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.
23 * describe properties of the OSD cluster.
24 * disks, disk groups, total # osds,
33 #include <boost/smart_ptr/local_shared_ptr.hpp>
34 #include "include/btree_map.h"
35 #include "include/common_fwd.h"
36 #include "include/types.h"
37 #include "common/ceph_releases.h"
38 #include "osd_types.h"
40 //#include "include/ceph_features.h"
41 #include "crush/CrushWrapper.h"
43 // forward declaration
45 class health_check_map_t
;
48 * we track up to two intervals during which the osd was alive and
49 * healthy. the most recent is [up_from,up_thru), where up_thru is
50 * the last epoch the osd is known to have _started_. i.e., a lower
51 * bound on the actual osd death. down_at (if it is > up_from) is an
52 * upper bound on the actual osd death.
54 * the second is the last_clean interval [first,last]. in that case,
55 * the last interval is the last epoch known to have been either
56 * _finished_, or during which the osd cleanly shut down. when
57 * possible, we push this forward to the epoch the osd was eventually
60 * the lost_at is used to allow build_prior to proceed without waiting
61 * for an osd to recover. In certain cases, progress may be blocked
62 * because an osd is down that may contain updates (i.e., a pg may have
63 * gone rw during an interval). If the osd can't be brought online, we
64 * can force things to proceed knowing that we _might_ be losing some
65 * acked writes. If the osd comes back to life later, that's fine to,
66 * but those writes will still be lost (the divergent objects will be
70 epoch_t last_clean_begin
; // last interval that ended with a clean osd shutdown
71 epoch_t last_clean_end
;
72 epoch_t up_from
; // epoch osd marked up
73 epoch_t up_thru
; // lower bound on actual osd death (if > up_from)
74 epoch_t down_at
; // upper bound on actual osd death (if > up_from)
75 epoch_t lost_at
; // last epoch we decided data was "lost"
77 osd_info_t() : last_clean_begin(0), last_clean_end(0),
78 up_from(0), up_thru(0), down_at(0), lost_at(0) {}
80 void dump(ceph::Formatter
*f
) const;
81 void encode(ceph::buffer::list
& bl
) const;
82 void decode(ceph::buffer::list::const_iterator
& bl
);
83 static void generate_test_instances(std::list
<osd_info_t
*>& o
);
85 WRITE_CLASS_ENCODER(osd_info_t
)
87 std::ostream
& operator<<(std::ostream
& out
, const osd_info_t
& info
);
90 utime_t down_stamp
; ///< timestamp when we were last marked down
91 float laggy_probability
; ///< encoded as __u32: 0 = definitely not laggy, 0xffffffff definitely laggy
92 __u32 laggy_interval
; ///< average interval between being marked laggy and recovering
93 uint64_t features
; ///< features supported by this osd we should know about
94 __u32 old_weight
; ///< weight prior to being auto marked out
95 utime_t last_purged_snaps_scrub
; ///< last scrub of purged_snaps
96 epoch_t dead_epoch
= 0; ///< last epoch we were confirmed dead (not just down)
98 osd_xinfo_t() : laggy_probability(0), laggy_interval(0),
99 features(0), old_weight(0) {}
101 void dump(ceph::Formatter
*f
) const;
102 void encode(ceph::buffer::list
& bl
, uint64_t features
) const;
103 void decode(ceph::buffer::list::const_iterator
& bl
);
104 static void generate_test_instances(std::list
<osd_xinfo_t
*>& o
);
106 WRITE_CLASS_ENCODER_FEATURES(osd_xinfo_t
)
108 std::ostream
& operator<<(std::ostream
& out
, const osd_xinfo_t
& xi
);
113 ceph::buffer::list data
;
114 typedef btree::btree_map
<pg_t
,ceph_le32
*> map_t
;
117 void encode(ceph::buffer::list
& bl
) const {
119 uint32_t n
= map
.size();
121 for (auto &p
: map
) {
123 bl
.append((char*)p
.second
, (*p
.second
+ 1) * sizeof(ceph_le32
));
126 void decode(ceph::buffer::list::const_iterator
& p
) {
135 size_t start_off
= pstart
.get_off();
136 std::vector
<std::pair
<pg_t
,size_t>> offsets
;
138 for (unsigned i
=0; i
<n
; ++i
) {
141 offsets
[i
].first
= pgid
;
142 offsets
[i
].second
= p
.get_off() - start_off
;
145 p
+= vn
* sizeof(int32_t);
147 size_t len
= p
.get_off() - start_off
;
148 pstart
.copy(len
, data
);
149 if (data
.get_num_buffers() > 1) {
153 char *start
= data
.c_str();
154 for (auto i
: offsets
) {
155 map
.insert(map
.end(), std::make_pair(i
.first
, (ceph_le32
*)(start
+ i
.second
)));
159 ceph::buffer::list bl
;
161 auto p
= std::cbegin(bl
);
164 friend bool operator==(const PGTempMap
& l
, const PGTempMap
& r
) {
166 l
.map
.size() == r
.map
.size() &&
167 l
.data
.contents_equal(r
.data
);
171 map_t::const_iterator it
;
172 map_t::const_iterator end
;
173 std::pair
<pg_t
,std::vector
<int32_t>> current
;
174 void init_current() {
176 current
.first
= it
->first
;
177 ceph_assert(it
->second
);
178 current
.second
.resize(*it
->second
);
179 ceph_le32
*p
= it
->second
+ 1;
180 for (uint32_t n
= 0; n
< *it
->second
; ++n
, ++p
) {
181 current
.second
[n
] = *p
;
186 iterator(map_t::const_iterator p
,
187 map_t::const_iterator e
)
192 const std::pair
<pg_t
,std::vector
<int32_t>>& operator*() const {
195 const std::pair
<pg_t
,std::vector
<int32_t>>* operator->() const {
198 friend bool operator==(const iterator
& l
, const iterator
& r
) {
201 friend bool operator!=(const iterator
& l
, const iterator
& r
) {
204 iterator
& operator++() {
210 iterator
operator++(int) {
218 iterator
begin() const {
219 return iterator(map
.begin(), map
.end());
221 iterator
end() const {
222 return iterator(map
.end(), map
.end());
224 iterator
find(pg_t pgid
) const {
225 return iterator(map
.find(pgid
), map
.end());
227 size_t size() const {
230 size_t count(pg_t pgid
) const {
231 return map
.count(pgid
);
233 void erase(pg_t pgid
) {
240 void set(pg_t pgid
, const mempool::osdmap::vector
<int32_t>& v
) {
242 size_t need
= sizeof(ceph_le32
) * (1 + v
.size());
243 if (need
< data
.get_append_buffer_unused_tail_length()) {
244 ceph::buffer::ptr
z(data
.get_append_buffer_unused_tail_length());
246 data
.append(z
.c_str(), z
.length());
249 map
[pgid
] = (ceph_le32
*)(data
.back().end_c_str()) - (1 + v
.size());
251 mempool::osdmap::vector
<int32_t> get(pg_t pgid
) {
252 mempool::osdmap::vector
<int32_t> v
;
253 ceph_le32
*p
= map
[pgid
];
256 for (size_t i
= 0; i
< n
; ++i
, ++p
) {
262 // trivial implementation
263 mempool::osdmap::map
<pg_t
,mempool::osdmap::vector
<int32_t> > pg_temp
;
265 void encode(ceph::buffer::list
& bl
) const {
268 void decode(ceph::buffer::list::const_iterator
& p
) {
271 friend bool operator==(const PGTempMap
& l
, const PGTempMap
& r
) {
273 l
.pg_temp
.size() == r
.pg_temp
.size() &&
274 l
.pg_temp
== r
.pg_temp
;
278 mempool::osdmap::map
<pg_t
,mempool::osdmap::vector
<int32_t> >::const_iterator it
;
280 iterator(mempool::osdmap::map
<pg_t
,
281 mempool::osdmap::vector
<int32_t> >::const_iterator p
)
284 std::pair
<pg_t
,const mempool::osdmap::vector
<int32_t>&> operator*() const {
287 const std::pair
<const pg_t
,mempool::osdmap::vector
<int32_t>>* operator->() const {
290 friend bool operator==(const iterator
& l
, const iterator
& r
) {
293 friend bool operator!=(const iterator
& l
, const iterator
& r
) {
296 iterator
& operator++() {
300 iterator
operator++(int) {
306 iterator
begin() const {
307 return iterator(pg_temp
.cbegin());
309 iterator
end() const {
310 return iterator(pg_temp
.cend());
312 iterator
find(pg_t pgid
) const {
313 return iterator(pg_temp
.find(pgid
));
315 size_t size() const {
316 return pg_temp
.size();
318 size_t count(pg_t pgid
) const {
319 return pg_temp
.count(pgid
);
321 void erase(pg_t pgid
) {
327 void set(pg_t pgid
, const mempool::osdmap::vector
<int32_t>& v
) {
330 const mempool::osdmap::vector
<int32_t>& get(pg_t pgid
) {
331 return pg_temp
.at(pgid
);
334 void dump(ceph::Formatter
*f
) const {
335 for (const auto &pg
: *this) {
336 f
->open_object_section("osds");
337 f
->dump_stream("pgid") << pg
.first
;
338 f
->open_array_section("osds");
339 for (const auto osd
: pg
.second
)
340 f
->dump_int("osd", osd
);
346 WRITE_CLASS_ENCODER(PGTempMap
)
352 MEMPOOL_CLASS_HELPERS();
356 MEMPOOL_CLASS_HELPERS();
358 /// feature bits we were encoded with. the subsequent OSDMap
359 /// encoding should match.
360 uint64_t encode_features
;
362 epoch_t epoch
; // new epoch; we are a diff from epoch-1 to epoch
364 int64_t new_pool_max
; //incremented by the OSDMonitor on each pool create
366 ceph_release_t new_require_osd_release
{0xff};
369 ceph::buffer::list fullmap
; // in lieu of below.
370 ceph::buffer::list crush
;
374 mempool::osdmap::map
<int64_t,pg_pool_t
> new_pools
;
375 mempool::osdmap::map
<int64_t,std::string
> new_pool_names
;
376 mempool::osdmap::set
<int64_t> old_pools
;
377 mempool::osdmap::map
<std::string
,std::map
<std::string
,std::string
> > new_erasure_code_profiles
;
378 mempool::osdmap::vector
<std::string
> old_erasure_code_profiles
;
379 mempool::osdmap::map
<int32_t,entity_addrvec_t
> new_up_client
;
380 mempool::osdmap::map
<int32_t,entity_addrvec_t
> new_up_cluster
;
381 mempool::osdmap::map
<int32_t,uint32_t> new_state
; // XORed onto previous state.
382 mempool::osdmap::map
<int32_t,uint32_t> new_weight
;
383 mempool::osdmap::map
<pg_t
,mempool::osdmap::vector
<int32_t> > new_pg_temp
; // [] to remove
384 mempool::osdmap::map
<pg_t
, int32_t> new_primary_temp
; // [-1] to remove
385 mempool::osdmap::map
<int32_t,uint32_t> new_primary_affinity
;
386 mempool::osdmap::map
<int32_t,epoch_t
> new_up_thru
;
387 mempool::osdmap::map
<int32_t,std::pair
<epoch_t
,epoch_t
> > new_last_clean_interval
;
388 mempool::osdmap::map
<int32_t,epoch_t
> new_lost
;
389 mempool::osdmap::map
<int32_t,uuid_d
> new_uuid
;
390 mempool::osdmap::map
<int32_t,osd_xinfo_t
> new_xinfo
;
392 mempool::osdmap::map
<entity_addr_t
,utime_t
> new_blacklist
;
393 mempool::osdmap::vector
<entity_addr_t
> old_blacklist
;
394 mempool::osdmap::map
<int32_t, entity_addrvec_t
> new_hb_back_up
;
395 mempool::osdmap::map
<int32_t, entity_addrvec_t
> new_hb_front_up
;
397 mempool::osdmap::map
<pg_t
,mempool::osdmap::vector
<int32_t>> new_pg_upmap
;
398 mempool::osdmap::map
<pg_t
,mempool::osdmap::vector
<std::pair
<int32_t,int32_t>>> new_pg_upmap_items
;
399 mempool::osdmap::set
<pg_t
> old_pg_upmap
, old_pg_upmap_items
;
400 mempool::osdmap::map
<int64_t, snap_interval_set_t
> new_removed_snaps
;
401 mempool::osdmap::map
<int64_t, snap_interval_set_t
> new_purged_snaps
;
403 mempool::osdmap::map
<int32_t,uint32_t> new_crush_node_flags
;
404 mempool::osdmap::map
<int32_t,uint32_t> new_device_class_flags
;
406 std::string cluster_snapshot
;
408 float new_nearfull_ratio
= -1;
409 float new_backfillfull_ratio
= -1;
410 float new_full_ratio
= -1;
412 ceph_release_t new_require_min_compat_client
{0xff};
414 utime_t new_last_up_change
, new_last_in_change
;
416 mutable bool have_crc
; ///< crc values are defined
417 uint32_t full_crc
; ///< crc of the resulting OSDMap
418 mutable uint32_t inc_crc
; ///< crc of this incremental
420 int get_net_marked_out(const OSDMap
*previous
) const;
421 int get_net_marked_down(const OSDMap
*previous
) const;
422 int identify_osd(uuid_d u
) const;
424 void encode_client_old(ceph::buffer::list
& bl
) const;
425 void encode_classic(ceph::buffer::list
& bl
, uint64_t features
) const;
426 void encode(ceph::buffer::list
& bl
, uint64_t features
=CEPH_FEATURES_ALL
) const;
427 void decode_classic(ceph::buffer::list::const_iterator
&p
);
428 void decode(ceph::buffer::list::const_iterator
&bl
);
429 void dump(ceph::Formatter
*f
) const;
430 static void generate_test_instances(std::list
<Incremental
*>& o
);
432 explicit Incremental(epoch_t e
=0) :
434 epoch(e
), new_pool_max(-1), new_flags(-1), new_max_osd(-1),
435 have_crc(false), full_crc(0), inc_crc(0) {
437 explicit Incremental(ceph::buffer::list
&bl
) {
438 auto p
= std::cbegin(bl
);
441 explicit Incremental(ceph::buffer::list::const_iterator
&p
) {
445 pg_pool_t
*get_new_pool(int64_t pool
, const pg_pool_t
*orig
) {
446 if (new_pools
.count(pool
) == 0)
447 new_pools
[pool
] = *orig
;
448 return &new_pools
[pool
];
450 bool has_erasure_code_profile(const std::string
&name
) const {
451 auto i
= new_erasure_code_profiles
.find(name
);
452 return i
!= new_erasure_code_profiles
.end();
454 void set_erasure_code_profile(const std::string
&name
,
455 const std::map
<std::string
,std::string
>& profile
) {
456 new_erasure_code_profiles
[name
] = profile
;
458 mempool::osdmap::map
<std::string
,std::map
<std::string
,std::string
>> get_erasure_code_profiles() const {
459 return new_erasure_code_profiles
;
462 /// propagate update pools' snap metadata to any of their tiers
463 int propagate_snaps_to_tiers(CephContext
*cct
, const OSDMap
&base
);
465 /// filter out osds with any pending state changing
466 size_t get_pending_state_osds(std::vector
<int> *osds
) {
470 for (auto &p
: new_state
) {
471 osds
->push_back(p
.first
);
477 bool pending_osd_has_state(int osd
, unsigned state
) {
478 return new_state
.count(osd
) && (new_state
[osd
] & state
) != 0;
481 bool pending_osd_state_set(int osd
, unsigned state
) {
482 if (pending_osd_has_state(osd
, state
))
484 new_state
[osd
] |= state
;
488 // cancel the specified pending osd state if there is any
489 // return ture on success, false otherwise.
490 bool pending_osd_state_clear(int osd
, unsigned state
) {
491 if (!pending_osd_has_state(osd
, state
)) {
492 // never has been set or already has been cancelled.
496 new_state
[osd
] &= ~state
;
497 if (!new_state
[osd
]) {
499 new_state
.erase(osd
);
504 bool in_new_removed_snaps(int64_t pool
, snapid_t snap
) const {
505 auto p
= new_removed_snaps
.find(pool
);
506 if (p
== new_removed_snaps
.end()) {
509 return p
->second
.contains(snap
);
515 epoch_t epoch
; // what epoch of the osd cluster descriptor is this
516 utime_t created
, modified
; // epoch start time
517 int32_t pool_max
; // the largest pool num, ever
521 int num_osd
; // not saved; see calc_num_osds
522 int num_up_osd
; // not saved; see calc_num_osds
523 int num_in_osd
; // not saved; see calc_num_osds
526 std::vector
<uint32_t> osd_state
;
528 mempool::osdmap::map
<int32_t,uint32_t> crush_node_flags
; // crush node -> CEPH_OSD_* flags
529 mempool::osdmap::map
<int32_t,uint32_t> device_class_flags
; // device class -> CEPH_OSD_* flags
531 utime_t last_up_change
, last_in_change
;
533 // These features affect OSDMap[::Incremental] encoding, or the
534 // encoding of some type embedded therein (CrushWrapper, something
535 // from osd_types, etc.).
536 static constexpr uint64_t SIGNIFICANT_FEATURES
=
537 CEPH_FEATUREMASK_PGID64
|
538 CEPH_FEATUREMASK_PGPOOL3
|
539 CEPH_FEATUREMASK_OSDENC
|
540 CEPH_FEATUREMASK_OSDMAP_ENC
|
541 CEPH_FEATUREMASK_OSD_POOLRESEND
|
542 CEPH_FEATUREMASK_NEW_OSDOP_ENCODING
|
543 CEPH_FEATUREMASK_MSG_ADDR2
|
544 CEPH_FEATUREMASK_CRUSH_TUNABLES5
|
545 CEPH_FEATUREMASK_CRUSH_CHOOSE_ARGS
|
546 CEPH_FEATUREMASK_SERVER_LUMINOUS
|
547 CEPH_FEATUREMASK_SERVER_MIMIC
|
548 CEPH_FEATUREMASK_SERVER_NAUTILUS
|
549 CEPH_FEATUREMASK_SERVER_OCTOPUS
;
552 mempool::osdmap::vector
<std::shared_ptr
<entity_addrvec_t
> > client_addrs
;
553 mempool::osdmap::vector
<std::shared_ptr
<entity_addrvec_t
> > cluster_addrs
;
554 mempool::osdmap::vector
<std::shared_ptr
<entity_addrvec_t
> > hb_back_addrs
;
555 mempool::osdmap::vector
<std::shared_ptr
<entity_addrvec_t
> > hb_front_addrs
;
557 std::shared_ptr
<addrs_s
> osd_addrs
;
559 entity_addrvec_t _blank_addrvec
;
561 mempool::osdmap::vector
<__u32
> osd_weight
; // 16.16 fixed point, 0x10000 = "in", 0 = "out"
562 mempool::osdmap::vector
<osd_info_t
> osd_info
;
563 std::shared_ptr
<PGTempMap
> pg_temp
; // temp pg mapping (e.g. while we rebuild)
564 std::shared_ptr
< mempool::osdmap::map
<pg_t
,int32_t > > primary_temp
; // temp primary mapping (e.g. while we rebuild)
565 std::shared_ptr
< mempool::osdmap::vector
<__u32
> > osd_primary_affinity
; ///< 16.16 fixed point, 0x10000 = baseline
567 // remap (post-CRUSH, pre-up)
568 mempool::osdmap::map
<pg_t
,mempool::osdmap::vector
<int32_t>> pg_upmap
; ///< remap pg
569 mempool::osdmap::map
<pg_t
,mempool::osdmap::vector
<std::pair
<int32_t,int32_t>>> pg_upmap_items
; ///< remap osds in up set
571 mempool::osdmap::map
<int64_t,pg_pool_t
> pools
;
572 mempool::osdmap::map
<int64_t,std::string
> pool_name
;
573 mempool::osdmap::map
<std::string
, std::map
<std::string
,std::string
>> erasure_code_profiles
;
574 mempool::osdmap::map
<std::string
,int64_t> name_pool
;
576 std::shared_ptr
< mempool::osdmap::vector
<uuid_d
> > osd_uuid
;
577 mempool::osdmap::vector
<osd_xinfo_t
> osd_xinfo
;
579 mempool::osdmap::unordered_map
<entity_addr_t
,utime_t
> blacklist
;
581 /// queue of snaps to remove
582 mempool::osdmap::map
<int64_t, snap_interval_set_t
> removed_snaps_queue
;
584 /// removed_snaps additions this epoch
585 mempool::osdmap::map
<int64_t, snap_interval_set_t
> new_removed_snaps
;
587 /// removed_snaps removals this epoch
588 mempool::osdmap::map
<int64_t, snap_interval_set_t
> new_purged_snaps
;
590 epoch_t cluster_snapshot_epoch
;
591 std::string cluster_snapshot
;
592 bool new_blacklist_entries
;
594 float full_ratio
= 0, backfillfull_ratio
= 0, nearfull_ratio
= 0;
596 /// min compat client we want to support
597 ceph_release_t require_min_compat_client
{ceph_release_t::unknown
};
600 /// require osds to run at least this release
601 ceph_release_t require_osd_release
{ceph_release_t::unknown
};
604 mutable uint64_t cached_up_osd_features
;
606 mutable bool crc_defined
;
607 mutable uint32_t crc
;
609 void _calc_up_osd_features();
612 bool have_crc() const { return crc_defined
; }
613 uint32_t get_crc() const { return crc
; }
615 std::shared_ptr
<CrushWrapper
> crush
; // hierarchical map
617 uint32_t crush_version
= 1;
619 friend class OSDMonitor
;
625 num_osd(0), num_up_osd(0), num_in_osd(0),
627 osd_addrs(std::make_shared
<addrs_s
>()),
628 pg_temp(std::make_shared
<PGTempMap
>()),
629 primary_temp(std::make_shared
<mempool::osdmap::map
<pg_t
,int32_t>>()),
630 osd_uuid(std::make_shared
<mempool::osdmap::vector
<uuid_d
>>()),
631 cluster_snapshot_epoch(0),
632 new_blacklist_entries(false),
633 cached_up_osd_features(0),
634 crc_defined(false), crc(0),
635 crush(std::make_shared
<CrushWrapper
>()) {
639 OSDMap(const OSDMap
& other
) = default;
640 OSDMap
& operator=(const OSDMap
& other
) = default;
643 /// return feature mask subset that is relevant to OSDMap encoding
644 static uint64_t get_significant_features(uint64_t features
) {
645 return SIGNIFICANT_FEATURES
& features
;
648 uint64_t get_encoding_features() const;
650 void deepish_copy_from(const OSDMap
& o
) {
652 primary_temp
.reset(new mempool::osdmap::map
<pg_t
,int32_t>(*o
.primary_temp
));
653 pg_temp
.reset(new PGTempMap(*o
.pg_temp
));
654 osd_uuid
.reset(new mempool::osdmap::vector
<uuid_d
>(*o
.osd_uuid
));
656 if (o
.osd_primary_affinity
)
657 osd_primary_affinity
.reset(new mempool::osdmap::vector
<__u32
>(*o
.osd_primary_affinity
));
659 // NOTE: this still references shared entity_addrvec_t's.
660 osd_addrs
.reset(new addrs_s(*o
.osd_addrs
));
662 // NOTE: we do not copy crush. note that apply_incremental will
663 // allocate a new CrushWrapper, though.
667 const uuid_d
& get_fsid() const { return fsid
; }
668 void set_fsid(uuid_d
& f
) { fsid
= f
; }
670 epoch_t
get_epoch() const { return epoch
; }
671 void inc_epoch() { epoch
++; }
673 void set_epoch(epoch_t e
);
675 uint32_t get_crush_version() const {
676 return crush_version
;
680 const utime_t
& get_created() const { return created
; }
681 const utime_t
& get_modified() const { return modified
; }
683 bool is_blacklisted(const entity_addr_t
& a
) const;
684 bool is_blacklisted(const entity_addrvec_t
& a
) const;
685 void get_blacklist(std::list
<std::pair
<entity_addr_t
,utime_t
> > *bl
) const;
686 void get_blacklist(std::set
<entity_addr_t
> *bl
) const;
688 std::string
get_cluster_snapshot() const {
689 if (cluster_snapshot_epoch
== epoch
)
690 return cluster_snapshot
;
691 return std::string();
694 float get_full_ratio() const {
697 float get_backfillfull_ratio() const {
698 return backfillfull_ratio
;
700 float get_nearfull_ratio() const {
701 return nearfull_ratio
;
703 void get_full_pools(CephContext
*cct
,
704 std::set
<int64_t> *full
,
705 std::set
<int64_t> *backfillfull
,
706 std::set
<int64_t> *nearfull
) const;
707 void get_full_osd_counts(std::set
<int> *full
, std::set
<int> *backfill
,
708 std::set
<int> *nearfull
) const;
711 /***** cluster state *****/
713 int get_max_osd() const { return max_osd
; }
714 void set_max_osd(int m
);
716 unsigned get_num_osds() const {
719 unsigned get_num_up_osds() const {
722 unsigned get_num_in_osds() const {
725 /// recalculate cached values for get_num{,_up,_in}_osds
728 void get_all_osds(std::set
<int32_t>& ls
) const;
729 void get_up_osds(std::set
<int32_t>& ls
) const;
730 void get_out_existing_osds(std::set
<int32_t>& ls
) const;
731 unsigned get_num_pg_temp() const {
732 return pg_temp
->size();
735 int get_flags() const { return flags
; }
736 bool test_flag(int f
) const { return flags
& f
; }
737 void set_flag(int f
) { flags
|= f
; }
738 void clear_flag(int f
) { flags
&= ~f
; }
740 void get_flag_set(std::set
<std::string
> *flagset
) const;
742 static void calc_state_set(int state
, std::set
<std::string
>& st
);
744 int get_state(int o
) const {
745 ceph_assert(o
< max_osd
);
748 int get_state(int o
, std::set
<std::string
>& st
) const {
749 ceph_assert(o
< max_osd
);
750 unsigned t
= osd_state
[o
];
751 calc_state_set(t
, st
);
754 void set_state(int o
, unsigned s
) {
755 ceph_assert(o
< max_osd
);
758 void set_weight(int o
, unsigned w
) {
759 ceph_assert(o
< max_osd
);
762 osd_state
[o
] |= CEPH_OSD_EXISTS
;
764 unsigned get_weight(int o
) const {
765 ceph_assert(o
< max_osd
);
766 return osd_weight
[o
];
768 float get_weightf(int o
) const {
769 return (float)get_weight(o
) / (float)CEPH_OSD_IN
;
771 void adjust_osd_weights(const std::map
<int,double>& weights
, Incremental
& inc
) const;
773 void set_primary_affinity(int o
, int w
) {
774 ceph_assert(o
< max_osd
);
775 if (!osd_primary_affinity
)
776 osd_primary_affinity
.reset(
777 new mempool::osdmap::vector
<__u32
>(
778 max_osd
, CEPH_OSD_DEFAULT_PRIMARY_AFFINITY
));
779 (*osd_primary_affinity
)[o
] = w
;
781 unsigned get_primary_affinity(int o
) const {
782 ceph_assert(o
< max_osd
);
783 if (!osd_primary_affinity
)
784 return CEPH_OSD_DEFAULT_PRIMARY_AFFINITY
;
785 return (*osd_primary_affinity
)[o
];
787 float get_primary_affinityf(int o
) const {
788 return (float)get_primary_affinity(o
) / (float)CEPH_OSD_MAX_PRIMARY_AFFINITY
;
791 bool has_erasure_code_profile(const std::string
&name
) const {
792 auto i
= erasure_code_profiles
.find(name
);
793 return i
!= erasure_code_profiles
.end();
795 int get_erasure_code_profile_default(CephContext
*cct
,
796 std::map
<std::string
,std::string
> &profile_map
,
798 void set_erasure_code_profile(const std::string
&name
,
799 const std::map
<std::string
,std::string
>& profile
) {
800 erasure_code_profiles
[name
] = profile
;
802 const std::map
<std::string
,std::string
> &get_erasure_code_profile(
803 const std::string
&name
) const {
804 static std::map
<std::string
,std::string
> empty
;
805 auto i
= erasure_code_profiles
.find(name
);
806 if (i
== erasure_code_profiles
.end())
811 const mempool::osdmap::map
<std::string
,std::map
<std::string
,std::string
>> &get_erasure_code_profiles() const {
812 return erasure_code_profiles
;
815 bool exists(int osd
) const {
817 return osd
>= 0 && osd
< max_osd
&& (osd_state
[osd
] & CEPH_OSD_EXISTS
);
820 bool is_destroyed(int osd
) const {
821 return exists(osd
) && (osd_state
[osd
] & CEPH_OSD_DESTROYED
);
824 bool is_up(int osd
) const {
825 return exists(osd
) && (osd_state
[osd
] & CEPH_OSD_UP
);
828 bool has_been_up_since(int osd
, epoch_t epoch
) const {
829 return is_up(osd
) && get_up_from(osd
) <= epoch
;
832 bool is_down(int osd
) const {
836 bool is_stop(int osd
) const {
837 return exists(osd
) && is_down(osd
) &&
838 (osd_state
[osd
] & CEPH_OSD_STOP
);
841 bool is_out(int osd
) const {
842 return !exists(osd
) || get_weight(osd
) == CEPH_OSD_OUT
;
845 bool is_in(int osd
) const {
849 bool is_dead(int osd
) const {
851 return false; // unclear if they know they are removed from map
853 return get_xinfo(osd
).dead_epoch
> get_info(osd
).up_from
;
856 unsigned get_osd_crush_node_flags(int osd
) const;
857 unsigned get_crush_node_flags(int id
) const;
858 unsigned get_device_class_flags(int id
) const;
860 bool is_noup_by_osd(int osd
) const {
861 return exists(osd
) && (osd_state
[osd
] & CEPH_OSD_NOUP
);
864 bool is_nodown_by_osd(int osd
) const {
865 return exists(osd
) && (osd_state
[osd
] & CEPH_OSD_NODOWN
);
868 bool is_noin_by_osd(int osd
) const {
869 return exists(osd
) && (osd_state
[osd
] & CEPH_OSD_NOIN
);
872 bool is_noout_by_osd(int osd
) const {
873 return exists(osd
) && (osd_state
[osd
] & CEPH_OSD_NOOUT
);
876 bool is_noup(int osd
) const {
877 if (test_flag(CEPH_OSDMAP_NOUP
)) // global?
879 if (is_noup_by_osd(osd
)) // by osd?
881 if (get_osd_crush_node_flags(osd
) & CEPH_OSD_NOUP
) // by crush-node?
883 if (auto class_id
= crush
->get_item_class_id(osd
); class_id
>= 0 &&
884 get_device_class_flags(class_id
) & CEPH_OSD_NOUP
) // by device-class?
889 bool is_nodown(int osd
) const {
890 if (test_flag(CEPH_OSDMAP_NODOWN
))
892 if (is_nodown_by_osd(osd
))
894 if (get_osd_crush_node_flags(osd
) & CEPH_OSD_NODOWN
)
896 if (auto class_id
= crush
->get_item_class_id(osd
); class_id
>= 0 &&
897 get_device_class_flags(class_id
) & CEPH_OSD_NODOWN
)
902 bool is_noin(int osd
) const {
903 if (test_flag(CEPH_OSDMAP_NOIN
))
905 if (is_noin_by_osd(osd
))
907 if (get_osd_crush_node_flags(osd
) & CEPH_OSD_NOIN
)
909 if (auto class_id
= crush
->get_item_class_id(osd
); class_id
>= 0 &&
910 get_device_class_flags(class_id
) & CEPH_OSD_NOIN
)
915 bool is_noout(int osd
) const {
916 if (test_flag(CEPH_OSDMAP_NOOUT
))
918 if (is_noout_by_osd(osd
))
920 if (get_osd_crush_node_flags(osd
) & CEPH_OSD_NOOUT
)
922 if (auto class_id
= crush
->get_item_class_id(osd
); class_id
>= 0 &&
923 get_device_class_flags(class_id
) & CEPH_OSD_NOOUT
)
929 * check if an entire crush subtree is down
931 bool subtree_is_down(int id
, std::set
<int> *down_cache
) const;
932 bool containing_subtree_is_down(CephContext
*cct
, int osd
, int subtree_type
, std::set
<int> *down_cache
) const;
934 bool subtree_type_is_down(CephContext
*cct
, int id
, int subtree_type
, std::set
<int> *down_in_osds
, std::set
<int> *up_in_osds
,
935 std::set
<int> *subtree_up
, std::unordered_map
<int, std::set
<int> > *subtree_type_down
) const;
937 int identify_osd(const entity_addr_t
& addr
) const;
938 int identify_osd(const uuid_d
& u
) const;
939 int identify_osd_on_all_channels(const entity_addr_t
& addr
) const;
941 bool have_addr(const entity_addr_t
& addr
) const {
942 return identify_osd(addr
) >= 0;
944 int find_osd_on_ip(const entity_addr_t
& ip
) const;
946 const entity_addrvec_t
& get_addrs(int osd
) const {
947 ceph_assert(exists(osd
));
948 return osd_addrs
->client_addrs
[osd
] ?
949 *osd_addrs
->client_addrs
[osd
] : _blank_addrvec
;
951 const entity_addrvec_t
& get_most_recent_addrs(int osd
) const {
952 return get_addrs(osd
);
954 const entity_addrvec_t
&get_cluster_addrs(int osd
) const {
955 ceph_assert(exists(osd
));
956 return osd_addrs
->cluster_addrs
[osd
] ?
957 *osd_addrs
->cluster_addrs
[osd
] : _blank_addrvec
;
959 const entity_addrvec_t
&get_hb_back_addrs(int osd
) const {
960 ceph_assert(exists(osd
));
961 return osd_addrs
->hb_back_addrs
[osd
] ?
962 *osd_addrs
->hb_back_addrs
[osd
] : _blank_addrvec
;
964 const entity_addrvec_t
&get_hb_front_addrs(int osd
) const {
965 ceph_assert(exists(osd
));
966 return osd_addrs
->hb_front_addrs
[osd
] ?
967 *osd_addrs
->hb_front_addrs
[osd
] : _blank_addrvec
;
970 const uuid_d
& get_uuid(int osd
) const {
971 ceph_assert(exists(osd
));
972 return (*osd_uuid
)[osd
];
975 const epoch_t
& get_up_from(int osd
) const {
976 ceph_assert(exists(osd
));
977 return osd_info
[osd
].up_from
;
979 const epoch_t
& get_up_thru(int osd
) const {
980 ceph_assert(exists(osd
));
981 return osd_info
[osd
].up_thru
;
983 const epoch_t
& get_down_at(int osd
) const {
984 ceph_assert(exists(osd
));
985 return osd_info
[osd
].down_at
;
987 const osd_info_t
& get_info(int osd
) const {
988 ceph_assert(osd
< max_osd
);
989 return osd_info
[osd
];
992 const osd_xinfo_t
& get_xinfo(int osd
) const {
993 ceph_assert(osd
< max_osd
);
994 return osd_xinfo
[osd
];
997 int get_next_up_osd_after(int n
) const {
998 if (get_max_osd() == 0)
1000 for (int i
= n
+ 1; i
!= n
; ++i
) {
1001 if (i
>= get_max_osd())
1011 int get_previous_up_osd_before(int n
) const {
1012 if (get_max_osd() == 0)
1014 for (int i
= n
- 1; i
!= n
; --i
) {
1016 i
= get_max_osd() - 1;
1026 void get_random_up_osds_by_subtree(int n
, // whoami
1027 std::string
&subtree
,
1028 int limit
, // how many
1030 std::set
<int> *want
) const;
1033 * get feature bits required by the current structure
1035 * @param entity_type [in] what entity type we are asking about
1036 * @param mask [out] std::set of all possible map-related features we could std::set
1037 * @return feature bits used by this map
1039 uint64_t get_features(int entity_type
, uint64_t *mask
) const;
1042 * get oldest *client* version (firefly, hammer, etc.) that can connect given
1043 * the feature bits required (according to get_features()).
1045 ceph_release_t
get_min_compat_client() const;
1048 * gets the required minimum *client* version that can connect to the cluster.
1050 ceph_release_t
get_require_min_compat_client() const;
1053 * get intersection of features supported by up osds
1055 uint64_t get_up_osd_features() const;
1057 void get_upmap_pgs(vector
<pg_t
> *upmap_pgs
) const;
1058 bool check_pg_upmaps(
1060 const vector
<pg_t
>& to_check
,
1061 vector
<pg_t
> *to_cancel
,
1062 map
<pg_t
, mempool::osdmap::vector
<pair
<int,int>>> *to_remap
) const;
1063 void clean_pg_upmaps(
1065 Incremental
*pending_inc
,
1066 const vector
<pg_t
>& to_cancel
,
1067 const map
<pg_t
, mempool::osdmap::vector
<pair
<int,int>>>& to_remap
) const;
1068 bool clean_pg_upmaps(CephContext
*cct
, Incremental
*pending_inc
) const;
1070 int apply_incremental(const Incremental
&inc
);
1072 /// try to re-use/reference addrs in oldmap from newmap
1073 static void dedup(const OSDMap
*oldmap
, OSDMap
*newmap
);
1075 static void clean_temps(CephContext
*cct
,
1076 const OSDMap
& oldmap
,
1077 const OSDMap
& nextmap
,
1078 Incremental
*pending_inc
);
1080 // serialize, unserialize
1082 void encode_client_old(ceph::buffer::list
& bl
) const;
1083 void encode_classic(ceph::buffer::list
& bl
, uint64_t features
) const;
1084 void decode_classic(ceph::buffer::list::const_iterator
& p
);
1087 void encode(ceph::buffer::list
& bl
, uint64_t features
=CEPH_FEATURES_ALL
) const;
1088 void decode(ceph::buffer::list
& bl
);
1089 void decode(ceph::buffer::list::const_iterator
& bl
);
1092 /**** mapping facilities ****/
1095 const std::string
& name
,
1096 const std::string
& key
,
1097 const std::string
& nspace
,
1099 int object_locator_to_pg(const object_t
& oid
, const object_locator_t
& loc
,
1101 pg_t
object_locator_to_pg(const object_t
& oid
,
1102 const object_locator_t
& loc
) const {
1104 int ret
= object_locator_to_pg(oid
, loc
, pg
);
1105 ceph_assert(ret
== 0);
1110 static object_locator_t
file_to_object_locator(const file_layout_t
& layout
) {
1111 return object_locator_t(layout
.pool_id
, layout
.pool_ns
);
1114 ceph_object_layout
file_to_object_layout(object_t oid
,
1115 file_layout_t
& layout
) const {
1116 return make_object_layout(oid
, layout
.pool_id
, layout
.pool_ns
);
1119 ceph_object_layout
make_object_layout(object_t oid
, int pg_pool
,
1120 std::string nspace
) const;
1122 int get_pg_num(int pg_pool
) const
1124 const pg_pool_t
*pool
= get_pg_pool(pg_pool
);
1125 ceph_assert(NULL
!= pool
);
1126 return pool
->get_pg_num();
1129 bool pg_exists(pg_t pgid
) const {
1130 const pg_pool_t
*p
= get_pg_pool(pgid
.pool());
1131 return p
&& pgid
.ps() < p
->get_pg_num();
1134 int get_pg_pool_min_size(pg_t pgid
) const {
1135 if (!pg_exists(pgid
)) {
1138 const pg_pool_t
*p
= get_pg_pool(pgid
.pool());
1140 return p
->get_min_size();
1143 int get_pg_pool_size(pg_t pgid
) const {
1144 if (!pg_exists(pgid
)) {
1147 const pg_pool_t
*p
= get_pg_pool(pgid
.pool());
1149 return p
->get_size();
1152 int get_pg_pool_crush_rule(pg_t pgid
) const {
1153 if (!pg_exists(pgid
)) {
1156 const pg_pool_t
*p
= get_pg_pool(pgid
.pool());
1158 return p
->get_crush_rule();
1162 /// pg -> (raw osd std::list)
1163 void _pg_to_raw_osds(
1164 const pg_pool_t
& pool
, pg_t pg
,
1165 std::vector
<int> *osds
,
1167 int _pick_primary(const std::vector
<int>& osds
) const;
1168 void _remove_nonexistent_osds(const pg_pool_t
& pool
, std::vector
<int>& osds
) const;
1170 void _apply_primary_affinity(ps_t seed
, const pg_pool_t
& pool
,
1171 std::vector
<int> *osds
, int *primary
) const;
1173 /// apply pg_upmap[_items] mappings
1174 void _apply_upmap(const pg_pool_t
& pi
, pg_t pg
, std::vector
<int> *raw
) const;
1176 /// pg -> (up osd std::list)
1177 void _raw_to_up_osds(const pg_pool_t
& pool
, const std::vector
<int>& raw
,
1178 std::vector
<int> *up
) const;
1182 * Get the pg and primary temp, if they are specified.
1183 * @param temp_pg [out] Will be empty or contain the temp PG mapping on return
1184 * @param temp_primary [out] Will be the value in primary_temp, or a value derived
1185 * from the pg_temp (if specified), or -1 if you should use the calculated (up_)primary.
1187 void _get_temp_osds(const pg_pool_t
& pool
, pg_t pg
,
1188 std::vector
<int> *temp_pg
, int *temp_primary
) const;
1191 * map to up and acting. Fills in whatever fields are non-NULL.
1193 void _pg_to_up_acting_osds(const pg_t
& pg
, std::vector
<int> *up
, int *up_primary
,
1194 std::vector
<int> *acting
, int *acting_primary
,
1195 bool raw_pg_to_pg
= true) const;
1199 * This is suitable only for looking at raw CRUSH outputs. It skips
1200 * applying the temp and up checks and should not be used
1201 * by anybody for data mapping purposes.
1202 * raw and primary must be non-NULL
1204 void pg_to_raw_osds(pg_t pg
, std::vector
<int> *raw
, int *primary
) const;
1205 void pg_to_raw_upmap(pg_t pg
, std::vector
<int> *raw
,
1206 std::vector
<int> *raw_upmap
) const;
1207 /// map a pg to its acting set. @return acting set size
1208 void pg_to_acting_osds(const pg_t
& pg
, std::vector
<int> *acting
,
1209 int *acting_primary
) const {
1210 _pg_to_up_acting_osds(pg
, NULL
, NULL
, acting
, acting_primary
);
1212 void pg_to_acting_osds(pg_t pg
, std::vector
<int>& acting
) const {
1213 return pg_to_acting_osds(pg
, &acting
, NULL
);
1216 * This does not apply temp overrides and should not be used
1217 * by anybody for data mapping purposes. Specify both pointers.
1219 void pg_to_raw_up(pg_t pg
, std::vector
<int> *up
, int *primary
) const;
1221 * map a pg to its acting set as well as its up set. You must use
1222 * the acting set for data mapping purposes, but some users will
1223 * also find the up set useful for things like deciding what to
1225 * Each of these pointers must be non-NULL.
1227 void pg_to_up_acting_osds(pg_t pg
, std::vector
<int> *up
, int *up_primary
,
1228 std::vector
<int> *acting
, int *acting_primary
) const {
1229 _pg_to_up_acting_osds(pg
, up
, up_primary
, acting
, acting_primary
);
1231 void pg_to_up_acting_osds(pg_t pg
, std::vector
<int>& up
, std::vector
<int>& acting
) const {
1232 int up_primary
, acting_primary
;
1233 pg_to_up_acting_osds(pg
, &up
, &up_primary
, &acting
, &acting_primary
);
1235 bool pg_is_ec(pg_t pg
) const {
1236 auto i
= pools
.find(pg
.pool());
1237 ceph_assert(i
!= pools
.end());
1238 return i
->second
.is_erasure();
1240 bool get_primary_shard(const pg_t
& pgid
, spg_t
*out
) const {
1241 auto i
= get_pools().find(pgid
.pool());
1242 if (i
== get_pools().end()) {
1245 if (!i
->second
.is_erasure()) {
1250 std::vector
<int> acting
;
1251 pg_to_acting_osds(pgid
, &acting
, &primary
);
1252 for (uint8_t i
= 0; i
< acting
.size(); ++i
) {
1253 if (acting
[i
] == primary
) {
1254 *out
= spg_t(pgid
, shard_id_t(i
));
1260 bool get_primary_shard(const pg_t
& pgid
, int *primary
, spg_t
*out
) const {
1261 auto i
= get_pools().find(pgid
.pool());
1262 if (i
== get_pools().end()) {
1265 std::vector
<int> acting
;
1266 pg_to_acting_osds(pgid
, &acting
, primary
);
1267 if (i
->second
.is_erasure()) {
1268 for (uint8_t i
= 0; i
< acting
.size(); ++i
) {
1269 if (acting
[i
] == *primary
) {
1270 *out
= spg_t(pgid
, shard_id_t(i
));
1281 bool in_removed_snaps_queue(int64_t pool
, snapid_t snap
) const {
1282 auto p
= removed_snaps_queue
.find(pool
);
1283 if (p
== removed_snaps_queue
.end()) {
1286 return p
->second
.contains(snap
);
1289 const mempool::osdmap::map
<int64_t,snap_interval_set_t
>&
1290 get_removed_snaps_queue() const {
1291 return removed_snaps_queue
;
1293 const mempool::osdmap::map
<int64_t,snap_interval_set_t
>&
1294 get_new_removed_snaps() const {
1295 return new_removed_snaps
;
1297 const mempool::osdmap::map
<int64_t,snap_interval_set_t
>&
1298 get_new_purged_snaps() const {
1299 return new_purged_snaps
;
1302 int64_t lookup_pg_pool_name(const std::string
& name
) const {
1303 auto p
= name_pool
.find(name
);
1304 if (p
== name_pool
.end())
1309 int64_t get_pool_max() const {
1312 const mempool::osdmap::map
<int64_t,pg_pool_t
>& get_pools() const {
1315 mempool::osdmap::map
<int64_t,pg_pool_t
>& get_pools() {
1318 void get_pool_ids_by_rule(int rule_id
, std::set
<int64_t> *pool_ids
) const {
1319 ceph_assert(pool_ids
);
1320 for (auto &p
: pools
) {
1321 if (p
.second
.get_crush_rule() == rule_id
) {
1322 pool_ids
->insert(p
.first
);
1326 void get_pool_ids_by_osd(CephContext
*cct
,
1328 std::set
<int64_t> *pool_ids
) const;
1329 const std::string
& get_pool_name(int64_t p
) const {
1330 auto i
= pool_name
.find(p
);
1331 ceph_assert(i
!= pool_name
.end());
1334 const mempool::osdmap::map
<int64_t,std::string
>& get_pool_names() const {
1337 bool have_pg_pool(int64_t p
) const {
1338 return pools
.count(p
);
1340 const pg_pool_t
* get_pg_pool(int64_t p
) const {
1341 auto i
= pools
.find(p
);
1342 if (i
!= pools
.end())
1346 unsigned get_pg_size(pg_t pg
) const {
1347 auto p
= pools
.find(pg
.pool());
1348 ceph_assert(p
!= pools
.end());
1349 return p
->second
.get_size();
1351 int get_pg_type(pg_t pg
) const {
1352 auto p
= pools
.find(pg
.pool());
1353 ceph_assert(p
!= pools
.end());
1354 return p
->second
.get_type();
1356 int get_pool_crush_rule(int64_t pool_id
) const {
1357 auto pool
= get_pg_pool(pool_id
);
1360 return pool
->get_crush_rule();
1364 pg_t
raw_pg_to_pg(pg_t pg
) const {
1365 auto p
= pools
.find(pg
.pool());
1366 ceph_assert(p
!= pools
.end());
1367 return p
->second
.raw_pg_to_pg(pg
);
1370 // pg -> acting primary osd
1371 int get_pg_acting_primary(pg_t pg
) const {
1373 _pg_to_up_acting_osds(pg
, nullptr, nullptr, nullptr, &primary
);
1378 * check whether an spg_t maps to a particular osd
1380 bool is_up_acting_osd_shard(spg_t pg
, int osd
) const {
1381 std::vector
<int> up
, acting
;
1382 _pg_to_up_acting_osds(pg
.pgid
, &up
, NULL
, &acting
, NULL
, false);
1383 if (calc_pg_role(pg_shard_t(osd
, pg
.shard
), acting
) >= 0 ||
1384 calc_pg_role(pg_shard_t(osd
, pg
.shard
), up
) >= 0) {
1391 static int calc_pg_role_broken(int osd
, const std::vector
<int>& acting
, int nrep
=0);
1392 static int calc_pg_role(pg_shard_t who
, const std::vector
<int>& acting
);
1393 static bool primary_changed_broken(
1395 const std::vector
<int> &oldacting
,
1397 const std::vector
<int> &newacting
);
1399 /* rank is -1 (stray), 0 (primary), 1,2,3,... (replica) */
1400 int get_pg_acting_role(spg_t pg
, int osd
) const {
1401 std::vector
<int> group
;
1402 pg_to_acting_osds(pg
.pgid
, group
);
1403 return calc_pg_role(pg_shard_t(osd
, pg
.shard
), group
);
1408 pg_t pg
, ///< pg to potentially remap
1409 const std::set
<int>& overfull
, ///< osds we'd want to evacuate
1410 const std::vector
<int>& underfull
, ///< osds to move to, in order of preference
1411 const std::vector
<int>& more_underfull
, ///< less full osds to move to, in order of preference
1412 std::vector
<int> *orig
,
1413 std::vector
<int> *out
); ///< resulting alternative mapping
1417 uint32_t max_deviation
, ///< max deviation from target (value >= 1)
1418 int max_iterations
, ///< max iterations to run
1419 const std::set
<int64_t>& pools
, ///< [optional] restrict to pool
1420 Incremental
*pending_inc
1423 int get_osds_by_bucket_name(const std::string
&name
, std::set
<int> *osds
) const;
1425 bool have_pg_upmaps(pg_t pg
) const {
1426 return pg_upmap
.count(pg
) ||
1427 pg_upmap_items
.count(pg
);
1430 bool check_full(const set
<pg_shard_t
> &missing_on
) const {
1431 for (auto shard
: missing_on
) {
1432 if (get_state(shard
.osd
) & CEPH_OSD_FULL
)
1439 * handy helpers to build simple maps...
1442 * Build an OSD map suitable for basic usage. If **num_osd** is >= 0
1443 * it will be initialized with the specified number of OSDs in a
1444 * single host. If **num_osd** is < 0 the layout of the OSD map will
1445 * be built by reading the content of the configuration file.
1447 * @param cct [in] in core ceph context
1448 * @param e [in] initial epoch
1449 * @param fsid [in] id of the cluster
1450 * @param num_osd [in] number of OSDs if >= 0 or read from conf if < 0
1451 * @return **0** on success, negative errno on error.
1454 int build_simple_optioned(CephContext
*cct
, epoch_t e
, uuid_d
&fsid
,
1455 int num_osd
, int pg_bits
, int pgp_bits
,
1458 int build_simple(CephContext
*cct
, epoch_t e
, uuid_d
&fsid
,
1460 return build_simple_optioned(cct
, e
, fsid
, num_osd
, 0, 0, false);
1462 int build_simple_with_pool(CephContext
*cct
, epoch_t e
, uuid_d
&fsid
,
1463 int num_osd
, int pg_bits
, int pgp_bits
) {
1464 return build_simple_optioned(cct
, e
, fsid
, num_osd
,
1465 pg_bits
, pgp_bits
, true);
1467 static int _build_crush_types(CrushWrapper
& crush
);
1468 static int build_simple_crush_map(CephContext
*cct
, CrushWrapper
& crush
,
1469 int num_osd
, std::ostream
*ss
);
1470 static int build_simple_crush_map_from_conf(CephContext
*cct
,
1471 CrushWrapper
& crush
,
1473 static int build_simple_crush_rules(
1474 CephContext
*cct
, CrushWrapper
& crush
,
1475 const std::string
& root
,
1478 bool crush_rule_in_use(int rule_id
) const;
1480 int validate_crush_rules(CrushWrapper
*crush
, std::ostream
*ss
) const;
1484 primary_temp
->clear();
1488 void print_osd_line(int cur
, std::ostream
*out
, ceph::Formatter
*f
) const;
1490 void print(std::ostream
& out
) const;
1491 void print_osd(int id
, std::ostream
& out
) const;
1492 void print_osds(std::ostream
& out
) const;
1493 void print_pools(std::ostream
& out
) const;
1494 void print_summary(ceph::Formatter
*f
, std::ostream
& out
,
1495 const std::string
& prefix
, bool extra
=false) const;
1496 void print_oneline_summary(std::ostream
& out
) const;
1499 DUMP_IN
= 1, // only 'in' osds
1500 DUMP_OUT
= 2, // only 'out' osds
1501 DUMP_UP
= 4, // only 'up' osds
1502 DUMP_DOWN
= 8, // only 'down' osds
1503 DUMP_DESTROYED
= 16, // only 'destroyed' osds
1505 void print_tree(ceph::Formatter
*f
, std::ostream
*out
,
1506 unsigned dump_flags
=0, std::string bucket
="") const;
1508 int summarize_mapping_stats(
1510 const std::set
<int64_t> *pools
,
1512 ceph::Formatter
*f
) const;
1514 std::string
get_flag_string() const;
1515 static std::string
get_flag_string(unsigned flags
);
1516 static void dump_erasure_code_profiles(
1517 const mempool::osdmap::map
<std::string
,std::map
<std::string
,std::string
> > &profiles
,
1518 ceph::Formatter
*f
);
1519 void dump(ceph::Formatter
*f
) const;
1520 void dump_osd(int id
, ceph::Formatter
*f
) const;
1521 void dump_osds(ceph::Formatter
*f
) const;
1522 static void generate_test_instances(std::list
<OSDMap
*>& o
);
1523 bool check_new_blacklist_entries() const { return new_blacklist_entries
; }
1525 void check_health(CephContext
*cct
, health_check_map_t
*checks
) const;
1527 int parse_osd_id_list(const std::vector
<std::string
>& ls
,
1529 std::ostream
*ss
) const;
1531 float pool_raw_used_rate(int64_t poolid
) const;
1534 WRITE_CLASS_ENCODER_FEATURES(OSDMap
)
1535 WRITE_CLASS_ENCODER_FEATURES(OSDMap::Incremental
)
1538 using OSDMapRef
= boost::local_shared_ptr
<const OSDMap
>;
1540 using OSDMapRef
= std::shared_ptr
<const OSDMap
>;
1544 inline std::ostream
& operator<<(std::ostream
& out
, const OSDMap
& m
) {
1545 m
.print_oneline_summary(out
);
1551 void print_osd_utilization(const OSDMap
& osdmap
,
1556 const std::string
& filter
);