poolid
);
1842 int OSDMap::object_locator_to_pg(
1843 const object_t
& oid
, const object_locator_t
& loc
, pg_t
&pg
) const
1845 if (loc
.hash
>= 0) {
1846 if (!get_pg_pool(loc
.get_pool())) {
1849 pg
= pg_t(loc
.hash
, loc
.get_pool());
1852 return map_to_pg(loc
.get_pool(), oid
.name
, loc
.key
, loc
.nspace
, &pg
);
1855 ceph_object_layout
OSDMap::make_object_layout(
1856 object_t oid
, int pg_pool
, string nspace
) const
1858 object_locator_t
loc(pg_pool
, nspace
);
1860 ceph_object_layout ol
;
1861 pg_t pgid
= object_locator_to_pg(oid
, loc
);
1862 ol
.ol_pgid
= pgid
.get_old_pg().v
;
1863 ol
.ol_stripe_unit
= 0;
1867 void OSDMap::_remove_nonexistent_osds(const pg_pool_t
& pool
,
1868 vector
<int>& osds
) const
1870 if (pool
.can_shift_osds()) {
1871 unsigned removed
= 0;
1872 for (unsigned i
= 0; i
< osds
.size(); i
++) {
1873 if (!exists(osds
[i
])) {
1878 osds
[i
- removed
] = osds
[i
];
1882 osds
.resize(osds
.size() - removed
);
1884 for (auto& osd
: osds
) {
1886 osd
= CRUSH_ITEM_NONE
;
1891 void OSDMap::_pg_to_raw_osds(
1892 const pg_pool_t
& pool
, pg_t pg
,
1897 ps_t pps
= pool
.raw_pg_to_pps(pg
); // placement ps
1898 unsigned size
= pool
.get_size();
1901 int ruleno
= crush
->find_rule(pool
.get_crush_rule(), pool
.get_type(), size
);
1903 crush
->do_rule(ruleno
, pps
, *osds
, size
, osd_weight
, pg
.pool());
1905 _remove_nonexistent_osds(pool
, *osds
);
1911 int OSDMap::_pick_primary(const vector
<int>& osds
) const
1913 for (auto osd
: osds
) {
1914 if (osd
!= CRUSH_ITEM_NONE
) {
1921 void OSDMap::_apply_upmap(const pg_pool_t
& pi
, pg_t raw_pg
, vector
<int> *raw
) const
1923 pg_t pg
= pi
.raw_pg_to_pg(raw_pg
);
1924 auto p
= pg_upmap
.find(pg
);
1925 if (p
!= pg_upmap
.end()) {
1926 // make sure targets aren't marked out
1927 for (auto osd
: p
->second
) {
1928 if (osd
!= CRUSH_ITEM_NONE
&& osd
< max_osd
&& osd_weight
[osd
] == 0) {
1929 // reject/ignore the explicit mapping
1933 *raw
= vector
<int>(p
->second
.begin(), p
->second
.end());
1934 // continue to check and apply pg_upmap_items if any
1937 auto q
= pg_upmap_items
.find(pg
);
1938 if (q
!= pg_upmap_items
.end()) {
1939 // NOTE: this approach does not allow a bidirectional swap,
1940 // e.g., [[1,2],[2,1]] applied to [0,1,2] -> [0,2,1].
1941 for (auto& r
: q
->second
) {
1942 // make sure the replacement value doesn't already appear
1943 bool exists
= false;
1945 for (unsigned i
= 0; i
< raw
->size(); ++i
) {
1946 int osd
= (*raw
)[i
];
1947 if (osd
== r
.second
) {
1951 // ignore mapping if target is marked out (or invalid osd id)
1952 if (osd
== r
.first
&&
1954 !(r
.second
!= CRUSH_ITEM_NONE
&& r
.second
< max_osd
&&
1955 osd_weight
[r
.second
] == 0)) {
1959 if (!exists
&& pos
>= 0) {
1960 (*raw
)[pos
] = r
.second
;
1966 // pg -> (up osd list)
1967 void OSDMap::_raw_to_up_osds(const pg_pool_t
& pool
, const vector
<int>& raw
,
1968 vector
<int> *up
) const
1970 if (pool
.can_shift_osds()) {
1973 up
->reserve(raw
.size());
1974 for (unsigned i
=0; i
<raw
.size(); i
++) {
1975 if (!exists(raw
[i
]) || is_down(raw
[i
]))
1977 up
->push_back(raw
[i
]);
1980 // set down/dne devices to NONE
1981 up
->resize(raw
.size());
1982 for (int i
= raw
.size() - 1; i
>= 0; --i
) {
1983 if (!exists(raw
[i
]) || is_down(raw
[i
])) {
1984 (*up
)[i
] = CRUSH_ITEM_NONE
;
1992 void OSDMap::_apply_primary_affinity(ps_t seed
,
1993 const pg_pool_t
& pool
,
1997 // do we have any non-default primary_affinity values for these osds?
1998 if (!osd_primary_affinity
)
2002 for (const auto osd
: *osds
) {
2003 if (osd
!= CRUSH_ITEM_NONE
&&
2004 (*osd_primary_affinity
)[osd
] != CEPH_OSD_DEFAULT_PRIMARY_AFFINITY
) {
2012 // pick the primary. feed both the seed (for the pg) and the osd
2013 // into the hash/rng so that a proportional fraction of an osd's pgs
2014 // get rejected as primary.
2016 for (unsigned i
= 0; i
< osds
->size(); ++i
) {
2018 if (o
== CRUSH_ITEM_NONE
)
2020 unsigned a
= (*osd_primary_affinity
)[o
];
2021 if (a
< CEPH_OSD_MAX_PRIMARY_AFFINITY
&&
2022 (crush_hash32_2(CRUSH_HASH_RJENKINS1
,
2023 seed
, o
) >> 16) >= a
) {
2024 // we chose not to use this primary. note it anyway as a
2025 // fallback in case we don't pick anyone else, but keep looking.
2036 *primary
= (*osds
)[pos
];
2038 if (pool
.can_shift_osds() && pos
> 0) {
2039 // move the new primary to the front.
2040 for (int i
= pos
; i
> 0; --i
) {
2041 (*osds
)[i
] = (*osds
)[i
-1];
2043 (*osds
)[0] = *primary
;
2047 void OSDMap::_get_temp_osds(const pg_pool_t
& pool
, pg_t pg
,
2048 vector
<int> *temp_pg
, int *temp_primary
) const
2050 pg
= pool
.raw_pg_to_pg(pg
);
2051 const auto p
= pg_temp
->find(pg
);
2053 if (p
!= pg_temp
->end()) {
2054 for (unsigned i
=0; i
<p
->second
.size(); i
++) {
2055 if (!exists(p
->second
[i
]) || is_down(p
->second
[i
])) {
2056 if (pool
.can_shift_osds()) {
2059 temp_pg
->push_back(CRUSH_ITEM_NONE
);
2062 temp_pg
->push_back(p
->second
[i
]);
2066 const auto &pp
= primary_temp
->find(pg
);
2068 if (pp
!= primary_temp
->end()) {
2069 *temp_primary
= pp
->second
;
2070 } else if (!temp_pg
->empty()) { // apply pg_temp's primary
2071 for (unsigned i
= 0; i
< temp_pg
->size(); ++i
) {
2072 if ((*temp_pg
)[i
] != CRUSH_ITEM_NONE
) {
2073 *temp_primary
= (*temp_pg
)[i
];
2080 void OSDMap::pg_to_raw_osds(pg_t pg
, vector
<int> *raw
, int *primary
) const
2084 const pg_pool_t
*pool
= get_pg_pool(pg
.pool());
2087 _pg_to_raw_osds(*pool
, pg
, raw
, NULL
);
2089 *primary
= _pick_primary(*raw
);
2092 void OSDMap::pg_to_raw_up(pg_t pg
, vector
<int> *up
, int *primary
) const
2094 const pg_pool_t
*pool
= get_pg_pool(pg
.pool());
2104 _pg_to_raw_osds(*pool
, pg
, &raw
, &pps
);
2105 _apply_upmap(*pool
, pg
, &raw
);
2106 _raw_to_up_osds(*pool
, raw
, up
);
2107 *primary
= _pick_primary(raw
);
2108 _apply_primary_affinity(pps
, *pool
, up
, primary
);
2111 void OSDMap::_pg_to_up_acting_osds(
2112 const pg_t
& pg
, vector
<int> *up
, int *up_primary
,
2113 vector
<int> *acting
, int *acting_primary
,
2114 bool raw_pg_to_pg
) const
2116 const pg_pool_t
*pool
= get_pg_pool(pg
.pool());
2118 (!raw_pg_to_pg
&& pg
.ps() >= pool
->get_pg_num())) {
2126 *acting_primary
= -1;
2131 vector
<int> _acting
;
2133 int _acting_primary
;
2135 _get_temp_osds(*pool
, pg
, &_acting
, &_acting_primary
);
2136 if (_acting
.empty() || up
|| up_primary
) {
2137 _pg_to_raw_osds(*pool
, pg
, &raw
, &pps
);
2138 _apply_upmap(*pool
, pg
, &raw
);
2139 _raw_to_up_osds(*pool
, raw
, &_up
);
2140 _up_primary
= _pick_primary(_up
);
2141 _apply_primary_affinity(pps
, *pool
, &_up
, &_up_primary
);
2142 if (_acting
.empty()) {
2144 if (_acting_primary
== -1) {
2145 _acting_primary
= _up_primary
;
2152 *up_primary
= _up_primary
;
2156 acting
->swap(_acting
);
2158 *acting_primary
= _acting_primary
;
2161 int OSDMap::calc_pg_rank(int osd
, const vector
<int>& acting
, int nrep
)
2164 nrep
= acting
.size();
2165 for (int i
=0; i
<nrep
; i
++)
2166 if (acting
[i
] == osd
)
2171 int OSDMap::calc_pg_role(int osd
, const vector
<int>& acting
, int nrep
)
2173 return calc_pg_rank(osd
, acting
, nrep
);
2176 bool OSDMap::primary_changed(
2178 const vector
<int> &oldacting
,
2180 const vector
<int> &newacting
)
2182 if (oldacting
.empty() && newacting
.empty())
2183 return false; // both still empty
2184 if (oldacting
.empty() ^ newacting
.empty())
2185 return true; // was empty, now not, or vice versa
2186 if (oldprimary
!= newprimary
)
2187 return true; // primary changed
2188 if (calc_pg_rank(oldprimary
, oldacting
) !=
2189 calc_pg_rank(newprimary
, newacting
))
2191 return false; // same primary (tho replicas may have changed)
2195 // serialize, unserialize
2196 void OSDMap::encode_client_old(bufferlist
& bl
) const
2203 ::encode(epoch
, bl
);
2204 ::encode(created
, bl
);
2205 ::encode(modified
, bl
);
2207 // for ::encode(pools, bl);
2208 __u32 n
= pools
.size();
2211 for (const auto &pool
: pools
) {
2214 ::encode(pool
.second
, bl
, 0);
2216 // for ::encode(pool_name, bl);
2217 n
= pool_name
.size();
2219 for (const auto &pname
: pool_name
) {
2222 ::encode(pname
.second
, bl
);
2224 // for ::encode(pool_max, bl);
2228 ::encode(flags
, bl
);
2230 ::encode(max_osd
, bl
);
2232 uint32_t n
= osd_state
.size();
2234 for (auto s
: osd_state
) {
2235 ::encode((uint8_t)s
, bl
);
2238 ::encode(osd_weight
, bl
);
2239 ::encode(osd_addrs
->client_addr
, bl
, 0);
2241 // for ::encode(pg_temp, bl);
2242 n
= pg_temp
->size();
2244 for (const auto pg
: *pg_temp
) {
2245 old_pg_t opg
= pg
.first
.get_old_pg();
2247 ::encode(pg
.second
, bl
);
2252 crush
->encode(cbl
, 0 /* legacy (no) features */);
2256 void OSDMap::encode_classic(bufferlist
& bl
, uint64_t features
) const
2258 if ((features
& CEPH_FEATURE_PGID64
) == 0) {
2259 encode_client_old(bl
);
2268 ::encode(epoch
, bl
);
2269 ::encode(created
, bl
);
2270 ::encode(modified
, bl
);
2272 ::encode(pools
, bl
, features
);
2273 ::encode(pool_name
, bl
);
2274 ::encode(pool_max
, bl
);
2276 ::encode(flags
, bl
);
2278 ::encode(max_osd
, bl
);
2280 uint32_t n
= osd_state
.size();
2282 for (auto s
: osd_state
) {
2283 ::encode((uint8_t)s
, bl
);
2286 ::encode(osd_weight
, bl
);
2287 ::encode(osd_addrs
->client_addr
, bl
, features
);
2289 ::encode(*pg_temp
, bl
);
2293 crush
->encode(cbl
, 0 /* legacy (no) features */);
2299 ::encode(osd_addrs
->hb_back_addr
, bl
, features
);
2300 ::encode(osd_info
, bl
);
2301 ::encode(blacklist
, bl
, features
);
2302 ::encode(osd_addrs
->cluster_addr
, bl
, features
);
2303 ::encode(cluster_snapshot_epoch
, bl
);
2304 ::encode(cluster_snapshot
, bl
);
2305 ::encode(*osd_uuid
, bl
);
2306 ::encode(osd_xinfo
, bl
);
2307 ::encode(osd_addrs
->hb_front_addr
, bl
, features
);
2310 void OSDMap::encode(bufferlist
& bl
, uint64_t features
) const
2312 if ((features
& CEPH_FEATURE_OSDMAP_ENC
) == 0) {
2313 encode_classic(bl
, features
);
2317 // only a select set of callers should *ever* be encoding new
2318 // OSDMaps. others should be passing around the canonical encoded
2319 // buffers from on high. select out those callers by passing in an
2320 // "impossible" feature bit.
2321 assert(features
& CEPH_FEATURE_RESERVED
);
2322 features
&= ~CEPH_FEATURE_RESERVED
;
2324 size_t start_offset
= bl
.length();
2326 buffer::list::iterator crc_it
;
2328 // meta-encoding: how we include client-used and osd-specific data
2329 ENCODE_START(8, 7, bl
);
2333 if (!HAVE_FEATURE(features
, SERVER_LUMINOUS
)) {
2336 ENCODE_START(v
, 1, bl
); // client-usable data
2339 ::encode(epoch
, bl
);
2340 ::encode(created
, bl
);
2341 ::encode(modified
, bl
);
2343 ::encode(pools
, bl
, features
);
2344 ::encode(pool_name
, bl
);
2345 ::encode(pool_max
, bl
);
2348 decltype(flags
) f
= flags
;
2349 if (require_osd_release
>= CEPH_RELEASE_LUMINOUS
)
2350 f
|= CEPH_OSDMAP_REQUIRE_LUMINOUS
;
2351 else if (require_osd_release
== CEPH_RELEASE_KRAKEN
)
2352 f
|= CEPH_OSDMAP_REQUIRE_KRAKEN
;
2353 else if (require_osd_release
== CEPH_RELEASE_JEWEL
)
2354 f
|= CEPH_OSDMAP_REQUIRE_JEWEL
;
2357 ::encode(flags
, bl
);
2360 ::encode(max_osd
, bl
);
2362 ::encode(osd_state
, bl
);
2364 uint32_t n
= osd_state
.size();
2366 for (auto s
: osd_state
) {
2367 ::encode((uint8_t)s
, bl
);
2370 ::encode(osd_weight
, bl
);
2371 ::encode(osd_addrs
->client_addr
, bl
, features
);
2373 ::encode(*pg_temp
, bl
);
2374 ::encode(*primary_temp
, bl
);
2375 if (osd_primary_affinity
) {
2376 ::encode(*osd_primary_affinity
, bl
);
2384 crush
->encode(cbl
, features
);
2386 ::encode(erasure_code_profiles
, bl
);
2389 ::encode(pg_upmap
, bl
);
2390 ::encode(pg_upmap_items
, bl
);
2392 assert(pg_upmap
.empty());
2393 assert(pg_upmap_items
.empty());
2396 ::encode(crush_version
, bl
);
2398 ENCODE_FINISH(bl
); // client-usable data
2402 uint8_t target_v
= 5;
2403 if (!HAVE_FEATURE(features
, SERVER_LUMINOUS
)) {
2406 ENCODE_START(target_v
, 1, bl
); // extended, osd-only data
2407 ::encode(osd_addrs
->hb_back_addr
, bl
, features
);
2408 ::encode(osd_info
, bl
);
2410 // put this in a sorted, ordered map<> so that we encode in a
2411 // deterministic order.
2412 map
<entity_addr_t
,utime_t
> blacklist_map
;
2413 for (const auto &addr
: blacklist
)
2414 blacklist_map
.insert(make_pair(addr
.first
, addr
.second
));
2415 ::encode(blacklist_map
, bl
, features
);
2417 ::encode(osd_addrs
->cluster_addr
, bl
, features
);
2418 ::encode(cluster_snapshot_epoch
, bl
);
2419 ::encode(cluster_snapshot
, bl
);
2420 ::encode(*osd_uuid
, bl
);
2421 ::encode(osd_xinfo
, bl
);
2422 ::encode(osd_addrs
->hb_front_addr
, bl
, features
);
2423 if (target_v
>= 2) {
2424 ::encode(nearfull_ratio
, bl
);
2425 ::encode(full_ratio
, bl
);
2426 ::encode(backfillfull_ratio
, bl
);
2428 // 4 was string-based new_require_min_compat_client
2429 if (target_v
>= 5) {
2430 ::encode(require_min_compat_client
, bl
);
2431 ::encode(require_osd_release
, bl
);
2433 ENCODE_FINISH(bl
); // osd-only data
2436 ::encode((uint32_t)0, bl
); // dummy crc
2439 tail_offset
= bl
.length();
2441 ENCODE_FINISH(bl
); // meta-encoding wrapper
2445 front
.substr_of(bl
, start_offset
, crc_it
.get_off() - start_offset
);
2446 crc
= front
.crc32c(-1);
2447 if (tail_offset
< bl
.length()) {
2449 tail
.substr_of(bl
, tail_offset
, bl
.length() - tail_offset
);
2450 crc
= tail
.crc32c(crc
);
2454 crc_it
.copy_in(4, (char*)&crc_le
);
2458 void OSDMap::decode(bufferlist
& bl
)
2460 auto p
= bl
.begin();
2464 void OSDMap::decode_classic(bufferlist::iterator
& p
)
2473 ::decode(created
, p
);
2474 ::decode(modified
, p
);
2478 int32_t max_pools
= 0;
2479 ::decode(max_pools
, p
);
2480 pool_max
= max_pools
;
2486 ::decode(pools
[t
], p
);
2491 } else if (v
== 5) {
2496 ::decode(pool_name
[t
], p
);
2503 ::decode(pool_name
, p
);
2504 ::decode(pool_max
, p
);
2506 // kludge around some old bug that zeroed out pool_max (#2307)
2507 if (pools
.size() && pool_max
< pools
.rbegin()->first
) {
2508 pool_max
= pools
.rbegin()->first
;
2513 ::decode(max_osd
, p
);
2517 osd_state
.resize(os
.size());
2518 for (unsigned i
= 0; i
< os
.size(); ++i
) {
2519 osd_state
[i
] = os
[i
];
2522 ::decode(osd_weight
, p
);
2523 ::decode(osd_addrs
->client_addr
, p
);
2529 ::decode_raw(opg
, p
);
2530 mempool::osdmap::vector
<int32_t> v
;
2532 pg_temp
->set(pg_t(opg
), v
);
2535 ::decode(*pg_temp
, p
);
2541 auto cblp
= cbl
.begin();
2542 crush
->decode(cblp
);
2548 ::decode(osd_addrs
->hb_back_addr
, p
);
2549 ::decode(osd_info
, p
);
2551 ::decode(pool_name
, p
);
2553 ::decode(blacklist
, p
);
2555 ::decode(osd_addrs
->cluster_addr
, p
);
2557 osd_addrs
->cluster_addr
.resize(osd_addrs
->client_addr
.size());
2560 ::decode(cluster_snapshot_epoch
, p
);
2561 ::decode(cluster_snapshot
, p
);
2565 ::decode(*osd_uuid
, p
);
2567 osd_uuid
->resize(max_osd
);
2570 ::decode(osd_xinfo
, p
);
2572 osd_xinfo
.resize(max_osd
);
2575 ::decode(osd_addrs
->hb_front_addr
, p
);
2577 osd_addrs
->hb_front_addr
.resize(osd_addrs
->hb_back_addr
.size());
2579 osd_primary_affinity
.reset();
2584 void OSDMap::decode(bufferlist::iterator
& bl
)
2587 * Older encodings of the OSDMap had a single struct_v which
2588 * covered the whole encoding, and was prior to our modern
2589 * stuff which includes a compatv and a size. So if we see
2590 * a struct_v < 7, we must rewind to the beginning and use our
2593 size_t start_offset
= bl
.get_off();
2594 size_t tail_offset
= 0;
2595 bufferlist crc_front
, crc_tail
;
2597 DECODE_START_LEGACY_COMPAT_LEN(8, 7, 7, bl
); // wrapper
2599 int struct_v_size
= sizeof(struct_v
);
2600 bl
.advance(-struct_v_size
);
2605 * Since we made it past that hurdle, we can use our normal paths.
2608 DECODE_START(6, bl
); // client-usable data
2611 ::decode(epoch
, bl
);
2612 ::decode(created
, bl
);
2613 ::decode(modified
, bl
);
2615 ::decode(pools
, bl
);
2616 ::decode(pool_name
, bl
);
2617 ::decode(pool_max
, bl
);
2619 ::decode(flags
, bl
);
2621 ::decode(max_osd
, bl
);
2622 if (struct_v
>= 5) {
2623 ::decode(osd_state
, bl
);
2627 osd_state
.resize(os
.size());
2628 for (unsigned i
= 0; i
< os
.size(); ++i
) {
2629 osd_state
[i
] = os
[i
];
2632 ::decode(osd_weight
, bl
);
2633 ::decode(osd_addrs
->client_addr
, bl
);
2635 ::decode(*pg_temp
, bl
);
2636 ::decode(*primary_temp
, bl
);
2637 if (struct_v
>= 2) {
2638 osd_primary_affinity
.reset(new mempool::osdmap::vector
<__u32
>);
2639 ::decode(*osd_primary_affinity
, bl
);
2640 if (osd_primary_affinity
->empty())
2641 osd_primary_affinity
.reset();
2643 osd_primary_affinity
.reset();
2649 auto cblp
= cbl
.begin();
2650 crush
->decode(cblp
);
2651 if (struct_v
>= 3) {
2652 ::decode(erasure_code_profiles
, bl
);
2654 erasure_code_profiles
.clear();
2656 if (struct_v
>= 4) {
2657 ::decode(pg_upmap
, bl
);
2658 ::decode(pg_upmap_items
, bl
);
2661 pg_upmap_items
.clear();
2663 if (struct_v
>= 6) {
2664 ::decode(crush_version
, bl
);
2666 DECODE_FINISH(bl
); // client-usable data
2670 DECODE_START(5, bl
); // extended, osd-only data
2671 ::decode(osd_addrs
->hb_back_addr
, bl
);
2672 ::decode(osd_info
, bl
);
2673 ::decode(blacklist
, bl
);
2674 ::decode(osd_addrs
->cluster_addr
, bl
);
2675 ::decode(cluster_snapshot_epoch
, bl
);
2676 ::decode(cluster_snapshot
, bl
);
2677 ::decode(*osd_uuid
, bl
);
2678 ::decode(osd_xinfo
, bl
);
2679 ::decode(osd_addrs
->hb_front_addr
, bl
);
2680 if (struct_v
>= 2) {
2681 ::decode(nearfull_ratio
, bl
);
2682 ::decode(full_ratio
, bl
);
2687 if (struct_v
>= 3) {
2688 ::decode(backfillfull_ratio
, bl
);
2690 backfillfull_ratio
= 0;
2692 if (struct_v
== 4) {
2696 require_min_compat_client
= ceph_release_from_name(r
.c_str());
2698 if (struct_v
>= 5) {
2699 ::decode(require_min_compat_client
, bl
);
2700 ::decode(require_osd_release
, bl
);
2701 if (require_osd_release
>= CEPH_RELEASE_LUMINOUS
) {
2702 flags
&= ~(CEPH_OSDMAP_LEGACY_REQUIRE_FLAGS
);
2705 if (flags
& CEPH_OSDMAP_REQUIRE_LUMINOUS
) {
2706 // only for compat with post-kraken pre-luminous test clusters
2707 require_osd_release
= CEPH_RELEASE_LUMINOUS
;
2708 flags
&= ~(CEPH_OSDMAP_LEGACY_REQUIRE_FLAGS
);
2709 } else if (flags
& CEPH_OSDMAP_REQUIRE_KRAKEN
) {
2710 require_osd_release
= CEPH_RELEASE_KRAKEN
;
2711 } else if (flags
& CEPH_OSDMAP_REQUIRE_JEWEL
) {
2712 require_osd_release
= CEPH_RELEASE_JEWEL
;
2714 require_osd_release
= 0;
2717 DECODE_FINISH(bl
); // osd-only data
2720 if (struct_v
>= 8) {
2721 crc_front
.substr_of(bl
.get_bl(), start_offset
, bl
.get_off() - start_offset
);
2723 tail_offset
= bl
.get_off();
2726 crc_defined
= false;
2730 DECODE_FINISH(bl
); // wrapper
2734 uint32_t actual
= crc_front
.crc32c(-1);
2735 if (tail_offset
< bl
.get_off()) {
2737 tail
.substr_of(bl
.get_bl(), tail_offset
, bl
.get_off() - tail_offset
);
2738 actual
= tail
.crc32c(actual
);
2740 if (crc
!= actual
) {
2742 ss
<< "bad crc, actual " << actual
<< " != expected " << crc
;
2743 string s
= ss
.str();
2744 throw buffer::malformed_input(s
.c_str());
2751 void OSDMap::post_decode()
2755 for (const auto &pname
: pool_name
) {
2756 name_pool
[pname
.second
] = pname
.first
;
2760 _calc_up_osd_features();
2763 void OSDMap::dump_erasure_code_profiles(
2764 const mempool::osdmap::map
<string
,map
<string
,string
>>& profiles
,
2767 f
->open_object_section("erasure_code_profiles");
2768 for (const auto &profile
: profiles
) {
2769 f
->open_object_section(profile
.first
.c_str());
2770 for (const auto &profm
: profile
.second
) {
2771 f
->dump_string(profm
.first
.c_str(), profm
.second
.c_str());
2778 void OSDMap::dump(Formatter
*f
) const
2780 f
->dump_int("epoch", get_epoch());
2781 f
->dump_stream("fsid") << get_fsid();
2782 f
->dump_stream("created") << get_created();
2783 f
->dump_stream("modified") << get_modified();
2784 f
->dump_string("flags", get_flag_string());
2785 f
->dump_unsigned("crush_version", get_crush_version());
2786 f
->dump_float("full_ratio", full_ratio
);
2787 f
->dump_float("backfillfull_ratio", backfillfull_ratio
);
2788 f
->dump_float("nearfull_ratio", nearfull_ratio
);
2789 f
->dump_string("cluster_snapshot", get_cluster_snapshot());
2790 f
->dump_int("pool_max", get_pool_max());
2791 f
->dump_int("max_osd", get_max_osd());
2792 f
->dump_string("require_min_compat_client",
2793 ceph_release_name(require_min_compat_client
));
2794 f
->dump_string("min_compat_client",
2795 ceph_release_name(get_min_compat_client()));
2796 f
->dump_string("require_osd_release",
2797 ceph_release_name(require_osd_release
));
2799 f
->open_array_section("pools");
2800 for (const auto &pool
: pools
) {
2801 std::string
name("<unknown>");
2802 const auto &pni
= pool_name
.find(pool
.first
);
2803 if (pni
!= pool_name
.end())
2805 f
->open_object_section("pool");
2806 f
->dump_int("pool", pool
.first
);
2807 f
->dump_string("pool_name", name
);
2808 pool
.second
.dump(f
);
2813 f
->open_array_section("osds");
2814 for (int i
=0; i
<get_max_osd(); i
++)
2816 f
->open_object_section("osd_info");
2817 f
->dump_int("osd", i
);
2818 f
->dump_stream("uuid") << get_uuid(i
);
2819 f
->dump_int("up", is_up(i
));
2820 f
->dump_int("in", is_in(i
));
2821 f
->dump_float("weight", get_weightf(i
));
2822 f
->dump_float("primary_affinity", get_primary_affinityf(i
));
2823 get_info(i
).dump(f
);
2824 f
->dump_stream("public_addr") << get_addr(i
);
2825 f
->dump_stream("cluster_addr") << get_cluster_addr(i
);
2826 f
->dump_stream("heartbeat_back_addr") << get_hb_back_addr(i
);
2827 f
->dump_stream("heartbeat_front_addr") << get_hb_front_addr(i
);
2831 f
->open_array_section("state");
2832 for (const auto &state
: st
)
2833 f
->dump_string("state", state
);
2840 f
->open_array_section("osd_xinfo");
2841 for (int i
=0; i
<get_max_osd(); i
++) {
2843 f
->open_object_section("xinfo");
2844 f
->dump_int("osd", i
);
2845 osd_xinfo
[i
].dump(f
);
2851 f
->open_array_section("pg_upmap");
2852 for (auto& p
: pg_upmap
) {
2853 f
->open_object_section("mapping");
2854 f
->dump_stream("pgid") << p
.first
;
2855 f
->open_array_section("osds");
2856 for (auto q
: p
.second
) {
2857 f
->dump_int("osd", q
);
2863 f
->open_array_section("pg_upmap_items");
2864 for (auto& p
: pg_upmap_items
) {
2865 f
->open_object_section("mapping");
2866 f
->dump_stream("pgid") << p
.first
;
2867 f
->open_array_section("mappings");
2868 for (auto& q
: p
.second
) {
2869 f
->open_object_section("mapping");
2870 f
->dump_int("from", q
.first
);
2871 f
->dump_int("to", q
.second
);
2878 f
->open_array_section("pg_temp");
2882 f
->open_array_section("primary_temp");
2883 for (const auto &pg
: *primary_temp
) {
2884 f
->dump_stream("pgid") << pg
.first
;
2885 f
->dump_int("osd", pg
.second
);
2887 f
->close_section(); // primary_temp
2889 f
->open_object_section("blacklist");
2890 for (const auto &addr
: blacklist
) {
2893 f
->dump_stream(ss
.str().c_str()) << addr
.second
;
2897 dump_erasure_code_profiles(erasure_code_profiles
, f
);
2900 void OSDMap::generate_test_instances(list
<OSDMap
*>& o
)
2902 o
.push_back(new OSDMap
);
2904 CephContext
*cct
= new CephContext(CODE_ENVIRONMENT_UTILITY
);
2905 o
.push_back(new OSDMap
);
2907 o
.back()->build_simple(cct
, 1, fsid
, 16);
2908 o
.back()->created
= o
.back()->modified
= utime_t(1, 2); // fix timestamp
2909 o
.back()->blacklist
[entity_addr_t()] = utime_t(5, 6);
2913 string
OSDMap::get_flag_string(unsigned f
)
2916 if ( f
& CEPH_OSDMAP_NEARFULL
)
2918 if (f
& CEPH_OSDMAP_FULL
)
2920 if (f
& CEPH_OSDMAP_PAUSERD
)
2922 if (f
& CEPH_OSDMAP_PAUSEWR
)
2924 if (f
& CEPH_OSDMAP_PAUSEREC
)
2926 if (f
& CEPH_OSDMAP_NOUP
)
2928 if (f
& CEPH_OSDMAP_NODOWN
)
2930 if (f
& CEPH_OSDMAP_NOOUT
)
2932 if (f
& CEPH_OSDMAP_NOIN
)
2934 if (f
& CEPH_OSDMAP_NOBACKFILL
)
2936 if (f
& CEPH_OSDMAP_NOREBALANCE
)
2937 s
+= ",norebalance";
2938 if (f
& CEPH_OSDMAP_NORECOVER
)
2940 if (f
& CEPH_OSDMAP_NOSCRUB
)
2942 if (f
& CEPH_OSDMAP_NODEEP_SCRUB
)
2943 s
+= ",nodeep-scrub";
2944 if (f
& CEPH_OSDMAP_NOTIERAGENT
)
2945 s
+= ",notieragent";
2946 if (f
& CEPH_OSDMAP_SORTBITWISE
)
2947 s
+= ",sortbitwise";
2948 if (f
& CEPH_OSDMAP_REQUIRE_JEWEL
)
2949 s
+= ",require_jewel_osds";
2950 if (f
& CEPH_OSDMAP_REQUIRE_KRAKEN
)
2951 s
+= ",require_kraken_osds";
2952 if (f
& CEPH_OSDMAP_REQUIRE_LUMINOUS
)
2953 s
+= ",require_luminous_osds";
2959 string
OSDMap::get_flag_string() const
2961 return get_flag_string(flags
);
2968 qi() : item(0), depth(0), weight(0) {}
2969 qi(int i
, int d
, float w
) : item(i
), depth(d
), weight(w
) {}
2972 void OSDMap::print_pools(ostream
& out
) const
2974 for (const auto &pool
: pools
) {
2975 std::string
name("<unknown>");
2976 const auto &pni
= pool_name
.find(pool
.first
);
2977 if (pni
!= pool_name
.end())
2979 out
<< "pool " << pool
.first
2981 << "' " << pool
.second
<< "\n";
2983 for (const auto &snap
: pool
.second
.snaps
)
2984 out
<< "\tsnap " << snap
.second
.snapid
<< " '" << snap
.second
.name
<< "' " << snap
.second
.stamp
<< "\n";
2986 if (!pool
.second
.removed_snaps
.empty())
2987 out
<< "\tremoved_snaps " << pool
.second
.removed_snaps
<< "\n";
2992 void OSDMap::print(ostream
& out
) const
2994 out
<< "epoch " << get_epoch() << "\n"
2995 << "fsid " << get_fsid() << "\n"
2996 << "created " << get_created() << "\n"
2997 << "modified " << get_modified() << "\n";
2999 out
<< "flags " << get_flag_string() << "\n";
3000 out
<< "crush_version " << get_crush_version() << "\n";
3001 out
<< "full_ratio " << full_ratio
<< "\n";
3002 out
<< "backfillfull_ratio " << backfillfull_ratio
<< "\n";
3003 out
<< "nearfull_ratio " << nearfull_ratio
<< "\n";
3004 if (require_min_compat_client
> 0) {
3005 out
<< "require_min_compat_client "
3006 << ceph_release_name(require_min_compat_client
) << "\n";
3008 out
<< "min_compat_client " << ceph_release_name(get_min_compat_client())
3010 if (require_osd_release
> 0) {
3011 out
<< "require_osd_release " << ceph_release_name(require_osd_release
)
3014 if (get_cluster_snapshot().length())
3015 out
<< "cluster_snapshot " << get_cluster_snapshot() << "\n";
3020 out
<< "max_osd " << get_max_osd() << "\n";
3021 for (int i
=0; i
<get_max_osd(); i
++) {
3024 out
<< (is_up(i
) ? " up ":" down");
3025 out
<< (is_in(i
) ? " in ":" out");
3026 out
<< " weight " << get_weightf(i
);
3027 if (get_primary_affinity(i
) != CEPH_OSD_DEFAULT_PRIMARY_AFFINITY
)
3028 out
<< " primary_affinity " << get_primary_affinityf(i
);
3029 const osd_info_t
& info(get_info(i
));
3031 out
<< " " << get_addr(i
) << " " << get_cluster_addr(i
) << " " << get_hb_back_addr(i
)
3032 << " " << get_hb_front_addr(i
);
3036 if (!get_uuid(i
).is_zero())
3037 out
<< " " << get_uuid(i
);
3043 for (auto& p
: pg_upmap
) {
3044 out
<< "pg_upmap " << p
.first
<< " " << p
.second
<< "\n";
3046 for (auto& p
: pg_upmap_items
) {
3047 out
<< "pg_upmap_items " << p
.first
<< " " << p
.second
<< "\n";
3050 for (const auto pg
: *pg_temp
)
3051 out
<< "pg_temp " << pg
.first
<< " " << pg
.second
<< "\n";
3053 for (const auto pg
: *primary_temp
)
3054 out
<< "primary_temp " << pg
.first
<< " " << pg
.second
<< "\n";
3056 for (const auto &addr
: blacklist
)
3057 out
<< "blacklist " << addr
.first
<< " expires " << addr
.second
<< "\n";
3059 // ignore pg_swap_primary
3062 class OSDTreePlainDumper
: public CrushTreeDumper::Dumper
<TextTable
> {
3064 typedef CrushTreeDumper::Dumper
<TextTable
> Parent
;
3066 OSDTreePlainDumper(const CrushWrapper
*crush
, const OSDMap
*osdmap_
,
3068 : Parent(crush
), osdmap(osdmap_
), filter(f
) { }
3070 bool should_dump_leaf(int i
) const override
{
3071 if (((filter
& OSDMap::DUMP_UP
) && !osdmap
->is_up(i
)) ||
3072 ((filter
& OSDMap::DUMP_DOWN
) && !osdmap
->is_down(i
)) ||
3073 ((filter
& OSDMap::DUMP_IN
) && !osdmap
->is_in(i
)) ||
3074 ((filter
& OSDMap::DUMP_OUT
) && !osdmap
->is_out(i
))) {
3080 bool should_dump_empty_bucket() const override
{
3084 void dump(TextTable
*tbl
) {
3085 tbl
->define_column("ID", TextTable::LEFT
, TextTable::RIGHT
);
3086 tbl
->define_column("CLASS", TextTable::LEFT
, TextTable::RIGHT
);
3087 tbl
->define_column("WEIGHT", TextTable::LEFT
, TextTable::RIGHT
);
3088 tbl
->define_column("TYPE NAME", TextTable::LEFT
, TextTable::LEFT
);
3089 tbl
->define_column("UP/DOWN", TextTable::LEFT
, TextTable::RIGHT
);
3090 tbl
->define_column("REWEIGHT", TextTable::LEFT
, TextTable::RIGHT
);
3091 tbl
->define_column("PRI-AFF", TextTable::LEFT
, TextTable::RIGHT
);
3095 for (int i
= 0; i
< osdmap
->get_max_osd(); i
++) {
3096 if (osdmap
->exists(i
) && !is_touched(i
) && should_dump_leaf(i
)) {
3097 dump_item(CrushTreeDumper::Item(i
, 0, 0), tbl
);
3103 void dump_item(const CrushTreeDumper::Item
&qi
, TextTable
*tbl
) override
{
3104 const char *c
= crush
->get_item_class(qi
.id
);
3109 << weightf_t(qi
.weight
);
3112 for (int k
= 0; k
< qi
.depth
; k
++)
3114 if (qi
.is_bucket()) {
3115 name
<< crush
->get_type_name(crush
->get_bucket_type(qi
.id
)) << " "
3116 << crush
->get_item_name(qi
.id
);
3118 name
<< "osd." << qi
.id
;
3122 if (!qi
.is_bucket()) {
3123 if (!osdmap
->exists(qi
.id
)) {
3127 *tbl
<< (osdmap
->is_up(qi
.id
) ? "up" : "down")
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(osdmap_
), filter(f
) { }
3148 bool should_dump_leaf(int i
) const override
{
3149 if (((filter
& OSDMap::DUMP_UP
) && !osdmap
->is_up(i
)) ||
3150 ((filter
& OSDMap::DUMP_DOWN
) && !osdmap
->is_down(i
)) ||
3151 ((filter
& OSDMap::DUMP_IN
) && !osdmap
->is_in(i
)) ||
3152 ((filter
& OSDMap::DUMP_OUT
) && !osdmap
->is_out(i
))) {
3158 bool should_dump_empty_bucket() const override
{
3162 void dump(Formatter
*f
) {
3163 f
->open_array_section("nodes");
3166 f
->open_array_section("stray");
3167 for (int i
= 0; i
< osdmap
->get_max_osd(); i
++) {
3168 if (osdmap
->exists(i
) && !is_touched(i
) && should_dump_leaf(i
))
3169 dump_item(CrushTreeDumper::Item(i
, 0, 0), f
);
3175 void dump_item_fields(const CrushTreeDumper::Item
&qi
, Formatter
*f
) override
{
3176 Parent::dump_item_fields(qi
, f
);
3177 if (!qi
.is_bucket())
3179 f
->dump_unsigned("exists", (int)osdmap
->exists(qi
.id
));
3180 f
->dump_string("status", osdmap
->is_up(qi
.id
) ? "up" : "down");
3181 f
->dump_float("reweight", osdmap
->get_weightf(qi
.id
));
3182 f
->dump_float("primary_affinity", osdmap
->get_primary_affinityf(qi
.id
));
3187 const OSDMap
*osdmap
;
3188 const unsigned filter
;
3191 void OSDMap::print_tree(Formatter
*f
, ostream
*out
, unsigned filter
) const
3194 OSDTreeFormattingDumper(crush
.get(), this, filter
).dump(f
);
3198 OSDTreePlainDumper(crush
.get(), this, filter
).dump(&tbl
);
3203 void OSDMap::print_summary(Formatter
*f
, ostream
& out
,
3204 const string
& prefix
) const
3207 f
->open_object_section("osdmap");
3208 f
->dump_int("epoch", get_epoch());
3209 f
->dump_int("num_osds", get_num_osds());
3210 f
->dump_int("num_up_osds", get_num_up_osds());
3211 f
->dump_int("num_in_osds", get_num_in_osds());
3212 f
->dump_bool("full", test_flag(CEPH_OSDMAP_FULL
) ? true : false);
3213 f
->dump_bool("nearfull", test_flag(CEPH_OSDMAP_NEARFULL
) ? true : false);
3214 f
->dump_unsigned("num_remapped_pgs", get_num_pg_temp());
3217 out
<< get_num_osds() << " osds: "
3218 << get_num_up_osds() << " up, "
3219 << get_num_in_osds() << " in";
3220 if (get_num_pg_temp())
3221 out
<< "; " << get_num_pg_temp() << " remapped pgs";
3223 uint64_t important_flags
= flags
& ~CEPH_OSDMAP_SEMIHIDDEN_FLAGS
;
3224 if (important_flags
)
3225 out
<< prefix
<< "flags " << get_flag_string(important_flags
) << "\n";
3229 void OSDMap::print_oneline_summary(ostream
& out
) const
3231 out
<< "e" << get_epoch() << ": "
3232 << get_num_osds() << " total, "
3233 << get_num_up_osds() << " up, "
3234 << get_num_in_osds() << " in";
3235 if (test_flag(CEPH_OSDMAP_FULL
))
3237 else if (test_flag(CEPH_OSDMAP_NEARFULL
))
3241 bool OSDMap::crush_ruleset_in_use(int ruleset
) const
3243 for (const auto &pool
: pools
) {
3244 if (pool
.second
.crush_rule
== ruleset
)
3250 int OSDMap::build_simple_optioned(CephContext
*cct
, epoch_t e
, uuid_d
&fsid
,
3251 int nosd
, int pg_bits
, int pgp_bits
,
3254 ldout(cct
, 10) << "build_simple on " << nosd
3255 << " osds" << dendl
;
3258 created
= modified
= ceph_clock_now();
3265 const md_config_t
*conf
= cct
->_conf
;
3266 vector
<string
> sections
;
3267 conf
->get_all_sections(sections
);
3269 for (auto §ion
: sections
) {
3270 if (section
.find("osd.") != 0)
3273 const char *begin
= section
.c_str() + 4;
3274 char *end
= (char*)begin
;
3275 int o
= strtol(begin
, &end
, 10);
3279 if (o
> cct
->_conf
->mon_max_osd
) {
3280 lderr(cct
) << "[osd." << o
<< "] in config has id > mon_max_osd " << cct
->_conf
->mon_max_osd
<< dendl
;
3288 set_max_osd(maxosd
+ 1);
3295 r
= build_simple_crush_map(cct
, *crush
, nosd
, &ss
);
3297 r
= build_simple_crush_map_from_conf(cct
, *crush
, &ss
);
3300 int poolbase
= get_max_osd() ? get_max_osd() : 1;
3302 int const default_replicated_rule
= crush
->get_osd_pool_default_crush_replicated_ruleset(cct
);
3303 assert(default_replicated_rule
>= 0);
3306 // pgp_num <= pg_num
3307 if (pgp_bits
> pg_bits
)
3310 vector
<string
> pool_names
;
3311 pool_names
.push_back("rbd");
3312 for (auto &plname
: pool_names
) {
3313 int64_t pool
= ++pool_max
;
3314 pools
[pool
].type
= pg_pool_t::TYPE_REPLICATED
;
3315 pools
[pool
].flags
= cct
->_conf
->osd_pool_default_flags
;
3316 if (cct
->_conf
->osd_pool_default_flag_hashpspool
)
3317 pools
[pool
].set_flag(pg_pool_t::FLAG_HASHPSPOOL
);
3318 if (cct
->_conf
->osd_pool_default_flag_nodelete
)
3319 pools
[pool
].set_flag(pg_pool_t::FLAG_NODELETE
);
3320 if (cct
->_conf
->osd_pool_default_flag_nopgchange
)
3321 pools
[pool
].set_flag(pg_pool_t::FLAG_NOPGCHANGE
);
3322 if (cct
->_conf
->osd_pool_default_flag_nosizechange
)
3323 pools
[pool
].set_flag(pg_pool_t::FLAG_NOSIZECHANGE
);
3324 pools
[pool
].size
= cct
->_conf
->osd_pool_default_size
;
3325 pools
[pool
].min_size
= cct
->_conf
->get_osd_pool_default_min_size();
3326 pools
[pool
].crush_rule
= default_replicated_rule
;
3327 pools
[pool
].object_hash
= CEPH_STR_HASH_RJENKINS
;
3328 pools
[pool
].set_pg_num(poolbase
<< pg_bits
);
3329 pools
[pool
].set_pgp_num(poolbase
<< pgp_bits
);
3330 pools
[pool
].last_change
= epoch
;
3331 pool_name
[pool
] = plname
;
3332 name_pool
[plname
] = pool
;
3336 for (int i
=0; i
<get_max_osd(); i
++) {
3338 set_weight(i
, CEPH_OSD_OUT
);
3341 map
<string
,string
> profile_map
;
3342 r
= get_erasure_code_profile_default(cct
, profile_map
, &ss
);
3344 lderr(cct
) << ss
.str() << dendl
;
3347 set_erasure_code_profile("default", profile_map
);
3351 int OSDMap::get_erasure_code_profile_default(CephContext
*cct
,
3352 map
<string
,string
> &profile_map
,
3355 int r
= get_json_str_map(cct
->_conf
->osd_pool_default_erasure_code_profile
,
3361 int OSDMap::_build_crush_types(CrushWrapper
& crush
)
3363 crush
.set_type_name(0, "osd");
3364 crush
.set_type_name(1, "host");
3365 crush
.set_type_name(2, "chassis");
3366 crush
.set_type_name(3, "rack");
3367 crush
.set_type_name(4, "row");
3368 crush
.set_type_name(5, "pdu");
3369 crush
.set_type_name(6, "pod");
3370 crush
.set_type_name(7, "room");
3371 crush
.set_type_name(8, "datacenter");
3372 crush
.set_type_name(9, "region");
3373 crush
.set_type_name(10, "root");
3377 int OSDMap::build_simple_crush_map(CephContext
*cct
, CrushWrapper
& crush
,
3378 int nosd
, ostream
*ss
)
3383 int root_type
= _build_crush_types(crush
);
3385 int r
= crush
.add_bucket(0, 0, CRUSH_HASH_DEFAULT
,
3386 root_type
, 0, NULL
, NULL
, &rootid
);
3388 crush
.set_item_name(rootid
, "default");
3390 for (int o
=0; o
<nosd
; o
++) {
3391 map
<string
,string
> loc
;
3392 loc
["host"] = "localhost";
3393 loc
["rack"] = "localrack";
3394 loc
["root"] = "default";
3395 ldout(cct
, 10) << " adding osd." << o
<< " at " << loc
<< dendl
;
3397 snprintf(name
, sizeof(name
), "osd.%d", o
);
3398 crush
.insert_item(cct
, o
, 1.0, name
, loc
);
3401 build_simple_crush_rules(cct
, crush
, "default", ss
);
3408 int OSDMap::build_simple_crush_map_from_conf(CephContext
*cct
,
3409 CrushWrapper
& crush
,
3412 const md_config_t
*conf
= cct
->_conf
;
3417 int root_type
= _build_crush_types(crush
);
3419 int r
= crush
.add_bucket(0, 0,
3421 root_type
, 0, NULL
, NULL
, &rootid
);
3423 crush
.set_item_name(rootid
, "default");
3426 vector
<string
> sections
;
3427 conf
->get_all_sections(sections
);
3429 for (auto §ion
: sections
) {
3430 if (section
.find("osd.") != 0)
3433 const char *begin
= section
.c_str() + 4;
3434 char *end
= (char*)begin
;
3435 int o
= strtol(begin
, &end
, 10);
3439 string host
, rack
, row
, room
, dc
, pool
;
3440 vector
<string
> sectiontmp
;
3441 sectiontmp
.push_back("osd");
3442 sectiontmp
.push_back(section
);
3443 conf
->get_val_from_conf_file(sectiontmp
, "host", host
, false);
3444 conf
->get_val_from_conf_file(sectiontmp
, "rack", rack
, false);
3445 conf
->get_val_from_conf_file(sectiontmp
, "row", row
, false);
3446 conf
->get_val_from_conf_file(sectiontmp
, "room", room
, false);
3447 conf
->get_val_from_conf_file(sectiontmp
, "datacenter", dc
, false);
3448 conf
->get_val_from_conf_file(sectiontmp
, "root", pool
, false);
3450 if (host
.length() == 0)
3451 host
= "unknownhost";
3452 if (rack
.length() == 0)
3453 rack
= "unknownrack";
3455 map
<string
,string
> loc
;
3463 loc
["datacenter"] = dc
;
3464 loc
["root"] = "default";
3466 ldout(cct
, 5) << " adding osd." << o
<< " at " << loc
<< dendl
;
3467 crush
.insert_item(cct
, o
, 1.0, section
, loc
);
3470 build_simple_crush_rules(cct
, crush
, "default", ss
);
3478 int OSDMap::build_simple_crush_rules(
3480 CrushWrapper
& crush
,
3484 int crush_rule
= crush
.get_osd_pool_default_crush_replicated_ruleset(cct
);
3485 string failure_domain
=
3486 crush
.get_type_name(cct
->_conf
->osd_crush_chooseleaf_type
);
3489 r
= crush
.add_simple_rule_at(
3490 "replicated_rule", root
, failure_domain
, "",
3491 "firstn", pg_pool_t::TYPE_REPLICATED
,
3495 // do not add an erasure rule by default or else we will implicitly
3496 // require the crush_v2 feature of clients
3500 int OSDMap::summarize_mapping_stats(
3502 const set
<int64_t> *pools
,
3510 for (auto &p
: get_pools())
3514 unsigned total_pg
= 0;
3515 unsigned moved_pg
= 0;
3516 vector
<unsigned> base_by_osd(get_max_osd(), 0);
3517 vector
<unsigned> new_by_osd(get_max_osd(), 0);
3518 for (int64_t pool_id
: ls
) {
3519 const pg_pool_t
*pi
= get_pg_pool(pool_id
);
3520 vector
<int> up
, up2
;
3522 for (unsigned ps
= 0; ps
< pi
->get_pg_num(); ++ps
) {
3523 pg_t
pgid(ps
, pool_id
, -1);
3524 total_pg
+= pi
->get_size();
3525 pg_to_up_acting_osds(pgid
, &up
, &up_primary
, nullptr, nullptr);
3526 for (int osd
: up
) {
3527 if (osd
>= 0 && osd
< get_max_osd())
3531 newmap
->pg_to_up_acting_osds(pgid
, &up2
, &up_primary
, nullptr, nullptr);
3532 for (int osd
: up2
) {
3533 if (osd
>= 0 && osd
< get_max_osd())
3536 if (pi
->type
== pg_pool_t::TYPE_ERASURE
) {
3537 for (unsigned i
=0; i
<up
.size(); ++i
) {
3538 if (up
[i
] != up2
[i
]) {
3542 } else if (pi
->type
== pg_pool_t::TYPE_REPLICATED
) {
3543 for (int osd
: up
) {
3544 if (std::find(up2
.begin(), up2
.end(), osd
) == up2
.end()) {
3549 assert(0 == "unhandled pool type");
3555 unsigned num_up_in
= 0;
3556 for (int osd
= 0; osd
< get_max_osd(); ++osd
) {
3557 if (is_up(osd
) && is_in(osd
))
3564 float avg_pg
= (float)total_pg
/ (float)num_up_in
;
3565 float base_stddev
= 0, new_stddev
= 0;
3566 int min
= -1, max
= -1;
3567 unsigned min_base_pg
= 0, max_base_pg
= 0;
3568 unsigned min_new_pg
= 0, max_new_pg
= 0;
3569 for (int osd
= 0; osd
< get_max_osd(); ++osd
) {
3570 if (is_up(osd
) && is_in(osd
)) {
3571 float base_diff
= (float)base_by_osd
[osd
] - avg_pg
;
3572 base_stddev
+= base_diff
* base_diff
;
3573 float new_diff
= (float)new_by_osd
[osd
] - avg_pg
;
3574 new_stddev
+= new_diff
* new_diff
;
3575 if (min
< 0 || base_by_osd
[osd
] < min_base_pg
) {
3577 min_base_pg
= base_by_osd
[osd
];
3578 min_new_pg
= new_by_osd
[osd
];
3580 if (max
< 0 || base_by_osd
[osd
] > max_base_pg
) {
3582 max_base_pg
= base_by_osd
[osd
];
3583 max_new_pg
= new_by_osd
[osd
];
3587 base_stddev
= sqrt(base_stddev
/ num_up_in
);
3588 new_stddev
= sqrt(new_stddev
/ num_up_in
);
3590 float edev
= sqrt(avg_pg
* (1.0 - (1.0 / (double)num_up_in
)));
3594 f
->open_object_section("utilization");
3597 f
->dump_unsigned("moved_pgs", moved_pg
);
3598 f
->dump_unsigned("total_pgs", total_pg
);
3602 percent
= (float)moved_pg
* 100.0 / (float)total_pg
;
3603 ss
<< "moved " << moved_pg
<< " / " << total_pg
3604 << " (" << percent
<< "%)\n";
3608 f
->dump_float("avg_pgs", avg_pg
);
3609 f
->dump_float("std_dev", base_stddev
);
3610 f
->dump_float("expected_baseline_std_dev", edev
);
3612 f
->dump_float("new_std_dev", new_stddev
);
3614 ss
<< "avg " << avg_pg
<< "\n";
3615 ss
<< "stddev " << base_stddev
;
3617 ss
<< " -> " << new_stddev
;
3618 ss
<< " (expected baseline " << edev
<< ")\n";
3622 f
->dump_unsigned("min_osd", min
);
3623 f
->dump_unsigned("min_osd_pgs", min_base_pg
);
3625 f
->dump_unsigned("new_min_osd_pgs", min_new_pg
);
3627 ss
<< "min osd." << min
<< " with " << min_base_pg
;
3629 ss
<< " -> " << min_new_pg
;
3630 ss
<< " pgs (" << (float)min_base_pg
/ avg_pg
;
3632 ss
<< " -> " << (float)min_new_pg
/ avg_pg
;
3638 f
->dump_unsigned("max_osd", max
);
3639 f
->dump_unsigned("max_osd_pgs", max_base_pg
);
3641 f
->dump_unsigned("new_max_osd_pgs", max_new_pg
);
3643 ss
<< "max osd." << max
<< " with " << max_base_pg
;
3645 ss
<< " -> " << max_new_pg
;
3646 ss
<< " pgs (" << (float)max_base_pg
/ avg_pg
;
3648 ss
<< " -> " << (float)max_new_pg
/ avg_pg
;
3660 int OSDMap::clean_pg_upmaps(
3662 Incremental
*pending_inc
)
3664 ldout(cct
, 10) << __func__
<< dendl
;
3666 for (auto& p
: pg_upmap
) {
3669 pg_to_raw_osds(p
.first
, &raw
, &primary
);
3670 if (vectors_equal(raw
, p
.second
)) {
3671 ldout(cct
, 10) << " removing redundant pg_upmap " << p
.first
<< " "
3672 << p
.second
<< dendl
;
3673 pending_inc
->old_pg_upmap
.insert(p
.first
);
3677 for (auto& p
: pg_upmap_items
) {
3680 pg_to_raw_osds(p
.first
, &raw
, &primary
);
3681 mempool::osdmap::vector
<pair
<int,int>> newmap
;
3682 for (auto& q
: p
.second
) {
3683 if (std::find(raw
.begin(), raw
.end(), q
.first
) != raw
.end()) {
3684 newmap
.push_back(q
);
3687 if (newmap
.empty()) {
3688 ldout(cct
, 10) << " removing no-op pg_upmap_items " << p
.first
<< " "
3689 << p
.second
<< dendl
;
3690 pending_inc
->old_pg_upmap_items
.insert(p
.first
);
3692 } else if (newmap
!= p
.second
) {
3693 ldout(cct
, 10) << " simplifying partially no-op pg_upmap_items "
3694 << p
.first
<< " " << p
.second
<< " -> " << newmap
<< dendl
;
3695 pending_inc
->new_pg_upmap_items
[p
.first
] = newmap
;
3702 bool OSDMap::try_pg_upmap(
3704 pg_t pg
, ///< pg to potentially remap
3705 const set
<int>& overfull
, ///< osds we'd want to evacuate
3706 const vector
<int>& underfull
, ///< osds to move to, in order of preference
3708 vector
<int> *out
) ///< resulting alternative mapping
3710 const pg_pool_t
*pool
= get_pg_pool(pg
.pool());
3713 int rule
= crush
->find_rule(pool
->get_crush_rule(), pool
->get_type(),
3718 // get original mapping
3719 _pg_to_raw_osds(*pool
, pg
, orig
, NULL
);
3721 // make sure there is something there to remap
3723 for (auto osd
: *orig
) {
3724 if (overfull
.count(osd
)) {
3733 int r
= crush
->try_remap_rule(
3737 overfull
, underfull
,
3747 int OSDMap::calc_pg_upmaps(
3749 float max_deviation_ratio
,
3751 const set
<int64_t>& only_pools_orig
,
3752 OSDMap::Incremental
*pending_inc
)
3754 set
<int64_t> only_pools
;
3755 if (only_pools_orig
.empty()) {
3756 for (auto& i
: pools
) {
3757 only_pools
.insert(i
.first
);
3760 only_pools
= only_pools_orig
;
3763 tmp
.deepish_copy_from(*this);
3764 float start_deviation
= 0;
3765 float end_deviation
= 0;
3766 int num_changed
= 0;
3768 map
<int,set
<pg_t
>> pgs_by_osd
;
3770 float osd_weight_total
= 0;
3771 map
<int,float> osd_weight
;
3772 for (auto& i
: pools
) {
3773 if (!only_pools
.empty() && !only_pools
.count(i
.first
))
3775 for (unsigned ps
= 0; ps
< i
.second
.get_pg_num(); ++ps
) {
3776 pg_t
pg(ps
, i
.first
);
3778 tmp
.pg_to_up_acting_osds(pg
, &up
, nullptr, nullptr, nullptr);
3779 for (auto osd
: up
) {
3780 if (osd
!= CRUSH_ITEM_NONE
)
3781 pgs_by_osd
[osd
].insert(pg
);
3784 total_pgs
+= i
.second
.get_size() * i
.second
.get_pg_num();
3786 map
<int,float> pmap
;
3787 int ruleno
= tmp
.crush
->find_rule(i
.second
.get_crush_rule(),
3788 i
.second
.get_type(),
3789 i
.second
.get_size());
3790 tmp
.crush
->get_rule_weight_osd_map(ruleno
, &pmap
);
3791 ldout(cct
,30) << __func__
<< " pool " << i
.first
<< " ruleno " << ruleno
<< dendl
;
3792 for (auto p
: pmap
) {
3793 osd_weight
[p
.first
] += p
.second
;
3794 osd_weight_total
+= p
.second
;
3797 for (auto& i
: osd_weight
) {
3799 auto p
= pgs_by_osd
.find(i
.first
);
3800 if (p
!= pgs_by_osd
.end())
3801 pgs
= p
->second
.size();
3803 pgs_by_osd
.emplace(i
.first
, set
<pg_t
>());
3804 ldout(cct
, 20) << " osd." << i
.first
<< " weight " << i
.second
3805 << " pgs " << pgs
<< dendl
;
3808 if (osd_weight_total
== 0) {
3809 lderr(cct
) << __func__
<< " abort due to osd_weight_total == 0" << dendl
;
3812 float pgs_per_weight
= total_pgs
/ osd_weight_total
;
3813 ldout(cct
, 10) << " osd_weight_total " << osd_weight_total
<< dendl
;
3814 ldout(cct
, 10) << " pgs_per_weight " << pgs_per_weight
<< dendl
;
3817 float total_deviation
= 0;
3818 map
<int,float> osd_deviation
; // osd, deviation(pgs)
3819 multimap
<float,int> deviation_osd
; // deviation(pgs), osd
3821 for (auto& i
: pgs_by_osd
) {
3822 float target
= osd_weight
[i
.first
] * pgs_per_weight
;
3823 float deviation
= (float)i
.second
.size() - target
;
3824 ldout(cct
, 20) << " osd." << i
.first
3825 << "\tpgs " << i
.second
.size()
3826 << "\ttarget " << target
3827 << "\tdeviation " << deviation
3829 osd_deviation
[i
.first
] = deviation
;
3830 deviation_osd
.insert(make_pair(deviation
, i
.first
));
3831 if (deviation
>= 1.0)
3832 overfull
.insert(i
.first
);
3833 total_deviation
+= abs(deviation
);
3835 if (num_changed
== 0) {
3836 start_deviation
= total_deviation
;
3838 end_deviation
= total_deviation
;
3840 // build underfull, sorted from least-full to most-average
3841 vector
<int> underfull
;
3842 for (auto i
= deviation_osd
.begin();
3843 i
!= deviation_osd
.end();
3845 if (i
->first
>= -.999)
3847 underfull
.push_back(i
->second
);
3849 ldout(cct
, 10) << " total_deviation " << total_deviation
3850 << " overfull " << overfull
3851 << " underfull " << underfull
<< dendl
;
3852 if (overfull
.empty() || underfull
.empty())
3856 bool restart
= false;
3857 for (auto p
= deviation_osd
.rbegin(); p
!= deviation_osd
.rend(); ++p
) {
3858 int osd
= p
->second
;
3859 float deviation
= p
->first
;
3860 float target
= osd_weight
[osd
] * pgs_per_weight
;
3862 if (deviation
/target
< max_deviation_ratio
) {
3863 ldout(cct
, 10) << " osd." << osd
3864 << " target " << target
3865 << " deviation " << deviation
3866 << " -> ratio " << deviation
/target
3867 << " < max ratio " << max_deviation_ratio
<< dendl
;
3870 int num_to_move
= deviation
;
3871 ldout(cct
, 10) << " osd." << osd
<< " move " << num_to_move
<< dendl
;
3872 if (num_to_move
< 1)
3875 set
<pg_t
>& pgs
= pgs_by_osd
[osd
];
3877 // look for remaps we can un-remap
3878 for (auto pg
: pgs
) {
3879 auto p
= tmp
.pg_upmap_items
.find(pg
);
3880 if (p
!= tmp
.pg_upmap_items
.end()) {
3881 for (auto q
: p
->second
) {
3882 if (q
.second
== osd
) {
3883 ldout(cct
, 10) << " dropping pg_upmap_items " << pg
3884 << " " << p
->second
<< dendl
;
3885 tmp
.pg_upmap_items
.erase(p
);
3886 pending_inc
->old_pg_upmap_items
.insert(pg
);
3898 for (auto pg
: pgs
) {
3899 if (tmp
.pg_upmap
.count(pg
) ||
3900 tmp
.pg_upmap_items
.count(pg
)) {
3901 ldout(cct
, 20) << " already remapped " << pg
<< dendl
;
3904 ldout(cct
, 10) << " trying " << pg
<< dendl
;
3905 vector
<int> orig
, out
;
3906 if (!try_pg_upmap(cct
, pg
, overfull
, underfull
, &orig
, &out
)) {
3909 ldout(cct
, 10) << " " << pg
<< " " << orig
<< " -> " << out
<< dendl
;
3910 if (orig
.size() != out
.size()) {
3913 assert(orig
!= out
);
3914 auto& rmi
= tmp
.pg_upmap_items
[pg
];
3915 for (unsigned i
= 0; i
< out
.size(); ++i
) {
3916 if (orig
[i
] != out
[i
]) {
3917 rmi
.push_back(make_pair(orig
[i
], out
[i
]));
3920 pending_inc
->new_pg_upmap_items
[pg
] = rmi
;
3921 ldout(cct
, 10) << " " << pg
<< " pg_upmap_items " << rmi
<< dendl
;
3931 ldout(cct
, 10) << " failed to find any changes to make" << dendl
;
3935 ldout(cct
, 10) << " hit max iterations, stopping" << dendl
;
3939 ldout(cct
, 10) << " start deviation " << start_deviation
<< dendl
;
3940 ldout(cct
, 10) << " end deviation " << end_deviation
<< dendl
;
3944 int OSDMap::get_osds_by_bucket_name(const string
&name
, set
<int> *osds
) const
3946 return crush
->get_leaves(name
, osds
);
3949 template <typename F
>
3950 class OSDUtilizationDumper
: public CrushTreeDumper::Dumper
<F
> {
3952 typedef CrushTreeDumper::Dumper
<F
> Parent
;
3954 OSDUtilizationDumper(const CrushWrapper
*crush
, const OSDMap
*osdmap_
,
3955 const PGStatService
*pgs_
, bool tree_
) :
3960 average_util(average_utilization()),
3968 void dump_stray(F
*f
) {
3969 for (int i
= 0; i
< osdmap
->get_max_osd(); i
++) {
3970 if (osdmap
->exists(i
) && !this->is_touched(i
))
3971 dump_item(CrushTreeDumper::Item(i
, 0, 0), f
);
3975 void dump_item(const CrushTreeDumper::Item
&qi
, F
*f
) override
{
3976 if (!tree
&& qi
.is_bucket())
3979 float reweight
= qi
.is_bucket() ? -1 : osdmap
->get_weightf(qi
.id
);
3980 int64_t kb
= 0, kb_used
= 0, kb_avail
= 0;
3982 if (get_bucket_utilization(qi
.id
, &kb
, &kb_used
, &kb_avail
))
3984 util
= 100.0 * (double)kb_used
/ (double)kb
;
3988 var
= util
/ average_util
;
3990 size_t num_pgs
= qi
.is_bucket() ? 0 : pgs
->get_num_pg_by_osd(qi
.id
);
3992 dump_item(qi
, reweight
, kb
, kb_used
, kb_avail
, util
, var
, num_pgs
, f
);
3994 if (!qi
.is_bucket() && reweight
> 0) {
3995 if (min_var
< 0 || var
< min_var
)
3997 if (max_var
< 0 || var
> max_var
)
4000 double dev
= util
- average_util
;
4002 stddev
+= reweight
* dev
;
4007 virtual void dump_item(const CrushTreeDumper::Item
&qi
,
4014 const size_t num_pgs
,
4018 return sum
> 0 ? sqrt(stddev
/ sum
) : 0;
4021 double average_utilization() {
4022 int64_t kb
= 0, kb_used
= 0;
4023 for (int i
= 0; i
< osdmap
->get_max_osd(); i
++) {
4024 if (!osdmap
->exists(i
) || osdmap
->get_weight(i
) == 0)
4026 int64_t kb_i
, kb_used_i
, kb_avail_i
;
4027 if (get_osd_utilization(i
, &kb_i
, &kb_used_i
, &kb_avail_i
)) {
4029 kb_used
+= kb_used_i
;
4032 return kb
> 0 ? 100.0 * (double)kb_used
/ (double)kb
: 0;
4035 bool get_osd_utilization(int id
, int64_t* kb
, int64_t* kb_used
,
4036 int64_t* kb_avail
) const {
4037 const osd_stat_t
*p
= pgs
->get_osd_stat(id
);
4038 if (!p
) return false;
4040 *kb_used
= p
->kb_used
;
4041 *kb_avail
= p
->kb_avail
;
4045 bool get_bucket_utilization(int id
, int64_t* kb
, int64_t* kb_used
,
4046 int64_t* kb_avail
) const {
4048 if (osdmap
->is_out(id
)) {
4054 return get_osd_utilization(id
, kb
, kb_used
, kb_avail
);
4061 for (int k
= osdmap
->crush
->get_bucket_size(id
) - 1; k
>= 0; k
--) {
4062 int item
= osdmap
->crush
->get_bucket_item(id
, k
);
4063 int64_t kb_i
= 0, kb_used_i
= 0, kb_avail_i
= 0;
4064 if (!get_bucket_utilization(item
, &kb_i
, &kb_used_i
, &kb_avail_i
))
4067 *kb_used
+= kb_used_i
;
4068 *kb_avail
+= kb_avail_i
;
4074 const OSDMap
*osdmap
;
4075 const PGStatService
*pgs
;
4077 double average_util
;
4085 class OSDUtilizationPlainDumper
: public OSDUtilizationDumper
<TextTable
> {
4087 typedef OSDUtilizationDumper
<TextTable
> Parent
;
4089 OSDUtilizationPlainDumper(const CrushWrapper
*crush
, const OSDMap
*osdmap
,
4090 const PGStatService
*pgs
, bool tree
) :
4091 Parent(crush
, osdmap
, pgs
, tree
) {}
4093 void dump(TextTable
*tbl
) {
4094 tbl
->define_column("ID", TextTable::LEFT
, TextTable::RIGHT
);
4095 tbl
->define_column("CLASS", TextTable::LEFT
, TextTable::RIGHT
);
4096 tbl
->define_column("WEIGHT", TextTable::LEFT
, TextTable::RIGHT
);
4097 tbl
->define_column("REWEIGHT", TextTable::LEFT
, TextTable::RIGHT
);
4098 tbl
->define_column("SIZE", TextTable::LEFT
, TextTable::RIGHT
);
4099 tbl
->define_column("USE", TextTable::LEFT
, TextTable::RIGHT
);
4100 tbl
->define_column("AVAIL", TextTable::LEFT
, TextTable::RIGHT
);
4101 tbl
->define_column("%USE", TextTable::LEFT
, TextTable::RIGHT
);
4102 tbl
->define_column("VAR", TextTable::LEFT
, TextTable::RIGHT
);
4103 tbl
->define_column("PGS", TextTable::LEFT
, TextTable::RIGHT
);
4105 tbl
->define_column("TYPE NAME", TextTable::LEFT
, TextTable::LEFT
);
4114 << si_t(pgs
->get_osd_sum().kb
<< 10)
4115 << si_t(pgs
->get_osd_sum().kb_used
<< 10)
4116 << si_t(pgs
->get_osd_sum().kb_avail
<< 10)
4117 << lowprecision_t(average_util
)
4119 << TextTable::endrow
;
4123 struct lowprecision_t
{
4125 explicit lowprecision_t(float _v
) : v(_v
) {}
4127 friend std::ostream
&operator<<(ostream
& out
, const lowprecision_t
& v
);
4129 using OSDUtilizationDumper
<TextTable
>::dump_item
;
4130 void dump_item(const CrushTreeDumper::Item
&qi
,
4137 const size_t num_pgs
,
4138 TextTable
*tbl
) override
{
4139 const char *c
= crush
->get_item_class(qi
.id
);
4144 << weightf_t(qi
.weight
)
4145 << weightf_t(reweight
)
4147 << si_t(kb_used
<< 10)
4148 << si_t(kb_avail
<< 10)
4149 << lowprecision_t(util
)
4150 << lowprecision_t(var
);
4152 if (qi
.is_bucket()) {
4160 for (int k
= 0; k
< qi
.depth
; k
++)
4162 if (qi
.is_bucket()) {
4163 int type
= crush
->get_bucket_type(qi
.id
);
4164 name
<< crush
->get_type_name(type
) << " "
4165 << crush
->get_item_name(qi
.id
);
4167 name
<< "osd." << qi
.id
;
4172 *tbl
<< TextTable::endrow
;
4178 out
<< "MIN/MAX VAR: " << lowprecision_t(min_var
)
4179 << "/" << lowprecision_t(max_var
) << " "
4180 << "STDDEV: " << lowprecision_t(dev());
4185 ostream
& operator<<(ostream
& out
,
4186 const OSDUtilizationPlainDumper::lowprecision_t
& v
)
4190 } else if (v
.v
< 0.001) {
4193 std::streamsize p
= out
.precision();
4194 return out
<< std::fixed
<< std::setprecision(2) << v
.v
<< std::setprecision(p
);
4198 class OSDUtilizationFormatDumper
: public OSDUtilizationDumper
<Formatter
> {
4200 typedef OSDUtilizationDumper
<Formatter
> Parent
;
4202 OSDUtilizationFormatDumper(const CrushWrapper
*crush
, const OSDMap
*osdmap
,
4203 const PGStatService
*pgs
, bool tree
) :
4204 Parent(crush
, osdmap
, pgs
, tree
) {}
4206 void dump(Formatter
*f
) {
4207 f
->open_array_section("nodes");
4211 f
->open_array_section("stray");
4217 using OSDUtilizationDumper
<Formatter
>::dump_item
;
4218 void dump_item(const CrushTreeDumper::Item
&qi
,
4225 const size_t num_pgs
,
4226 Formatter
*f
) override
{
4227 f
->open_object_section("item");
4228 CrushTreeDumper::dump_item_fields(crush
, qi
, f
);
4229 f
->dump_float("reweight", reweight
);
4230 f
->dump_int("kb", kb
);
4231 f
->dump_int("kb_used", kb_used
);
4232 f
->dump_int("kb_avail", kb_avail
);
4233 f
->dump_float("utilization", util
);
4234 f
->dump_float("var", var
);
4235 f
->dump_unsigned("pgs", num_pgs
);
4236 CrushTreeDumper::dump_bucket_children(crush
, qi
, f
);
4241 void summary(Formatter
*f
) {
4242 f
->open_object_section("summary");
4243 f
->dump_int("total_kb", pgs
->get_osd_sum().kb
);
4244 f
->dump_int("total_kb_used", pgs
->get_osd_sum().kb_used
);
4245 f
->dump_int("total_kb_avail", pgs
->get_osd_sum().kb_avail
);
4246 f
->dump_float("average_utilization", average_util
);
4247 f
->dump_float("min_var", min_var
);
4248 f
->dump_float("max_var", max_var
);
4249 f
->dump_float("dev", dev());
4254 void print_osd_utilization(const OSDMap
& osdmap
,
4255 const PGStatService
*pgstat
,
4260 const CrushWrapper
*crush
= osdmap
.crush
.get();
4262 f
->open_object_section("df");
4263 OSDUtilizationFormatDumper
d(crush
, &osdmap
, pgstat
, tree
);
4269 OSDUtilizationPlainDumper
d(crush
, &osdmap
, pgstat
, tree
);
4272 out
<< tbl
<< d
.summary() << "\n";
4276 void OSDMap::check_health(health_check_map_t
*checks
) const
4278 int num_osds
= get_num_osds();
4281 // OSD_$subtree_DOWN
4283 if (num_osds
>= 0) {
4284 int num_in_osds
= 0;
4285 int num_down_in_osds
= 0;
4287 set
<int> down_in_osds
;
4288 set
<int> up_in_osds
;
4289 set
<int> subtree_up
;
4290 unordered_map
<int, set
<int> > subtree_type_down
;
4291 unordered_map
<int, int> num_osds_subtree
;
4292 int max_type
= crush
->get_max_type_id();
4294 for (int i
= 0; i
< get_max_osd(); i
++) {
4296 if (crush
->item_exists(i
)) {
4304 if (down_in_osds
.count(i
) || up_in_osds
.count(i
))
4307 down_in_osds
.insert(i
);
4310 for (int type
= 0; type
<= max_type
; type
++) {
4311 if (!crush
->get_type_name(type
))
4313 int r
= crush
->get_immediate_parent_id(current
, &parent_id
);
4316 // break early if this parent is already marked as up
4317 if (subtree_up
.count(parent_id
))
4319 type
= crush
->get_bucket_type(parent_id
);
4320 if (!subtree_type_is_down(
4321 g_ceph_context
, parent_id
, type
,
4322 &down_in_osds
, &up_in_osds
, &subtree_up
, &subtree_type_down
))
4324 current
= parent_id
;
4329 // calculate the number of down osds in each down subtree and
4330 // store it in num_osds_subtree
4331 for (int type
= 1; type
<= max_type
; type
++) {
4332 if (!crush
->get_type_name(type
))
4334 for (auto j
= subtree_type_down
[type
].begin();
4335 j
!= subtree_type_down
[type
].end();
4339 int num_children
= crush
->get_children(*j
, &children
);
4340 if (num_children
== 0)
4342 for (auto l
= children
.begin(); l
!= children
.end(); ++l
) {
4345 } else if (num_osds_subtree
[*l
] > 0) {
4346 num
= num
+ num_osds_subtree
[*l
];
4349 num_osds_subtree
[*j
] = num
;
4352 num_down_in_osds
= down_in_osds
.size();
4353 assert(num_down_in_osds
<= num_in_osds
);
4354 if (num_down_in_osds
> 0) {
4355 // summary of down subtree types and osds
4356 for (int type
= max_type
; type
> 0; type
--) {
4357 if (!crush
->get_type_name(type
))
4359 if (subtree_type_down
[type
].size() > 0) {
4361 ss
<< subtree_type_down
[type
].size() << " "
4362 << crush
->get_type_name(type
);
4363 if (subtree_type_down
[type
].size() > 1) {
4366 int sum_down_osds
= 0;
4367 for (auto j
= subtree_type_down
[type
].begin();
4368 j
!= subtree_type_down
[type
].end();
4370 sum_down_osds
= sum_down_osds
+ num_osds_subtree
[*j
];
4372 ss
<< " (" << sum_down_osds
<< " osds) down";
4373 string err
= string("OSD_") +
4374 string(crush
->get_type_name(type
)) + "_DOWN";
4375 boost::to_upper(err
);
4376 auto& d
= checks
->add(err
, HEALTH_WARN
, ss
.str());
4377 for (auto j
= subtree_type_down
[type
].rbegin();
4378 j
!= subtree_type_down
[type
].rend();
4381 ss
<< crush
->get_type_name(type
);
4383 ss
<< crush
->get_item_name(*j
);
4384 // at the top level, do not print location
4385 if (type
!= max_type
) {
4387 ss
<< crush
->get_full_location_ordered_string(*j
);
4390 int num
= num_osds_subtree
[*j
];
4391 ss
<< " (" << num
<< " osds)";
4393 d
.detail
.push_back(ss
.str());
4398 ss
<< down_in_osds
.size() << " osds down";
4399 auto& d
= checks
->add("OSD_DOWN", HEALTH_WARN
, ss
.str());
4400 for (auto it
= down_in_osds
.begin(); it
!= down_in_osds
.end(); ++it
) {
4402 ss
<< "osd." << *it
<< " (";
4403 ss
<< crush
->get_full_location_ordered_string(*it
);
4405 d
.detail
.push_back(ss
.str());
4409 if (!osds
.empty()) {
4411 ss
<< osds
.size() << " osds exist in the crush map but not in the osdmap";
4412 auto& d
= checks
->add("OSD_ORPHAN", HEALTH_WARN
, ss
.str());
4413 for (auto osd
: osds
) {
4415 ss
<< "osd." << osd
<< " exists in crush map but not in osdmap";
4416 d
.detail
.push_back(ss
.str());
4421 // OSD_OUT_OF_ORDER_FULL
4423 // An osd could configure failsafe ratio, to something different
4424 // but for now assume it is the same here.
4425 float fsr
= g_conf
->osd_failsafe_full_ratio
;
4426 if (fsr
> 1.0) fsr
/= 100;
4427 float fr
= get_full_ratio();
4428 float br
= get_backfillfull_ratio();
4429 float nr
= get_nearfull_ratio();
4431 list
<string
> detail
;
4432 // These checks correspond to how OSDService::check_full_status() in an OSD
4433 // handles the improper setting of these values.
4436 ss
<< "backfillfull_ratio (" << br
4437 << ") < nearfull_ratio (" << nr
<< "), increased";
4438 detail
.push_back(ss
.str());
4443 ss
<< "full_ratio (" << fr
<< ") < backfillfull_ratio (" << br
4445 detail
.push_back(ss
.str());
4450 ss
<< "osd_failsafe_full_ratio (" << fsr
<< ") < full_ratio (" << fr
4452 detail
.push_back(ss
.str());
4454 if (!detail
.empty()) {
4455 auto& d
= checks
->add("OSD_OUT_OF_ORDER_FULL", HEALTH_ERR
,
4456 "full ratio(s) out of order");
4457 d
.detail
.swap(detail
);
4464 // OSD_FAILSAFE_FULL
4466 set
<int> full
, backfillfull
, nearfull
;
4467 get_full_osd_counts(&full
, &backfillfull
, &nearfull
);
4470 ss
<< full
.size() << " full osd(s)";
4471 auto& d
= checks
->add("OSD_FULL", HEALTH_ERR
, ss
.str());
4472 for (auto& i
: full
) {
4474 ss
<< "osd." << i
<< " is full";
4475 d
.detail
.push_back(ss
.str());
4478 if (backfillfull
.size()) {
4480 ss
<< backfillfull
.size() << " backfillfull osd(s)";
4481 auto& d
= checks
->add("OSD_BACKFILLFULL", HEALTH_WARN
, ss
.str());
4482 for (auto& i
: backfillfull
) {
4484 ss
<< "osd." << i
<< " is backfill full";
4485 d
.detail
.push_back(ss
.str());
4488 if (nearfull
.size()) {
4490 ss
<< nearfull
.size() << " nearfull osd(s)";
4491 auto& d
= checks
->add("OSD_NEARFULL", HEALTH_WARN
, ss
.str());
4492 for (auto& i
: nearfull
) {
4494 ss
<< "osd." << i
<< " is near full";
4495 d
.detail
.push_back(ss
.str());
4503 uint64_t warn_flags
=
4505 CEPH_OSDMAP_PAUSERD
|
4506 CEPH_OSDMAP_PAUSEWR
|
4507 CEPH_OSDMAP_PAUSEREC
|
4509 CEPH_OSDMAP_NODOWN
|
4512 CEPH_OSDMAP_NOBACKFILL
|
4513 CEPH_OSDMAP_NORECOVER
|
4514 CEPH_OSDMAP_NOSCRUB
|
4515 CEPH_OSDMAP_NODEEP_SCRUB
|
4516 CEPH_OSDMAP_NOTIERAGENT
|
4517 CEPH_OSDMAP_NOREBALANCE
;
4518 if (test_flag(warn_flags
)) {
4520 ss
<< get_flag_string(get_flags() & warn_flags
)
4522 checks
->add("OSDMAP_FLAGS", HEALTH_WARN
, ss
.str());
4528 list
<string
> detail
;
4529 const unsigned flags
=
4534 for (int i
= 0; i
< max_osd
; ++i
) {
4535 if (osd_state
[i
] & flags
) {
4538 OSDMap::calc_state_set(osd_state
[i
] & flags
, states
);
4539 ss
<< "osd." << i
<< " has flags " << states
;
4540 detail
.push_back(ss
.str());
4543 if (!detail
.empty()) {
4545 ss
<< detail
.size() << " osd(s) have {NOUP,NODOWN,NOIN,NOOUT} flags set";
4546 auto& d
= checks
->add("OSD_FLAGS", HEALTH_WARN
, ss
.str());
4547 d
.detail
.swap(detail
);
4551 // OLD_CRUSH_TUNABLES
4552 if (g_conf
->mon_warn_on_legacy_crush_tunables
) {
4553 string min
= crush
->get_min_required_version();
4554 if (min
< g_conf
->mon_crush_min_required_version
) {
4556 ss
<< "crush map has legacy tunables (require " << min
4557 << ", min is " << g_conf
->mon_crush_min_required_version
<< ")";
4558 auto& d
= checks
->add("OLD_CRUSH_TUNABLES", HEALTH_WARN
, ss
.str());
4559 d
.detail
.push_back("see http://docs.ceph.com/docs/master/rados/operations/crush-map/#tunables");
4563 // OLD_CRUSH_STRAW_CALC_VERSION
4564 if (g_conf
->mon_warn_on_crush_straw_calc_version_zero
) {
4565 if (crush
->get_straw_calc_version() == 0) {
4567 ss
<< "crush map has straw_calc_version=0";
4568 auto& d
= checks
->add("OLD_CRUSH_STRAW_CALC_VERSION", HEALTH_WARN
, ss
.str());
4570 "see http://docs.ceph.com/docs/master/rados/operations/crush-map/#tunables");
4574 // CACHE_POOL_NO_HIT_SET
4575 if (g_conf
->mon_warn_on_cache_pools_without_hit_sets
) {
4576 list
<string
> detail
;
4577 for (map
<int64_t, pg_pool_t
>::const_iterator p
= pools
.begin();
4580 const pg_pool_t
& info
= p
->second
;
4581 if (info
.cache_mode_requires_hit_set() &&
4582 info
.hit_set_params
.get_type() == HitSet::TYPE_NONE
) {
4584 ss
<< "pool '" << get_pool_name(p
->first
)
4585 << "' with cache_mode " << info
.get_cache_mode_name()
4586 << " needs hit_set_type to be set but it is not";
4587 detail
.push_back(ss
.str());
4590 if (!detail
.empty()) {
4592 ss
<< detail
.size() << " cache pools are missing hit_sets";
4593 auto& d
= checks
->add("CACHE_POOL_NO_HIT_SET", HEALTH_WARN
, ss
.str());
4594 d
.detail
.swap(detail
);
4598 // OSD_NO_SORTBITWISE
4599 if (!test_flag(CEPH_OSDMAP_SORTBITWISE
) &&
4600 (get_up_osd_features() &
4601 CEPH_FEATURE_OSD_BITWISE_HOBJ_SORT
)) {
4603 ss
<< "no legacy OSD present but 'sortbitwise' flag is not set";
4604 checks
->add("OSD_NO_SORTBITWISE", HEALTH_WARN
, ss
.str());
4607 // OSD_UPGRADE_FINISHED
4608 // none of these (yet) since we don't run until luminous upgrade is done.
4612 list
<string
> detail
;
4613 for (auto it
: get_pools()) {
4614 const pg_pool_t
&pool
= it
.second
;
4615 if (pool
.has_flag(pg_pool_t::FLAG_FULL
)) {
4616 const string
& pool_name
= get_pool_name(it
.first
);
4618 ss
<< "pool '" << pool_name
<< "' is full";
4619 detail
.push_back(ss
.str());
4622 if (!detail
.empty()) {
4624 ss
<< detail
.size() << " pool(s) full";
4625 auto& d
= checks
->add("POOL_FULL", HEALTH_WARN
, ss
.str());
4626 d
.detail
.swap(detail
);