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 #ifndef CEPH_OSD_TYPES_H
19 #define CEPH_OSD_TYPES_H
24 #include <boost/scoped_ptr.hpp>
25 #include <boost/optional/optional_io.hpp>
26 #include <boost/variant.hpp>
28 #include "include/rados/rados_types.hpp"
29 #include "include/mempool.h"
31 #include "msg/msg_types.h"
32 #include "include/types.h"
33 #include "include/utime.h"
34 #include "include/CompatSet.h"
35 #include "common/histogram.h"
36 #include "include/interval_set.h"
37 #include "include/inline_memory.h"
38 #include "common/Formatter.h"
39 #include "common/bloom_filter.hpp"
40 #include "common/hobject.h"
41 #include "common/snap_types.h"
44 #include "include/cmp.h"
45 #include "librados/ListObjectImpl.h"
46 #include "compressor/Compressor.h"
49 #define CEPH_OSD_ONDISK_MAGIC "ceph osd volume v026"
51 #define CEPH_OSD_FEATURE_INCOMPAT_BASE CompatSet::Feature(1, "initial feature set(~v.18)")
52 #define CEPH_OSD_FEATURE_INCOMPAT_PGINFO CompatSet::Feature(2, "pginfo object")
53 #define CEPH_OSD_FEATURE_INCOMPAT_OLOC CompatSet::Feature(3, "object locator")
54 #define CEPH_OSD_FEATURE_INCOMPAT_LEC CompatSet::Feature(4, "last_epoch_clean")
55 #define CEPH_OSD_FEATURE_INCOMPAT_CATEGORIES CompatSet::Feature(5, "categories")
56 #define CEPH_OSD_FEATURE_INCOMPAT_HOBJECTPOOL CompatSet::Feature(6, "hobjectpool")
57 #define CEPH_OSD_FEATURE_INCOMPAT_BIGINFO CompatSet::Feature(7, "biginfo")
58 #define CEPH_OSD_FEATURE_INCOMPAT_LEVELDBINFO CompatSet::Feature(8, "leveldbinfo")
59 #define CEPH_OSD_FEATURE_INCOMPAT_LEVELDBLOG CompatSet::Feature(9, "leveldblog")
60 #define CEPH_OSD_FEATURE_INCOMPAT_SNAPMAPPER CompatSet::Feature(10, "snapmapper")
61 #define CEPH_OSD_FEATURE_INCOMPAT_SHARDS CompatSet::Feature(11, "sharded objects")
62 #define CEPH_OSD_FEATURE_INCOMPAT_HINTS CompatSet::Feature(12, "transaction hints")
63 #define CEPH_OSD_FEATURE_INCOMPAT_PGMETA CompatSet::Feature(13, "pg meta object")
64 #define CEPH_OSD_FEATURE_INCOMPAT_MISSING CompatSet::Feature(14, "explicit missing set")
65 #define CEPH_OSD_FEATURE_INCOMPAT_FASTINFO CompatSet::Feature(15, "fastinfo pg attr")
66 #define CEPH_OSD_FEATURE_INCOMPAT_RECOVERY_DELETES CompatSet::Feature(16, "deletes in missing set")
69 /// min recovery priority for MBackfillReserve
70 #define OSD_RECOVERY_PRIORITY_MIN 0
72 /// base backfill priority for MBackfillReserve
73 #define OSD_BACKFILL_PRIORITY_BASE 100
75 /// base backfill priority for MBackfillReserve (degraded PG)
76 #define OSD_BACKFILL_DEGRADED_PRIORITY_BASE 140
78 /// base recovery priority for MBackfillReserve
79 #define OSD_RECOVERY_PRIORITY_BASE 180
81 /// base backfill priority for MBackfillReserve (inactive PG)
82 #define OSD_BACKFILL_INACTIVE_PRIORITY_BASE 220
84 /// max manually/automatically set recovery priority for MBackfillReserve
85 #define OSD_RECOVERY_PRIORITY_MAX 254
87 /// max recovery priority for MBackfillReserve, only when forced manually
88 #define OSD_RECOVERY_PRIORITY_FORCED 255
91 typedef hobject_t collection_list_handle_t
;
93 /// convert a single CPEH_OSD_FLAG_* to a string
94 const char *ceph_osd_flag_name(unsigned flag
);
95 /// convert a single CEPH_OSD_OF_FLAG_* to a string
96 const char *ceph_osd_op_flag_name(unsigned flag
);
98 /// convert CEPH_OSD_FLAG_* op flags to a string
99 string
ceph_osd_flag_string(unsigned flags
);
100 /// conver CEPH_OSD_OP_FLAG_* op flags to a string
101 string
ceph_osd_op_flag_string(unsigned flags
);
102 /// conver CEPH_OSD_ALLOC_HINT_FLAG_* op flags to a string
103 string
ceph_osd_alloc_hint_flag_string(unsigned flags
);
107 * osd request identifier
109 * caller name + incarnation# + tid to unique identify this request.
112 entity_name_t name
; // who
114 int32_t inc
; // incarnation
119 osd_reqid_t(const osd_reqid_t
& other
)
120 : name(other
.name
), tid(other
.tid
), inc(other
.inc
)
122 osd_reqid_t(const entity_name_t
& a
, int i
, ceph_tid_t t
)
123 : name(a
), tid(t
), inc(i
)
126 DENC(osd_reqid_t
, v
, p
) {
133 void dump(Formatter
*f
) const;
134 static void generate_test_instances(list
<osd_reqid_t
*>& o
);
136 WRITE_CLASS_DENC(osd_reqid_t
)
143 pg_shard_t() : osd(-1), shard(shard_id_t::NO_SHARD
) {}
144 explicit pg_shard_t(int osd
) : osd(osd
), shard(shard_id_t::NO_SHARD
) {}
145 pg_shard_t(int osd
, shard_id_t shard
) : osd(osd
), shard(shard
) {}
146 bool is_undefined() const {
149 void encode(bufferlist
&bl
) const;
150 void decode(bufferlist::iterator
&bl
);
151 void dump(Formatter
*f
) const {
152 f
->dump_unsigned("osd", osd
);
153 if (shard
!= shard_id_t::NO_SHARD
) {
154 f
->dump_unsigned("shard", shard
);
158 WRITE_CLASS_ENCODER(pg_shard_t
)
159 WRITE_EQ_OPERATORS_2(pg_shard_t
, osd
, shard
)
160 WRITE_CMP_OPERATORS_2(pg_shard_t
, osd
, shard
)
161 ostream
&operator<<(ostream
&lhs
, const pg_shard_t
&rhs
);
163 class IsPGRecoverablePredicate
{
166 * have encodes the shards available
168 virtual bool operator()(const set
<pg_shard_t
> &have
) const = 0;
169 virtual ~IsPGRecoverablePredicate() {}
172 class IsPGReadablePredicate
{
175 * have encodes the shards available
177 virtual bool operator()(const set
<pg_shard_t
> &have
) const = 0;
178 virtual ~IsPGReadablePredicate() {}
181 inline ostream
& operator<<(ostream
& out
, const osd_reqid_t
& r
) {
182 return out
<< r
.name
<< "." << r
.inc
<< ":" << r
.tid
;
185 inline bool operator==(const osd_reqid_t
& l
, const osd_reqid_t
& r
) {
186 return (l
.name
== r
.name
) && (l
.inc
== r
.inc
) && (l
.tid
== r
.tid
);
188 inline bool operator!=(const osd_reqid_t
& l
, const osd_reqid_t
& r
) {
189 return (l
.name
!= r
.name
) || (l
.inc
!= r
.inc
) || (l
.tid
!= r
.tid
);
191 inline bool operator<(const osd_reqid_t
& l
, const osd_reqid_t
& r
) {
192 return (l
.name
< r
.name
) || (l
.inc
< r
.inc
) ||
193 (l
.name
== r
.name
&& l
.inc
== r
.inc
&& l
.tid
< r
.tid
);
195 inline bool operator<=(const osd_reqid_t
& l
, const osd_reqid_t
& r
) {
196 return (l
.name
< r
.name
) || (l
.inc
< r
.inc
) ||
197 (l
.name
== r
.name
&& l
.inc
== r
.inc
&& l
.tid
<= r
.tid
);
199 inline bool operator>(const osd_reqid_t
& l
, const osd_reqid_t
& r
) { return !(l
<= r
); }
200 inline bool operator>=(const osd_reqid_t
& l
, const osd_reqid_t
& r
) { return !(l
< r
); }
203 template<> struct hash
<osd_reqid_t
> {
204 size_t operator()(const osd_reqid_t
&r
) const {
205 static hash
<uint64_t> H
;
206 return H(r
.name
.num() ^ r
.tid
^ r
.inc
);
214 // a locator constrains the placement of an object. mainly, which pool
216 struct object_locator_t
{
217 // You specify either the hash or the key -- not both
218 int64_t pool
; ///< pool id
219 string key
; ///< key string (if non-empty)
220 string nspace
; ///< namespace
221 int64_t hash
; ///< hash position (if >= 0)
223 explicit object_locator_t()
224 : pool(-1), hash(-1) {}
225 explicit object_locator_t(int64_t po
)
226 : pool(po
), hash(-1) {}
227 explicit object_locator_t(int64_t po
, int64_t ps
)
228 : pool(po
), hash(ps
) {}
229 explicit object_locator_t(int64_t po
, string ns
)
230 : pool(po
), nspace(ns
), hash(-1) {}
231 explicit object_locator_t(int64_t po
, string ns
, int64_t ps
)
232 : pool(po
), nspace(ns
), hash(ps
) {}
233 explicit object_locator_t(int64_t po
, string ns
, string s
)
234 : pool(po
), key(s
), nspace(ns
), hash(-1) {}
235 explicit object_locator_t(const hobject_t
& soid
)
236 : pool(soid
.pool
), key(soid
.get_key()), nspace(soid
.nspace
), hash(-1) {}
238 int64_t get_pool() const {
253 void encode(bufferlist
& bl
) const;
254 void decode(bufferlist::iterator
& p
);
255 void dump(Formatter
*f
) const;
256 static void generate_test_instances(list
<object_locator_t
*>& o
);
258 WRITE_CLASS_ENCODER(object_locator_t
)
260 inline bool operator==(const object_locator_t
& l
, const object_locator_t
& r
) {
261 return l
.pool
== r
.pool
&& l
.key
== r
.key
&& l
.nspace
== r
.nspace
&& l
.hash
== r
.hash
;
263 inline bool operator!=(const object_locator_t
& l
, const object_locator_t
& r
) {
267 inline ostream
& operator<<(ostream
& out
, const object_locator_t
& loc
)
269 out
<< "@" << loc
.pool
;
270 if (loc
.nspace
.length())
271 out
<< ";" << loc
.nspace
;
272 if (loc
.key
.length())
273 out
<< ":" << loc
.key
;
277 struct request_redirect_t
{
279 object_locator_t redirect_locator
; ///< this is authoritative
280 string redirect_object
; ///< If non-empty, the request goes to this object name
281 bufferlist osd_instructions
; ///< a bufferlist for the OSDs, passed but not interpreted by clients
283 friend ostream
& operator<<(ostream
& out
, const request_redirect_t
& redir
);
286 request_redirect_t() {}
287 explicit request_redirect_t(const object_locator_t
& orig
, int64_t rpool
) :
288 redirect_locator(orig
) { redirect_locator
.pool
= rpool
; }
289 explicit request_redirect_t(const object_locator_t
& rloc
) :
290 redirect_locator(rloc
) {}
291 explicit request_redirect_t(const object_locator_t
& orig
,
292 const string
& robj
) :
293 redirect_locator(orig
), redirect_object(robj
) {}
295 void set_instructions(const bufferlist
& bl
) { osd_instructions
= bl
; }
296 const bufferlist
& get_instructions() { return osd_instructions
; }
298 bool empty() const { return redirect_locator
.empty() &&
299 redirect_object
.empty(); }
301 void combine_with_locator(object_locator_t
& orig
, string
& obj
) const {
302 orig
= redirect_locator
;
303 if (!redirect_object
.empty())
304 obj
= redirect_object
;
307 void encode(bufferlist
& bl
) const;
308 void decode(bufferlist::iterator
& bl
);
309 void dump(Formatter
*f
) const;
310 static void generate_test_instances(list
<request_redirect_t
*>& o
);
312 WRITE_CLASS_ENCODER(request_redirect_t
)
314 inline ostream
& operator<<(ostream
& out
, const request_redirect_t
& redir
) {
315 out
<< "object " << redir
.redirect_object
<< ", locator{" << redir
.redirect_locator
<< "}";
319 // Internal OSD op flags - set by the OSD based on the op types
321 CEPH_OSD_RMW_FLAG_READ
= (1 << 1),
322 CEPH_OSD_RMW_FLAG_WRITE
= (1 << 2),
323 CEPH_OSD_RMW_FLAG_CLASS_READ
= (1 << 3),
324 CEPH_OSD_RMW_FLAG_CLASS_WRITE
= (1 << 4),
325 CEPH_OSD_RMW_FLAG_PGOP
= (1 << 5),
326 CEPH_OSD_RMW_FLAG_CACHE
= (1 << 6),
327 CEPH_OSD_RMW_FLAG_FORCE_PROMOTE
= (1 << 7),
328 CEPH_OSD_RMW_FLAG_SKIP_HANDLE_CACHE
= (1 << 8),
329 CEPH_OSD_RMW_FLAG_SKIP_PROMOTE
= (1 << 9),
330 CEPH_OSD_RMW_FLAG_RWORDERED
= (1 << 10),
336 #define OSD_SUPERBLOCK_GOBJECT ghobject_t(hobject_t(sobject_t(object_t("osd_superblock"), 0)))
338 // placement seed (a hash value)
339 typedef uint32_t ps_t
;
341 // old (v1) pg_t encoding (wrap old struct ceph_pg)
344 void encode(bufferlist
& bl
) const {
347 void decode(bufferlist::iterator
& bl
) {
351 WRITE_CLASS_ENCODER(old_pg_t
)
353 // placement group id
359 pg_t() : m_pool(0), m_seed(0), m_preferred(-1) {}
360 pg_t(ps_t seed
, uint64_t pool
, int pref
=-1) :
361 m_pool(pool
), m_seed(seed
), m_preferred(pref
) {}
362 // cppcheck-suppress noExplicitConstructor
363 pg_t(const ceph_pg
& cpg
) :
364 m_pool(cpg
.pool
), m_seed(cpg
.ps
), m_preferred((__s16
)cpg
.preferred
) {}
366 // cppcheck-suppress noExplicitConstructor
367 pg_t(const old_pg_t
& opg
) {
371 old_pg_t
get_old_pg() const {
373 assert(m_pool
< 0xffffffffull
);
376 o
.v
.preferred
= (__s16
)m_preferred
;
383 uint64_t pool() const {
386 int32_t preferred() const {
390 static const uint8_t calc_name_buf_size
= 36; // max length for max values len("18446744073709551615.ffffffff") + future suffix len("_head") + '\0'
391 char *calc_name(char *buf
, const char *suffix_backwords
) const;
393 void set_ps(ps_t p
) {
396 void set_pool(uint64_t p
) {
399 void set_preferred(int32_t osd
) {
403 pg_t
get_parent() const;
404 pg_t
get_ancestor(unsigned old_pg_num
) const;
406 int print(char *o
, int maxlen
) const;
407 bool parse(const char *s
);
409 bool is_split(unsigned old_pg_num
, unsigned new_pg_num
, set
<pg_t
> *pchildren
) const;
412 * Returns b such that for all object o:
413 * ~((~0)<<b) & o.hash) == 0 iff o is in the pg for *this
415 unsigned get_split_bits(unsigned pg_num
) const;
417 bool contains(int bits
, const ghobject_t
& oid
) {
418 return oid
.match(bits
, ps());
420 bool contains(int bits
, const hobject_t
& oid
) {
421 return oid
.match(bits
, ps());
424 hobject_t
get_hobj_start() const;
425 hobject_t
get_hobj_end(unsigned pg_num
) const;
427 void encode(bufferlist
& bl
) const {
430 ::encode(m_pool
, bl
);
431 ::encode(m_seed
, bl
);
432 ::encode(m_preferred
, bl
);
434 void decode(bufferlist::iterator
& bl
) {
437 ::decode(m_pool
, bl
);
438 ::decode(m_seed
, bl
);
439 ::decode(m_preferred
, bl
);
441 void decode_old(bufferlist::iterator
& bl
) {
446 void dump(Formatter
*f
) const;
447 static void generate_test_instances(list
<pg_t
*>& o
);
449 WRITE_CLASS_ENCODER(pg_t
)
451 inline bool operator<(const pg_t
& l
, const pg_t
& r
) {
452 return l
.pool() < r
.pool() ||
453 (l
.pool() == r
.pool() && (l
.preferred() < r
.preferred() ||
454 (l
.preferred() == r
.preferred() && (l
.ps() < r
.ps()))));
456 inline bool operator<=(const pg_t
& l
, const pg_t
& r
) {
457 return l
.pool() < r
.pool() ||
458 (l
.pool() == r
.pool() && (l
.preferred() < r
.preferred() ||
459 (l
.preferred() == r
.preferred() && (l
.ps() <= r
.ps()))));
461 inline bool operator==(const pg_t
& l
, const pg_t
& r
) {
462 return l
.pool() == r
.pool() &&
463 l
.preferred() == r
.preferred() &&
466 inline bool operator!=(const pg_t
& l
, const pg_t
& r
) {
467 return l
.pool() != r
.pool() ||
468 l
.preferred() != r
.preferred() ||
471 inline bool operator>(const pg_t
& l
, const pg_t
& r
) {
472 return l
.pool() > r
.pool() ||
473 (l
.pool() == r
.pool() && (l
.preferred() > r
.preferred() ||
474 (l
.preferred() == r
.preferred() && (l
.ps() > r
.ps()))));
476 inline bool operator>=(const pg_t
& l
, const pg_t
& r
) {
477 return l
.pool() > r
.pool() ||
478 (l
.pool() == r
.pool() && (l
.preferred() > r
.preferred() ||
479 (l
.preferred() == r
.preferred() && (l
.ps() >= r
.ps()))));
482 ostream
& operator<<(ostream
& out
, const pg_t
&pg
);
485 template<> struct hash
< pg_t
>
487 size_t operator()( const pg_t
& x
) const
489 static hash
<uint32_t> H
;
490 return H((x
.pool() & 0xffffffff) ^ (x
.pool() >> 32) ^ x
.ps() ^ x
.preferred());
498 spg_t() : shard(shard_id_t::NO_SHARD
) {}
499 spg_t(pg_t pgid
, shard_id_t shard
) : pgid(pgid
), shard(shard
) {}
500 explicit spg_t(pg_t pgid
) : pgid(pgid
), shard(shard_id_t::NO_SHARD
) {}
501 unsigned get_split_bits(unsigned pg_num
) const {
502 return pgid
.get_split_bits(pg_num
);
504 spg_t
get_parent() const {
505 return spg_t(pgid
.get_parent(), shard
);
510 uint64_t pool() const {
513 int32_t preferred() const {
514 return pgid
.preferred();
517 static const uint8_t calc_name_buf_size
= pg_t::calc_name_buf_size
+ 4; // 36 + len('s') + len("255");
518 char *calc_name(char *buf
, const char *suffix_backwords
) const;
520 bool parse(const char *s
);
521 bool parse(const std::string
& s
) {
522 return parse(s
.c_str());
524 bool is_split(unsigned old_pg_num
, unsigned new_pg_num
,
525 set
<spg_t
> *pchildren
) const {
527 set
<pg_t
> *children
= pchildren
? &_children
: NULL
;
528 bool is_split
= pgid
.is_split(old_pg_num
, new_pg_num
, children
);
529 if (pchildren
&& is_split
) {
530 for (set
<pg_t
>::iterator i
= _children
.begin();
531 i
!= _children
.end();
533 pchildren
->insert(spg_t(*i
, shard
));
538 bool is_no_shard() const {
539 return shard
== shard_id_t::NO_SHARD
;
542 ghobject_t
make_pgmeta_oid() const {
543 return ghobject_t::make_pgmeta(pgid
.pool(), pgid
.ps(), shard
);
546 void encode(bufferlist
&bl
) const {
547 ENCODE_START(1, 1, bl
);
552 void decode(bufferlist::iterator
&bl
) {
559 ghobject_t
make_temp_ghobject(const string
& name
) const {
561 hobject_t(object_t(name
), "", CEPH_NOSNAP
,
563 hobject_t::POOL_TEMP_START
- pgid
.pool(), ""),
568 unsigned hash_to_shard(unsigned num_shards
) const {
569 return ps() % num_shards
;
572 WRITE_CLASS_ENCODER(spg_t
)
573 WRITE_EQ_OPERATORS_2(spg_t
, pgid
, shard
)
574 WRITE_CMP_OPERATORS_2(spg_t
, pgid
, shard
)
577 template<> struct hash
< spg_t
>
579 size_t operator()( const spg_t
& x
) const
581 static hash
<uint32_t> H
;
582 return H(hash
<pg_t
>()(x
.pgid
) ^ x
.shard
);
587 ostream
& operator<<(ostream
& out
, const spg_t
&pg
);
589 // ----------------------
594 TYPE_LEGACY_TEMP
= 1, /* no longer used */
600 uint64_t removal_seq
; // note: deprecated, not encoded
602 char _str_buff
[spg_t::calc_name_buf_size
];
607 coll_t(type_t t
, spg_t p
, uint64_t r
)
608 : type(t
), pgid(p
), removal_seq(r
) {
613 coll_t() : type(TYPE_META
), removal_seq(0)
618 coll_t(const coll_t
& other
)
619 : type(other
.type
), pgid(other
.pgid
), removal_seq(other
.removal_seq
) {
623 explicit coll_t(spg_t pgid
)
624 : type(TYPE_PG
), pgid(pgid
), removal_seq(0)
629 coll_t
& operator=(const coll_t
& rhs
)
631 this->type
= rhs
.type
;
632 this->pgid
= rhs
.pgid
;
633 this->removal_seq
= rhs
.removal_seq
;
638 // named constructors
639 static coll_t
meta() {
642 static coll_t
pg(spg_t p
) {
646 const std::string
to_str() const {
649 const char *c_str() const {
653 bool parse(const std::string
& s
);
655 int operator<(const coll_t
&rhs
) const {
656 return type
< rhs
.type
||
657 (type
== rhs
.type
&& pgid
< rhs
.pgid
);
660 bool is_meta() const {
661 return type
== TYPE_META
;
663 bool is_pg_prefix(spg_t
*pgid_
) const {
664 if (type
== TYPE_PG
|| type
== TYPE_PG_TEMP
) {
671 return type
== TYPE_PG
;
673 bool is_pg(spg_t
*pgid_
) const {
674 if (type
== TYPE_PG
) {
680 bool is_temp() const {
681 return type
== TYPE_PG_TEMP
;
683 bool is_temp(spg_t
*pgid_
) const {
684 if (type
== TYPE_PG_TEMP
) {
691 void encode(bufferlist
& bl
) const;
692 void decode(bufferlist::iterator
& bl
);
693 size_t encoded_size() const;
695 inline bool operator==(const coll_t
& rhs
) const {
696 // only compare type if meta
697 if (type
!= rhs
.type
)
699 if (type
== TYPE_META
)
701 return type
== rhs
.type
&& pgid
== rhs
.pgid
;
703 inline bool operator!=(const coll_t
& rhs
) const {
704 return !(*this == rhs
);
707 // get a TEMP collection that corresponds to the current collection,
708 // which we presume is a pg collection.
709 coll_t
get_temp() const {
710 assert(type
== TYPE_PG
);
711 return coll_t(TYPE_PG_TEMP
, pgid
, 0);
714 ghobject_t
get_min_hobj() const {
718 o
.hobj
.pool
= pgid
.pool();
719 o
.set_shard(pgid
.shard
);
730 unsigned hash_to_shard(unsigned num_shards
) const {
732 return pgid
.hash_to_shard(num_shards
);
733 return 0; // whatever.
736 void dump(Formatter
*f
) const;
737 static void generate_test_instances(list
<coll_t
*>& o
);
740 WRITE_CLASS_ENCODER(coll_t
)
742 inline ostream
& operator<<(ostream
& out
, const coll_t
& c
) {
748 template<> struct hash
<coll_t
> {
749 size_t operator()(const coll_t
&c
) const {
751 string
str(c
.to_str());
752 std::string::const_iterator
end(str
.end());
753 for (std::string::const_iterator s
= str
.begin(); s
!= end
; ++s
) {
766 inline ostream
& operator<<(ostream
& out
, const ceph_object_layout
&ol
)
768 out
<< pg_t(ol
.ol_pgid
);
769 int su
= ol
.ol_stripe_unit
;
777 // compound rados version type
778 /* WARNING: If add member in eversion_t, please make sure the encode/decode function
779 * work well. For little-endian machine, we should make sure there is no padding
780 * in 32-bit machine and 64-bit machine.
787 eversion_t() : version(0), epoch(0), __pad(0) {}
788 eversion_t(epoch_t e
, version_t v
) : version(v
), epoch(e
), __pad(0) {}
790 // cppcheck-suppress noExplicitConstructor
791 eversion_t(const ceph_eversion
& ce
) :
796 explicit eversion_t(bufferlist
& bl
) : __pad(0) { decode(bl
); }
798 static eversion_t
max() {
805 operator ceph_eversion() {
812 string
get_key_name() const;
814 void encode(bufferlist
&bl
) const {
815 #if defined(CEPH_LITTLE_ENDIAN)
816 bl
.append((char *)this, sizeof(version_t
) + sizeof(epoch_t
));
818 ::encode(version
, bl
);
822 void decode(bufferlist::iterator
&bl
) {
823 #if defined(CEPH_LITTLE_ENDIAN)
824 bl
.copy(sizeof(version_t
) + sizeof(epoch_t
), (char *)this);
826 ::decode(version
, bl
);
830 void decode(bufferlist
& bl
) {
831 bufferlist::iterator p
= bl
.begin();
835 WRITE_CLASS_ENCODER(eversion_t
)
837 inline bool operator==(const eversion_t
& l
, const eversion_t
& r
) {
838 return (l
.epoch
== r
.epoch
) && (l
.version
== r
.version
);
840 inline bool operator!=(const eversion_t
& l
, const eversion_t
& r
) {
841 return (l
.epoch
!= r
.epoch
) || (l
.version
!= r
.version
);
843 inline bool operator<(const eversion_t
& l
, const eversion_t
& r
) {
844 return (l
.epoch
== r
.epoch
) ? (l
.version
< r
.version
):(l
.epoch
< r
.epoch
);
846 inline bool operator<=(const eversion_t
& l
, const eversion_t
& r
) {
847 return (l
.epoch
== r
.epoch
) ? (l
.version
<= r
.version
):(l
.epoch
<= r
.epoch
);
849 inline bool operator>(const eversion_t
& l
, const eversion_t
& r
) {
850 return (l
.epoch
== r
.epoch
) ? (l
.version
> r
.version
):(l
.epoch
> r
.epoch
);
852 inline bool operator>=(const eversion_t
& l
, const eversion_t
& r
) {
853 return (l
.epoch
== r
.epoch
) ? (l
.version
>= r
.version
):(l
.epoch
>= r
.epoch
);
855 inline ostream
& operator<<(ostream
& out
, const eversion_t
& e
) {
856 return out
<< e
.epoch
<< "'" << e
.version
;
860 * objectstore_perf_stat_t
862 * current perf information about the osd
864 struct objectstore_perf_stat_t
{
865 // cur_op_latency is in ms since double add/sub are not associative
866 uint32_t os_commit_latency
;
867 uint32_t os_apply_latency
;
869 objectstore_perf_stat_t() :
870 os_commit_latency(0), os_apply_latency(0) {}
872 bool operator==(const objectstore_perf_stat_t
&r
) const {
873 return os_commit_latency
== r
.os_commit_latency
&&
874 os_apply_latency
== r
.os_apply_latency
;
877 void add(const objectstore_perf_stat_t
&o
) {
878 os_commit_latency
+= o
.os_commit_latency
;
879 os_apply_latency
+= o
.os_apply_latency
;
881 void sub(const objectstore_perf_stat_t
&o
) {
882 os_commit_latency
-= o
.os_commit_latency
;
883 os_apply_latency
-= o
.os_apply_latency
;
885 void dump(Formatter
*f
) const;
886 void encode(bufferlist
&bl
) const;
887 void decode(bufferlist::iterator
&bl
);
888 static void generate_test_instances(std::list
<objectstore_perf_stat_t
*>& o
);
890 WRITE_CLASS_ENCODER(objectstore_perf_stat_t
)
893 * aggregate stats for an osd
896 int64_t kb
, kb_used
, kb_avail
;
897 vector
<int> hb_peers
;
898 int32_t snap_trim_queue_len
, num_snap_trimming
;
900 pow2_hist_t op_queue_age_hist
;
902 objectstore_perf_stat_t os_perf_stat
;
907 uint32_t num_pgs
= 0;
909 osd_stat_t() : kb(0), kb_used(0), kb_avail(0),
910 snap_trim_queue_len(0), num_snap_trimming(0) {}
912 void add(const osd_stat_t
& o
) {
914 kb_used
+= o
.kb_used
;
915 kb_avail
+= o
.kb_avail
;
916 snap_trim_queue_len
+= o
.snap_trim_queue_len
;
917 num_snap_trimming
+= o
.num_snap_trimming
;
918 op_queue_age_hist
.add(o
.op_queue_age_hist
);
919 os_perf_stat
.add(o
.os_perf_stat
);
920 num_pgs
+= o
.num_pgs
;
922 void sub(const osd_stat_t
& o
) {
924 kb_used
-= o
.kb_used
;
925 kb_avail
-= o
.kb_avail
;
926 snap_trim_queue_len
-= o
.snap_trim_queue_len
;
927 num_snap_trimming
-= o
.num_snap_trimming
;
928 op_queue_age_hist
.sub(o
.op_queue_age_hist
);
929 os_perf_stat
.sub(o
.os_perf_stat
);
930 num_pgs
-= o
.num_pgs
;
933 void dump(Formatter
*f
) const;
934 void encode(bufferlist
&bl
) const;
935 void decode(bufferlist::iterator
&bl
);
936 static void generate_test_instances(std::list
<osd_stat_t
*>& o
);
938 WRITE_CLASS_ENCODER(osd_stat_t
)
940 inline bool operator==(const osd_stat_t
& l
, const osd_stat_t
& r
) {
941 return l
.kb
== r
.kb
&&
942 l
.kb_used
== r
.kb_used
&&
943 l
.kb_avail
== r
.kb_avail
&&
944 l
.snap_trim_queue_len
== r
.snap_trim_queue_len
&&
945 l
.num_snap_trimming
== r
.num_snap_trimming
&&
946 l
.hb_peers
== r
.hb_peers
&&
947 l
.op_queue_age_hist
== r
.op_queue_age_hist
&&
948 l
.os_perf_stat
== r
.os_perf_stat
&&
949 l
.num_pgs
== r
.num_pgs
;
951 inline bool operator!=(const osd_stat_t
& l
, const osd_stat_t
& r
) {
957 inline ostream
& operator<<(ostream
& out
, const osd_stat_t
& s
) {
958 return out
<< "osd_stat(" << kb_t(s
.kb_used
) << " used, "
959 << kb_t(s
.kb_avail
) << " avail, "
960 << kb_t(s
.kb
) << " total, "
961 << "peers " << s
.hb_peers
962 << " op hist " << s
.op_queue_age_hist
.h
970 #define PG_STATE_CREATING (1<<0) // creating
971 #define PG_STATE_ACTIVE (1<<1) // i am active. (primary: replicas too)
972 #define PG_STATE_CLEAN (1<<2) // peers are complete, clean of stray replicas.
973 #define PG_STATE_DOWN (1<<4) // a needed replica is down, PG offline
974 //#define PG_STATE_REPLAY (1<<5) // crashed, waiting for replay
975 //#define PG_STATE_STRAY (1<<6) // i must notify the primary i exist.
976 //#define PG_STATE_SPLITTING (1<<7) // i am splitting
977 #define PG_STATE_SCRUBBING (1<<8) // scrubbing
978 //#define PG_STATE_SCRUBQ (1<<9) // queued for scrub
979 #define PG_STATE_DEGRADED (1<<10) // pg contains objects with reduced redundancy
980 #define PG_STATE_INCONSISTENT (1<<11) // pg replicas are inconsistent (but shouldn't be)
981 #define PG_STATE_PEERING (1<<12) // pg is (re)peering
982 #define PG_STATE_REPAIR (1<<13) // pg should repair on next scrub
983 #define PG_STATE_RECOVERING (1<<14) // pg is recovering/migrating objects
984 #define PG_STATE_BACKFILL_WAIT (1<<15) // [active] reserving backfill
985 #define PG_STATE_INCOMPLETE (1<<16) // incomplete content, peering failed.
986 #define PG_STATE_STALE (1<<17) // our state for this pg is stale, unknown.
987 #define PG_STATE_REMAPPED (1<<18) // pg is explicitly remapped to different OSDs than CRUSH
988 #define PG_STATE_DEEP_SCRUB (1<<19) // deep scrub: check CRC32 on files
989 #define PG_STATE_BACKFILL (1<<20) // [active] backfilling pg content
990 #define PG_STATE_BACKFILL_TOOFULL (1<<21) // backfill can't proceed: too full
991 #define PG_STATE_RECOVERY_WAIT (1<<22) // waiting for recovery reservations
992 #define PG_STATE_UNDERSIZED (1<<23) // pg acting < pool size
993 #define PG_STATE_ACTIVATING (1<<24) // pg is peered but not yet active
994 #define PG_STATE_PEERED (1<<25) // peered, cannot go active, can recover
995 #define PG_STATE_SNAPTRIM (1<<26) // trimming snaps
996 #define PG_STATE_SNAPTRIM_WAIT (1<<27) // queued to trim snaps
997 #define PG_STATE_RECOVERY_TOOFULL (1<<28) // recovery can't proceed: too full
998 #define PG_STATE_SNAPTRIM_ERROR (1<<29) // error stopped trimming snaps
999 #define PG_STATE_FORCED_RECOVERY (1<<30) // force recovery of this pg before any other
1000 #define PG_STATE_FORCED_BACKFILL (1<<31) // force backfill of this pg before any other
1002 std::string
pg_state_string(int state
);
1003 std::string
pg_vector_string(const vector
<int32_t> &a
);
1004 int pg_string_state(const std::string
& state
);
1010 * attributes for a single pool snapshot.
1012 struct pool_snap_info_t
{
1017 void dump(Formatter
*f
) const;
1018 void encode(bufferlist
& bl
, uint64_t features
) const;
1019 void decode(bufferlist::iterator
& bl
);
1020 static void generate_test_instances(list
<pool_snap_info_t
*>& o
);
1022 WRITE_CLASS_ENCODER_FEATURES(pool_snap_info_t
)
1024 inline ostream
& operator<<(ostream
& out
, const pool_snap_info_t
& si
) {
1025 return out
<< si
.snapid
<< '(' << si
.name
<< ' ' << si
.stamp
<< ')';
1040 DEEP_SCRUB_INTERVAL
,
1042 RECOVERY_OP_PRIORITY
,
1045 COMPRESSION_ALGORITHM
,
1046 COMPRESSION_REQUIRED_RATIO
,
1047 COMPRESSION_MAX_BLOB_SIZE
,
1048 COMPRESSION_MIN_BLOB_SIZE
,
1064 opt_desc_t(key_t k
, type_t t
) : key(k
), type(t
) {}
1066 bool operator==(const opt_desc_t
& rhs
) const {
1067 return key
== rhs
.key
&& type
== rhs
.type
;
1071 typedef boost::variant
<std::string
,int,double> value_t
;
1073 static bool is_opt_name(const std::string
& name
);
1074 static opt_desc_t
get_opt_desc(const std::string
& name
);
1076 pool_opts_t() : opts() {}
1078 bool is_set(key_t key
) const;
1080 template<typename T
>
1081 void set(key_t key
, const T
&val
) {
1082 value_t value
= val
;
1086 template<typename T
>
1087 bool get(key_t key
, T
*val
) const {
1088 opts_t::const_iterator i
= opts
.find(key
);
1089 if (i
== opts
.end()) {
1092 *val
= boost::get
<T
>(i
->second
);
1096 const value_t
& get(key_t key
) const;
1098 bool unset(key_t key
);
1100 void dump(const std::string
& name
, Formatter
*f
) const;
1102 void dump(Formatter
*f
) const;
1103 void encode(bufferlist
&bl
) const;
1104 void decode(bufferlist::iterator
&bl
);
1107 typedef std::map
<key_t
, value_t
> opts_t
;
1110 friend ostream
& operator<<(ostream
& out
, const pool_opts_t
& opts
);
1112 WRITE_CLASS_ENCODER(pool_opts_t
)
1118 static const char *APPLICATION_NAME_CEPHFS
;
1119 static const char *APPLICATION_NAME_RBD
;
1120 static const char *APPLICATION_NAME_RGW
;
1123 TYPE_REPLICATED
= 1, // replication
1124 //TYPE_RAID4 = 2, // raid4 (never implemented)
1125 TYPE_ERASURE
= 3, // erasure-coded
1127 static const char *get_type_name(int t
) {
1129 case TYPE_REPLICATED
: return "replicated";
1130 //case TYPE_RAID4: return "raid4";
1131 case TYPE_ERASURE
: return "erasure";
1132 default: return "???";
1135 const char *get_type_name() const {
1136 return get_type_name(type
);
1140 FLAG_HASHPSPOOL
= 1<<0, // hash pg seed and pool together (instead of adding)
1141 FLAG_FULL
= 1<<1, // pool is full
1142 FLAG_EC_OVERWRITES
= 1<<2, // enables overwrites, once enabled, cannot be disabled
1143 FLAG_INCOMPLETE_CLONES
= 1<<3, // may have incomplete clones (bc we are/were an overlay)
1144 FLAG_NODELETE
= 1<<4, // pool can't be deleted
1145 FLAG_NOPGCHANGE
= 1<<5, // pool's pg and pgp num can't be changed
1146 FLAG_NOSIZECHANGE
= 1<<6, // pool's size and min size can't be changed
1147 FLAG_WRITE_FADVISE_DONTNEED
= 1<<7, // write mode with LIBRADOS_OP_FLAG_FADVISE_DONTNEED
1148 FLAG_NOSCRUB
= 1<<8, // block periodic scrub
1149 FLAG_NODEEP_SCRUB
= 1<<9, // block periodic deep-scrub
1152 static const char *get_flag_name(int f
) {
1154 case FLAG_HASHPSPOOL
: return "hashpspool";
1155 case FLAG_FULL
: return "full";
1156 case FLAG_EC_OVERWRITES
: return "ec_overwrites";
1157 case FLAG_INCOMPLETE_CLONES
: return "incomplete_clones";
1158 case FLAG_NODELETE
: return "nodelete";
1159 case FLAG_NOPGCHANGE
: return "nopgchange";
1160 case FLAG_NOSIZECHANGE
: return "nosizechange";
1161 case FLAG_WRITE_FADVISE_DONTNEED
: return "write_fadvise_dontneed";
1162 case FLAG_NOSCRUB
: return "noscrub";
1163 case FLAG_NODEEP_SCRUB
: return "nodeep-scrub";
1164 default: return "???";
1167 static string
get_flags_string(uint64_t f
) {
1169 for (unsigned n
=0; f
&& n
<64; ++n
) {
1170 if (f
& (1ull << n
)) {
1173 s
+= get_flag_name(1ull << n
);
1178 string
get_flags_string() const {
1179 return get_flags_string(flags
);
1181 static uint64_t get_flag_by_name(const string
& name
) {
1182 if (name
== "hashpspool")
1183 return FLAG_HASHPSPOOL
;
1186 if (name
== "ec_overwrites")
1187 return FLAG_EC_OVERWRITES
;
1188 if (name
== "incomplete_clones")
1189 return FLAG_INCOMPLETE_CLONES
;
1190 if (name
== "nodelete")
1191 return FLAG_NODELETE
;
1192 if (name
== "nopgchange")
1193 return FLAG_NOPGCHANGE
;
1194 if (name
== "nosizechange")
1195 return FLAG_NOSIZECHANGE
;
1196 if (name
== "write_fadvise_dontneed")
1197 return FLAG_WRITE_FADVISE_DONTNEED
;
1198 if (name
== "noscrub")
1199 return FLAG_NOSCRUB
;
1200 if (name
== "nodeep-scrub")
1201 return FLAG_NODEEP_SCRUB
;
1205 /// converts the acting/up vector to a set of pg shards
1206 void convert_to_pg_shards(const vector
<int> &from
, set
<pg_shard_t
>* to
) const;
1209 CACHEMODE_NONE
= 0, ///< no caching
1210 CACHEMODE_WRITEBACK
= 1, ///< write to cache, flush later
1211 CACHEMODE_FORWARD
= 2, ///< forward if not in cache
1212 CACHEMODE_READONLY
= 3, ///< handle reads, forward writes [not strongly consistent]
1213 CACHEMODE_READFORWARD
= 4, ///< forward reads, write to cache flush later
1214 CACHEMODE_READPROXY
= 5, ///< proxy reads, write to cache flush later
1215 CACHEMODE_PROXY
= 6, ///< proxy if not in cache
1217 static const char *get_cache_mode_name(cache_mode_t m
) {
1219 case CACHEMODE_NONE
: return "none";
1220 case CACHEMODE_WRITEBACK
: return "writeback";
1221 case CACHEMODE_FORWARD
: return "forward";
1222 case CACHEMODE_READONLY
: return "readonly";
1223 case CACHEMODE_READFORWARD
: return "readforward";
1224 case CACHEMODE_READPROXY
: return "readproxy";
1225 case CACHEMODE_PROXY
: return "proxy";
1226 default: return "unknown";
1229 static cache_mode_t
get_cache_mode_from_str(const string
& s
) {
1231 return CACHEMODE_NONE
;
1232 if (s
== "writeback")
1233 return CACHEMODE_WRITEBACK
;
1235 return CACHEMODE_FORWARD
;
1236 if (s
== "readonly")
1237 return CACHEMODE_READONLY
;
1238 if (s
== "readforward")
1239 return CACHEMODE_READFORWARD
;
1240 if (s
== "readproxy")
1241 return CACHEMODE_READPROXY
;
1243 return CACHEMODE_PROXY
;
1244 return (cache_mode_t
)-1;
1246 const char *get_cache_mode_name() const {
1247 return get_cache_mode_name(cache_mode
);
1249 bool cache_mode_requires_hit_set() const {
1250 switch (cache_mode
) {
1251 case CACHEMODE_NONE
:
1252 case CACHEMODE_FORWARD
:
1253 case CACHEMODE_READONLY
:
1254 case CACHEMODE_PROXY
:
1256 case CACHEMODE_WRITEBACK
:
1257 case CACHEMODE_READFORWARD
:
1258 case CACHEMODE_READPROXY
:
1261 assert(0 == "implement me");
1265 uint64_t flags
; ///< FLAG_*
1266 __u8 type
; ///< TYPE_*
1267 __u8 size
, min_size
; ///< number of osds in each pg
1268 __u8 crush_rule
; ///< crush placement rule
1269 __u8 object_hash
; ///< hash mapping object name to ps
1271 __u32 pg_num
, pgp_num
; ///< number of pgs
1275 map
<string
,string
> properties
; ///< OBSOLETE
1276 string erasure_code_profile
; ///< name of the erasure code profile in OSDMap
1277 epoch_t last_change
; ///< most recent epoch changed, exclusing snapshot changes
1278 epoch_t last_force_op_resend
; ///< last epoch that forced clients to resend
1279 /// last epoch that forced clients to resend (pre-luminous clients only)
1280 epoch_t last_force_op_resend_preluminous
;
1281 snapid_t snap_seq
; ///< seq for per-pool snapshot
1282 epoch_t snap_epoch
; ///< osdmap epoch of last snap
1283 uint64_t auid
; ///< who owns the pg
1284 __u32 crash_replay_interval
; ///< seconds to allow clients to replay ACKed but unCOMMITted requests
1286 uint64_t quota_max_bytes
; ///< maximum number of bytes for this pool
1287 uint64_t quota_max_objects
; ///< maximum number of objects for this pool
1290 * Pool snaps (global to this pool). These define a SnapContext for
1291 * the pool, unless the client manually specifies an alternate
1294 map
<snapid_t
, pool_snap_info_t
> snaps
;
1296 * Alternatively, if we are defining non-pool snaps (e.g. via the
1297 * Ceph MDS), we must track @removed_snaps (since @snaps is not
1298 * used). Snaps and removed_snaps are to be used exclusive of each
1301 interval_set
<snapid_t
> removed_snaps
;
1303 unsigned pg_num_mask
, pgp_num_mask
;
1305 set
<uint64_t> tiers
; ///< pools that are tiers of us
1306 int64_t tier_of
; ///< pool for which we are a tier
1307 // Note that write wins for read+write ops
1308 int64_t read_tier
; ///< pool/tier for objecter to direct reads to
1309 int64_t write_tier
; ///< pool/tier for objecter to direct writes to
1310 cache_mode_t cache_mode
; ///< cache pool mode
1312 bool is_tier() const { return tier_of
>= 0; }
1313 bool has_tiers() const { return !tiers
.empty(); }
1318 clear_tier_tunables();
1320 bool has_read_tier() const { return read_tier
>= 0; }
1321 void clear_read_tier() { read_tier
= -1; }
1322 bool has_write_tier() const { return write_tier
>= 0; }
1323 void clear_write_tier() { write_tier
= -1; }
1324 void clear_tier_tunables() {
1325 if (cache_mode
!= CACHEMODE_NONE
)
1326 flags
|= FLAG_INCOMPLETE_CLONES
;
1327 cache_mode
= CACHEMODE_NONE
;
1329 target_max_bytes
= 0;
1330 target_max_objects
= 0;
1331 cache_target_dirty_ratio_micro
= 0;
1332 cache_target_dirty_high_ratio_micro
= 0;
1333 cache_target_full_ratio_micro
= 0;
1334 hit_set_params
= HitSet::Params();
1337 hit_set_grade_decay_rate
= 0;
1338 hit_set_search_last_n
= 0;
1339 grade_table
.resize(0);
1342 uint64_t target_max_bytes
; ///< tiering: target max pool size
1343 uint64_t target_max_objects
; ///< tiering: target max pool size
1345 uint32_t cache_target_dirty_ratio_micro
; ///< cache: fraction of target to leave dirty
1346 uint32_t cache_target_dirty_high_ratio_micro
; ///<cache: fraction of target to flush with high speed
1347 uint32_t cache_target_full_ratio_micro
; ///< cache: fraction of target to fill before we evict in earnest
1349 uint32_t cache_min_flush_age
; ///< minimum age (seconds) before we can flush
1350 uint32_t cache_min_evict_age
; ///< minimum age (seconds) before we can evict
1352 HitSet::Params hit_set_params
; ///< The HitSet params to use on this pool
1353 uint32_t hit_set_period
; ///< periodicity of HitSet segments (seconds)
1354 uint32_t hit_set_count
; ///< number of periods to retain
1355 bool use_gmt_hitset
; ///< use gmt to name the hitset archive object
1356 uint32_t min_read_recency_for_promote
; ///< minimum number of HitSet to check before promote on read
1357 uint32_t min_write_recency_for_promote
; ///< minimum number of HitSet to check before promote on write
1358 uint32_t hit_set_grade_decay_rate
; ///< current hit_set has highest priority on objects
1359 ///temperature count,the follow hit_set's priority decay
1360 ///by this params than pre hit_set
1361 uint32_t hit_set_search_last_n
; ///<accumulate atmost N hit_sets for temperature
1363 uint32_t stripe_width
; ///< erasure coded stripe size in bytes
1365 uint64_t expected_num_objects
; ///< expected number of objects on this pool, a value of 0 indicates
1366 ///< user does not specify any expected value
1367 bool fast_read
; ///< whether turn on fast read on the pool or not
1369 pool_opts_t opts
; ///< options
1371 /// application -> key/value metadata
1372 map
<string
, std::map
<string
, string
>> application_metadata
;
1375 vector
<uint32_t> grade_table
;
1378 uint32_t get_grade(unsigned i
) const {
1379 if (grade_table
.size() <= i
)
1381 return grade_table
[i
];
1383 void calc_grade_table() {
1384 unsigned v
= 1000000;
1385 grade_table
.resize(hit_set_count
);
1386 for (unsigned i
= 0; i
< hit_set_count
; i
++) {
1387 v
= v
* (1 - (hit_set_grade_decay_rate
/ 100.0));
1393 : flags(0), type(0), size(0), min_size(0),
1394 crush_rule(0), object_hash(0),
1395 pg_num(0), pgp_num(0),
1397 last_force_op_resend(0),
1398 last_force_op_resend_preluminous(0),
1399 snap_seq(0), snap_epoch(0),
1401 crash_replay_interval(0),
1402 quota_max_bytes(0), quota_max_objects(0),
1403 pg_num_mask(0), pgp_num_mask(0),
1404 tier_of(-1), read_tier(-1), write_tier(-1),
1405 cache_mode(CACHEMODE_NONE
),
1406 target_max_bytes(0), target_max_objects(0),
1407 cache_target_dirty_ratio_micro(0),
1408 cache_target_dirty_high_ratio_micro(0),
1409 cache_target_full_ratio_micro(0),
1410 cache_min_flush_age(0),
1411 cache_min_evict_age(0),
1415 use_gmt_hitset(true),
1416 min_read_recency_for_promote(0),
1417 min_write_recency_for_promote(0),
1418 hit_set_grade_decay_rate(0),
1419 hit_set_search_last_n(0),
1421 expected_num_objects(0),
1426 void dump(Formatter
*f
) const;
1428 uint64_t get_flags() const { return flags
; }
1429 bool has_flag(uint64_t f
) const { return flags
& f
; }
1430 void set_flag(uint64_t f
) { flags
|= f
; }
1431 void unset_flag(uint64_t f
) { flags
&= ~f
; }
1433 bool ec_pool() const {
1434 return type
== TYPE_ERASURE
;
1436 bool require_rollback() const {
1440 /// true if incomplete clones may be present
1441 bool allow_incomplete_clones() const {
1442 return cache_mode
!= CACHEMODE_NONE
|| has_flag(FLAG_INCOMPLETE_CLONES
);
1445 unsigned get_type() const { return type
; }
1446 unsigned get_size() const { return size
; }
1447 unsigned get_min_size() const { return min_size
; }
1448 int get_crush_rule() const { return crush_rule
; }
1449 int get_object_hash() const { return object_hash
; }
1450 const char *get_object_hash_name() const {
1451 return ceph_str_hash_name(get_object_hash());
1453 epoch_t
get_last_change() const { return last_change
; }
1454 epoch_t
get_last_force_op_resend() const { return last_force_op_resend
; }
1455 epoch_t
get_last_force_op_resend_preluminous() const {
1456 return last_force_op_resend_preluminous
;
1458 epoch_t
get_snap_epoch() const { return snap_epoch
; }
1459 snapid_t
get_snap_seq() const { return snap_seq
; }
1460 uint64_t get_auid() const { return auid
; }
1461 unsigned get_crash_replay_interval() const { return crash_replay_interval
; }
1463 void set_snap_seq(snapid_t s
) { snap_seq
= s
; }
1464 void set_snap_epoch(epoch_t e
) { snap_epoch
= e
; }
1466 void set_stripe_width(uint32_t s
) { stripe_width
= s
; }
1467 uint32_t get_stripe_width() const { return stripe_width
; }
1469 bool is_replicated() const { return get_type() == TYPE_REPLICATED
; }
1470 bool is_erasure() const { return get_type() == TYPE_ERASURE
; }
1472 bool supports_omap() const {
1473 return !(get_type() == TYPE_ERASURE
);
1476 bool requires_aligned_append() const {
1477 return is_erasure() && !has_flag(FLAG_EC_OVERWRITES
);
1479 uint64_t required_alignment() const { return stripe_width
; }
1481 bool allows_ecoverwrites() const {
1482 return has_flag(FLAG_EC_OVERWRITES
);
1485 bool can_shift_osds() const {
1486 switch (get_type()) {
1487 case TYPE_REPLICATED
:
1492 assert(0 == "unhandled pool type");
1496 unsigned get_pg_num() const { return pg_num
; }
1497 unsigned get_pgp_num() const { return pgp_num
; }
1499 unsigned get_pg_num_mask() const { return pg_num_mask
; }
1500 unsigned get_pgp_num_mask() const { return pgp_num_mask
; }
1502 // if pg_num is not a multiple of two, pgs are not equally sized.
1503 // return, for a given pg, the fraction (denominator) of the total
1504 // pool size that it represents.
1505 unsigned get_pg_num_divisor(pg_t pgid
) const;
1507 void set_pg_num(int p
) {
1511 void set_pgp_num(int p
) {
1516 void set_quota_max_bytes(uint64_t m
) {
1517 quota_max_bytes
= m
;
1519 uint64_t get_quota_max_bytes() {
1520 return quota_max_bytes
;
1523 void set_quota_max_objects(uint64_t m
) {
1524 quota_max_objects
= m
;
1526 uint64_t get_quota_max_objects() {
1527 return quota_max_objects
;
1530 void set_last_force_op_resend(uint64_t t
) {
1531 last_force_op_resend
= t
;
1532 last_force_op_resend_preluminous
= t
;
1535 void calc_pg_masks();
1538 * we have two snap modes:
1539 * - pool global snaps
1540 * - snap existence/non-existence defined by snaps[] and snap_seq
1541 * - user managed snaps
1542 * - removal governed by removed_snaps
1544 * we know which mode we're using based on whether removed_snaps is empty.
1545 * If nothing has been created, both functions report false.
1547 bool is_pool_snaps_mode() const;
1548 bool is_unmanaged_snaps_mode() const;
1549 bool is_removed_snap(snapid_t s
) const;
1552 * build set of known-removed sets from either pool snaps or
1553 * explicit removed_snaps set.
1555 void build_removed_snaps(interval_set
<snapid_t
>& rs
) const;
1556 snapid_t
snap_exists(const char *s
) const;
1557 void add_snap(const char *n
, utime_t stamp
);
1558 void add_unmanaged_snap(uint64_t& snapid
);
1559 void remove_snap(snapid_t s
);
1560 void remove_unmanaged_snap(snapid_t s
);
1562 SnapContext
get_snap_context() const;
1564 /// hash a object name+namespace key to a hash position
1565 uint32_t hash_key(const string
& key
, const string
& ns
) const;
1567 /// round a hash position down to a pg num
1568 uint32_t raw_hash_to_pg(uint32_t v
) const;
1571 * map a raw pg (with full precision ps) into an actual pg, for storage
1573 pg_t
raw_pg_to_pg(pg_t pg
) const;
1576 * map raw pg (full precision ps) into a placement seed. include
1577 * pool id in that value so that different pools don't use the same
1580 ps_t
raw_pg_to_pps(pg_t pg
) const;
1582 /// choose a random hash position within a pg
1583 uint32_t get_random_pg_position(pg_t pgid
, uint32_t seed
) const;
1585 void encode(bufferlist
& bl
, uint64_t features
) const;
1586 void decode(bufferlist::iterator
& bl
);
1588 static void generate_test_instances(list
<pg_pool_t
*>& o
);
1590 WRITE_CLASS_ENCODER_FEATURES(pg_pool_t
)
1592 ostream
& operator<<(ostream
& out
, const pg_pool_t
& p
);
1596 * a summation of object stats
1598 * This is just a container for object stats; we don't know what for.
1600 * If you add members in object_stat_sum_t, you should make sure there are
1601 * not padding among these members.
1602 * You should also modify the padding_check function.
1605 struct object_stat_sum_t
{
1606 /**************************************************************************
1607 * WARNING: be sure to update operator==, floor, and split when
1608 * adding/removing fields!
1609 **************************************************************************/
1610 int64_t num_bytes
; // in bytes
1611 int64_t num_objects
;
1612 int64_t num_object_clones
;
1613 int64_t num_object_copies
; // num_objects * num_replicas
1614 int64_t num_objects_missing_on_primary
;
1615 int64_t num_objects_degraded
;
1616 int64_t num_objects_unfound
;
1621 int64_t num_scrub_errors
; // total deep and shallow scrub errors
1622 int64_t num_objects_recovered
;
1623 int64_t num_bytes_recovered
;
1624 int64_t num_keys_recovered
;
1625 int64_t num_shallow_scrub_errors
;
1626 int64_t num_deep_scrub_errors
;
1627 int64_t num_objects_dirty
;
1628 int64_t num_whiteouts
;
1629 int64_t num_objects_omap
;
1630 int64_t num_objects_hit_set_archive
;
1631 int64_t num_objects_misplaced
;
1632 int64_t num_bytes_hit_set_archive
;
1634 int64_t num_flush_kb
;
1636 int64_t num_evict_kb
;
1637 int64_t num_promote
;
1638 int32_t num_flush_mode_high
; // 1 when in high flush mode, otherwise 0
1639 int32_t num_flush_mode_low
; // 1 when in low flush mode, otherwise 0
1640 int32_t num_evict_mode_some
; // 1 when in evict some mode, otherwise 0
1641 int32_t num_evict_mode_full
; // 1 when in evict full mode, otherwise 0
1642 int64_t num_objects_pinned
;
1643 int64_t num_objects_missing
;
1644 int64_t num_legacy_snapsets
; ///< upper bound on pre-luminous-style SnapSets
1648 num_objects(0), num_object_clones(0), num_object_copies(0),
1649 num_objects_missing_on_primary(0), num_objects_degraded(0),
1650 num_objects_unfound(0),
1651 num_rd(0), num_rd_kb(0), num_wr(0), num_wr_kb(0),
1652 num_scrub_errors(0),
1653 num_objects_recovered(0),
1654 num_bytes_recovered(0),
1655 num_keys_recovered(0),
1656 num_shallow_scrub_errors(0),
1657 num_deep_scrub_errors(0),
1658 num_objects_dirty(0),
1660 num_objects_omap(0),
1661 num_objects_hit_set_archive(0),
1662 num_objects_misplaced(0),
1663 num_bytes_hit_set_archive(0),
1669 num_flush_mode_high(0), num_flush_mode_low(0),
1670 num_evict_mode_some(0), num_evict_mode_full(0),
1671 num_objects_pinned(0),
1672 num_objects_missing(0),
1673 num_legacy_snapsets(0)
1676 void floor(int64_t f
) {
1677 #define FLOOR(x) if (x < f) x = f
1680 FLOOR(num_object_clones
);
1681 FLOOR(num_object_copies
);
1682 FLOOR(num_objects_missing_on_primary
);
1683 FLOOR(num_objects_missing
);
1684 FLOOR(num_objects_degraded
);
1685 FLOOR(num_objects_misplaced
);
1686 FLOOR(num_objects_unfound
);
1691 FLOOR(num_scrub_errors
);
1692 FLOOR(num_shallow_scrub_errors
);
1693 FLOOR(num_deep_scrub_errors
);
1694 FLOOR(num_objects_recovered
);
1695 FLOOR(num_bytes_recovered
);
1696 FLOOR(num_keys_recovered
);
1697 FLOOR(num_objects_dirty
);
1698 FLOOR(num_whiteouts
);
1699 FLOOR(num_objects_omap
);
1700 FLOOR(num_objects_hit_set_archive
);
1701 FLOOR(num_bytes_hit_set_archive
);
1703 FLOOR(num_flush_kb
);
1705 FLOOR(num_evict_kb
);
1707 FLOOR(num_flush_mode_high
);
1708 FLOOR(num_flush_mode_low
);
1709 FLOOR(num_evict_mode_some
);
1710 FLOOR(num_evict_mode_full
);
1711 FLOOR(num_objects_pinned
);
1712 FLOOR(num_legacy_snapsets
);
1716 void split(vector
<object_stat_sum_t
> &out
) const {
1717 #define SPLIT(PARAM) \
1718 for (unsigned i = 0; i < out.size(); ++i) { \
1719 out[i].PARAM = PARAM / out.size(); \
1720 if (i < (PARAM % out.size())) { \
1724 #define SPLIT_PRESERVE_NONZERO(PARAM) \
1725 for (unsigned i = 0; i < out.size(); ++i) { \
1727 out[i].PARAM = 1 + PARAM / out.size(); \
1734 SPLIT(num_object_clones
);
1735 SPLIT(num_object_copies
);
1736 SPLIT(num_objects_missing_on_primary
);
1737 SPLIT(num_objects_missing
);
1738 SPLIT(num_objects_degraded
);
1739 SPLIT(num_objects_misplaced
);
1740 SPLIT(num_objects_unfound
);
1745 SPLIT(num_scrub_errors
);
1746 SPLIT(num_shallow_scrub_errors
);
1747 SPLIT(num_deep_scrub_errors
);
1748 SPLIT(num_objects_recovered
);
1749 SPLIT(num_bytes_recovered
);
1750 SPLIT(num_keys_recovered
);
1751 SPLIT(num_objects_dirty
);
1752 SPLIT(num_whiteouts
);
1753 SPLIT(num_objects_omap
);
1754 SPLIT(num_objects_hit_set_archive
);
1755 SPLIT(num_bytes_hit_set_archive
);
1757 SPLIT(num_flush_kb
);
1759 SPLIT(num_evict_kb
);
1761 SPLIT(num_flush_mode_high
);
1762 SPLIT(num_flush_mode_low
);
1763 SPLIT(num_evict_mode_some
);
1764 SPLIT(num_evict_mode_full
);
1765 SPLIT(num_objects_pinned
);
1766 SPLIT_PRESERVE_NONZERO(num_legacy_snapsets
);
1768 #undef SPLIT_PRESERVE_NONZERO
1772 memset(this, 0, sizeof(*this));
1775 void calc_copies(int nrep
) {
1776 num_object_copies
= nrep
* num_objects
;
1779 bool is_zero() const {
1780 return mem_is_zero((char*)this, sizeof(*this));
1783 void add(const object_stat_sum_t
& o
);
1784 void sub(const object_stat_sum_t
& o
);
1786 void dump(Formatter
*f
) const;
1787 void padding_check() {
1789 sizeof(object_stat_sum_t
) ==
1791 sizeof(num_objects
) +
1792 sizeof(num_object_clones
) +
1793 sizeof(num_object_copies
) +
1794 sizeof(num_objects_missing_on_primary
) +
1795 sizeof(num_objects_degraded
) +
1796 sizeof(num_objects_unfound
) +
1801 sizeof(num_scrub_errors
) +
1802 sizeof(num_objects_recovered
) +
1803 sizeof(num_bytes_recovered
) +
1804 sizeof(num_keys_recovered
) +
1805 sizeof(num_shallow_scrub_errors
) +
1806 sizeof(num_deep_scrub_errors
) +
1807 sizeof(num_objects_dirty
) +
1808 sizeof(num_whiteouts
) +
1809 sizeof(num_objects_omap
) +
1810 sizeof(num_objects_hit_set_archive
) +
1811 sizeof(num_objects_misplaced
) +
1812 sizeof(num_bytes_hit_set_archive
) +
1814 sizeof(num_flush_kb
) +
1816 sizeof(num_evict_kb
) +
1817 sizeof(num_promote
) +
1818 sizeof(num_flush_mode_high
) +
1819 sizeof(num_flush_mode_low
) +
1820 sizeof(num_evict_mode_some
) +
1821 sizeof(num_evict_mode_full
) +
1822 sizeof(num_objects_pinned
) +
1823 sizeof(num_objects_missing
) +
1824 sizeof(num_legacy_snapsets
)
1826 "object_stat_sum_t have padding");
1828 void encode(bufferlist
& bl
) const;
1829 void decode(bufferlist::iterator
& bl
);
1830 static void generate_test_instances(list
<object_stat_sum_t
*>& o
);
1832 WRITE_CLASS_ENCODER(object_stat_sum_t
)
1834 bool operator==(const object_stat_sum_t
& l
, const object_stat_sum_t
& r
);
1837 * a collection of object stat sums
1839 * This is a collection of stat sums over different categories.
1841 struct object_stat_collection_t
{
1842 /**************************************************************************
1843 * WARNING: be sure to update the operator== when adding/removing fields! *
1844 **************************************************************************/
1845 object_stat_sum_t sum
;
1847 void calc_copies(int nrep
) {
1848 sum
.calc_copies(nrep
);
1851 void dump(Formatter
*f
) const;
1852 void encode(bufferlist
& bl
) const;
1853 void decode(bufferlist::iterator
& bl
);
1854 static void generate_test_instances(list
<object_stat_collection_t
*>& o
);
1856 bool is_zero() const {
1857 return sum
.is_zero();
1864 void floor(int64_t f
) {
1868 void add(const object_stat_sum_t
& o
) {
1872 void add(const object_stat_collection_t
& o
) {
1875 void sub(const object_stat_collection_t
& o
) {
1879 WRITE_CLASS_ENCODER(object_stat_collection_t
)
1881 inline bool operator==(const object_stat_collection_t
& l
,
1882 const object_stat_collection_t
& r
) {
1883 return l
.sum
== r
.sum
;
1888 * aggregate stats for a single PG.
1891 /**************************************************************************
1892 * WARNING: be sure to update the operator== when adding/removing fields! *
1893 **************************************************************************/
1895 version_t reported_seq
; // sequence number
1896 epoch_t reported_epoch
; // epoch of this report
1898 utime_t last_fresh
; // last reported
1899 utime_t last_change
; // new state != previous state
1900 utime_t last_active
; // state & PG_STATE_ACTIVE
1901 utime_t last_peered
; // state & PG_STATE_ACTIVE || state & PG_STATE_PEERED
1902 utime_t last_clean
; // state & PG_STATE_CLEAN
1903 utime_t last_unstale
; // (state & PG_STATE_STALE) == 0
1904 utime_t last_undegraded
; // (state & PG_STATE_DEGRADED) == 0
1905 utime_t last_fullsized
; // (state & PG_STATE_UNDERSIZED) == 0
1907 eversion_t log_start
; // (log_start,version]
1908 eversion_t ondisk_log_start
; // there may be more on disk
1911 epoch_t last_epoch_clean
;
1913 __u32 parent_split_bits
;
1915 eversion_t last_scrub
;
1916 eversion_t last_deep_scrub
;
1917 utime_t last_scrub_stamp
;
1918 utime_t last_deep_scrub_stamp
;
1919 utime_t last_clean_scrub_stamp
;
1921 object_stat_collection_t stats
;
1924 int64_t ondisk_log_size
; // >= active_log_size
1926 vector
<int32_t> up
, acting
;
1927 epoch_t mapping_epoch
;
1929 vector
<int32_t> blocked_by
; ///< osds on which the pg is blocked
1931 utime_t last_became_active
;
1932 utime_t last_became_peered
;
1934 /// up, acting primaries
1936 int32_t acting_primary
;
1938 bool stats_invalid
:1;
1939 /// true if num_objects_dirty is not accurate (because it was not
1940 /// maintained starting from pool creation)
1941 bool dirty_stats_invalid
:1;
1942 bool omap_stats_invalid
:1;
1943 bool hitset_stats_invalid
:1;
1944 bool hitset_bytes_stats_invalid
:1;
1945 bool pin_stats_invalid
:1;
1951 created(0), last_epoch_clean(0),
1952 parent_split_bits(0),
1953 log_size(0), ondisk_log_size(0),
1957 stats_invalid(false),
1958 dirty_stats_invalid(false),
1959 omap_stats_invalid(false),
1960 hitset_stats_invalid(false),
1961 hitset_bytes_stats_invalid(false),
1962 pin_stats_invalid(false)
1965 epoch_t
get_effective_last_epoch_clean() const {
1966 if (state
& PG_STATE_CLEAN
) {
1967 // we are clean as of this report, and should thus take the
1969 return reported_epoch
;
1971 return last_epoch_clean
;
1975 pair
<epoch_t
, version_t
> get_version_pair() const {
1976 return make_pair(reported_epoch
, reported_seq
);
1979 void floor(int64_t f
) {
1983 if (ondisk_log_size
< f
)
1984 ondisk_log_size
= f
;
1987 void add(const pg_stat_t
& o
) {
1989 log_size
+= o
.log_size
;
1990 ondisk_log_size
+= o
.ondisk_log_size
;
1992 void sub(const pg_stat_t
& o
) {
1994 log_size
-= o
.log_size
;
1995 ondisk_log_size
-= o
.ondisk_log_size
;
1998 bool is_acting_osd(int32_t osd
, bool primary
) const;
1999 void dump(Formatter
*f
) const;
2000 void dump_brief(Formatter
*f
) const;
2001 void encode(bufferlist
&bl
) const;
2002 void decode(bufferlist::iterator
&bl
);
2003 static void generate_test_instances(list
<pg_stat_t
*>& o
);
2005 WRITE_CLASS_ENCODER(pg_stat_t
)
2007 bool operator==(const pg_stat_t
& l
, const pg_stat_t
& r
);
2010 * summation over an entire pool
2012 struct pool_stat_t
{
2013 object_stat_collection_t stats
;
2015 int64_t ondisk_log_size
; // >= active_log_size
2016 int32_t up
; ///< number of up replicas or shards
2017 int32_t acting
; ///< number of acting replicas or shards
2019 pool_stat_t() : log_size(0), ondisk_log_size(0), up(0), acting(0)
2022 void floor(int64_t f
) {
2026 if (ondisk_log_size
< f
)
2027 ondisk_log_size
= f
;
2034 void add(const pg_stat_t
& o
) {
2036 log_size
+= o
.log_size
;
2037 ondisk_log_size
+= o
.ondisk_log_size
;
2039 acting
+= o
.acting
.size();
2041 void sub(const pg_stat_t
& o
) {
2043 log_size
-= o
.log_size
;
2044 ondisk_log_size
-= o
.ondisk_log_size
;
2046 acting
-= o
.acting
.size();
2049 bool is_zero() const {
2050 return (stats
.is_zero() &&
2052 ondisk_log_size
== 0 &&
2057 void dump(Formatter
*f
) const;
2058 void encode(bufferlist
&bl
, uint64_t features
) const;
2059 void decode(bufferlist::iterator
&bl
);
2060 static void generate_test_instances(list
<pool_stat_t
*>& o
);
2062 WRITE_CLASS_ENCODER_FEATURES(pool_stat_t
)
2065 // -----------------------------------------
2068 * pg_hit_set_info_t - information about a single recorded HitSet
2070 * Track basic metadata about a HitSet, like the nubmer of insertions
2071 * and the time range it covers.
2073 struct pg_hit_set_info_t
{
2074 utime_t begin
, end
; ///< time interval
2075 eversion_t version
; ///< version this HitSet object was written
2076 bool using_gmt
; ///< use gmt for creating the hit_set archive object name
2078 friend bool operator==(const pg_hit_set_info_t
& l
,
2079 const pg_hit_set_info_t
& r
) {
2081 l
.begin
== r
.begin
&&
2083 l
.version
== r
.version
&&
2084 l
.using_gmt
== r
.using_gmt
;
2087 explicit pg_hit_set_info_t(bool using_gmt
= true)
2088 : using_gmt(using_gmt
) {}
2090 void encode(bufferlist
&bl
) const;
2091 void decode(bufferlist::iterator
&bl
);
2092 void dump(Formatter
*f
) const;
2093 static void generate_test_instances(list
<pg_hit_set_info_t
*>& o
);
2095 WRITE_CLASS_ENCODER(pg_hit_set_info_t
)
2098 * pg_hit_set_history_t - information about a history of hitsets
2100 * Include information about the currently accumulating hit set as well
2101 * as archived/historical ones.
2103 struct pg_hit_set_history_t
{
2104 eversion_t current_last_update
; ///< last version inserted into current set
2105 list
<pg_hit_set_info_t
> history
; ///< archived sets, sorted oldest -> newest
2107 friend bool operator==(const pg_hit_set_history_t
& l
,
2108 const pg_hit_set_history_t
& r
) {
2110 l
.current_last_update
== r
.current_last_update
&&
2111 l
.history
== r
.history
;
2114 void encode(bufferlist
&bl
) const;
2115 void decode(bufferlist::iterator
&bl
);
2116 void dump(Formatter
*f
) const;
2117 static void generate_test_instances(list
<pg_hit_set_history_t
*>& o
);
2119 WRITE_CLASS_ENCODER(pg_hit_set_history_t
)
2122 // -----------------------------------------
2125 * pg_history_t - information about recent pg peering/mapping history
2127 * This is aggressively shared between OSDs to bound the amount of past
2128 * history they need to worry about.
2130 struct pg_history_t
{
2131 epoch_t epoch_created
; // epoch in which *pg* was created (pool or pg)
2132 epoch_t epoch_pool_created
; // epoch in which *pool* was created
2133 // (note: may be pg creation epoch for
2134 // pre-luminous clusters)
2135 epoch_t last_epoch_started
; // lower bound on last epoch started (anywhere, not necessarily locally)
2136 epoch_t last_interval_started
; // first epoch of last_epoch_started interval
2137 epoch_t last_epoch_clean
; // lower bound on last epoch the PG was completely clean.
2138 epoch_t last_interval_clean
; // first epoch of last_epoch_clean interval
2139 epoch_t last_epoch_split
; // as parent or child
2140 epoch_t last_epoch_marked_full
; // pool or cluster
2143 * In the event of a map discontinuity, same_*_since may reflect the first
2144 * map the osd has seen in the new map sequence rather than the actual start
2145 * of the interval. This is ok since a discontinuity at epoch e means there
2146 * must have been a clean interval between e and now and that we cannot be
2147 * in the active set during the interval containing e.
2149 epoch_t same_up_since
; // same acting set since
2150 epoch_t same_interval_since
; // same acting AND up set since
2151 epoch_t same_primary_since
; // same primary at least back through this epoch.
2153 eversion_t last_scrub
;
2154 eversion_t last_deep_scrub
;
2155 utime_t last_scrub_stamp
;
2156 utime_t last_deep_scrub_stamp
;
2157 utime_t last_clean_scrub_stamp
;
2159 friend bool operator==(const pg_history_t
& l
, const pg_history_t
& r
) {
2161 l
.epoch_created
== r
.epoch_created
&&
2162 l
.epoch_pool_created
== r
.epoch_pool_created
&&
2163 l
.last_epoch_started
== r
.last_epoch_started
&&
2164 l
.last_interval_started
== r
.last_interval_started
&&
2165 l
.last_epoch_clean
== r
.last_epoch_clean
&&
2166 l
.last_interval_clean
== r
.last_interval_clean
&&
2167 l
.last_epoch_split
== r
.last_epoch_split
&&
2168 l
.last_epoch_marked_full
== r
.last_epoch_marked_full
&&
2169 l
.same_up_since
== r
.same_up_since
&&
2170 l
.same_interval_since
== r
.same_interval_since
&&
2171 l
.same_primary_since
== r
.same_primary_since
&&
2172 l
.last_scrub
== r
.last_scrub
&&
2173 l
.last_deep_scrub
== r
.last_deep_scrub
&&
2174 l
.last_scrub_stamp
== r
.last_scrub_stamp
&&
2175 l
.last_deep_scrub_stamp
== r
.last_deep_scrub_stamp
&&
2176 l
.last_clean_scrub_stamp
== r
.last_clean_scrub_stamp
;
2181 epoch_pool_created(0),
2182 last_epoch_started(0),
2183 last_interval_started(0),
2184 last_epoch_clean(0),
2185 last_interval_clean(0),
2186 last_epoch_split(0),
2187 last_epoch_marked_full(0),
2188 same_up_since(0), same_interval_since(0), same_primary_since(0) {}
2190 bool merge(const pg_history_t
&other
) {
2191 // Here, we only update the fields which cannot be calculated from the OSDmap.
2192 bool modified
= false;
2193 if (epoch_created
< other
.epoch_created
) {
2194 epoch_created
= other
.epoch_created
;
2197 if (epoch_pool_created
< other
.epoch_pool_created
) {
2198 // FIXME: for jewel compat only; this should either be 0 or always the
2199 // same value across all pg instances.
2200 epoch_pool_created
= other
.epoch_pool_created
;
2203 if (last_epoch_started
< other
.last_epoch_started
) {
2204 last_epoch_started
= other
.last_epoch_started
;
2207 if (last_interval_started
< other
.last_interval_started
) {
2208 last_interval_started
= other
.last_interval_started
;
2211 if (last_epoch_clean
< other
.last_epoch_clean
) {
2212 last_epoch_clean
= other
.last_epoch_clean
;
2215 if (last_interval_clean
< other
.last_interval_clean
) {
2216 last_interval_clean
= other
.last_interval_clean
;
2219 if (last_epoch_split
< other
.last_epoch_split
) {
2220 last_epoch_split
= other
.last_epoch_split
;
2223 if (last_epoch_marked_full
< other
.last_epoch_marked_full
) {
2224 last_epoch_marked_full
= other
.last_epoch_marked_full
;
2227 if (other
.last_scrub
> last_scrub
) {
2228 last_scrub
= other
.last_scrub
;
2231 if (other
.last_scrub_stamp
> last_scrub_stamp
) {
2232 last_scrub_stamp
= other
.last_scrub_stamp
;
2235 if (other
.last_deep_scrub
> last_deep_scrub
) {
2236 last_deep_scrub
= other
.last_deep_scrub
;
2239 if (other
.last_deep_scrub_stamp
> last_deep_scrub_stamp
) {
2240 last_deep_scrub_stamp
= other
.last_deep_scrub_stamp
;
2243 if (other
.last_clean_scrub_stamp
> last_clean_scrub_stamp
) {
2244 last_clean_scrub_stamp
= other
.last_clean_scrub_stamp
;
2250 void encode(bufferlist
& bl
) const;
2251 void decode(bufferlist::iterator
& p
);
2252 void dump(Formatter
*f
) const;
2253 static void generate_test_instances(list
<pg_history_t
*>& o
);
2255 WRITE_CLASS_ENCODER(pg_history_t
)
2257 inline ostream
& operator<<(ostream
& out
, const pg_history_t
& h
) {
2258 return out
<< "ec=" << h
.epoch_created
<< "/" << h
.epoch_pool_created
2259 << " lis/c " << h
.last_interval_started
2260 << "/" << h
.last_interval_clean
2261 << " les/c/f " << h
.last_epoch_started
<< "/" << h
.last_epoch_clean
2262 << "/" << h
.last_epoch_marked_full
2263 << " " << h
.same_up_since
2264 << "/" << h
.same_interval_since
2265 << "/" << h
.same_primary_since
;
2270 * pg_info_t - summary of PG statistics.
2273 * - last_complete implies we have all objects that existed as of that
2274 * stamp, OR a newer object, OR have already applied a later delete.
2275 * - if last_complete >= log.bottom, then we know pg contents thru log.head.
2276 * otherwise, we have no idea what the pg is supposed to contain.
2280 eversion_t last_update
; ///< last object version applied to store.
2281 eversion_t last_complete
; ///< last version pg was complete through.
2282 epoch_t last_epoch_started
; ///< last epoch at which this pg started on this osd
2283 epoch_t last_interval_started
; ///< first epoch of last_epoch_started interval
2285 version_t last_user_version
; ///< last user object version applied to store
2287 eversion_t log_tail
; ///< oldest log entry.
2289 hobject_t last_backfill
; ///< objects >= this and < last_complete may be missing
2290 bool last_backfill_bitwise
; ///< true if last_backfill reflects a bitwise (vs nibblewise) sort
2292 interval_set
<snapid_t
> purged_snaps
;
2296 pg_history_t history
;
2297 pg_hit_set_history_t hit_set
;
2299 friend bool operator==(const pg_info_t
& l
, const pg_info_t
& r
) {
2302 l
.last_update
== r
.last_update
&&
2303 l
.last_complete
== r
.last_complete
&&
2304 l
.last_epoch_started
== r
.last_epoch_started
&&
2305 l
.last_interval_started
== r
.last_interval_started
&&
2306 l
.last_user_version
== r
.last_user_version
&&
2307 l
.log_tail
== r
.log_tail
&&
2308 l
.last_backfill
== r
.last_backfill
&&
2309 l
.last_backfill_bitwise
== r
.last_backfill_bitwise
&&
2310 l
.purged_snaps
== r
.purged_snaps
&&
2311 l
.stats
== r
.stats
&&
2312 l
.history
== r
.history
&&
2313 l
.hit_set
== r
.hit_set
;
2317 : last_epoch_started(0),
2318 last_interval_started(0),
2319 last_user_version(0),
2320 last_backfill(hobject_t::get_max()),
2321 last_backfill_bitwise(false)
2323 // cppcheck-suppress noExplicitConstructor
2326 last_epoch_started(0),
2327 last_interval_started(0),
2328 last_user_version(0),
2329 last_backfill(hobject_t::get_max()),
2330 last_backfill_bitwise(false)
2333 void set_last_backfill(hobject_t pos
) {
2334 last_backfill
= pos
;
2335 last_backfill_bitwise
= true;
2338 bool is_empty() const { return last_update
.version
== 0; }
2339 bool dne() const { return history
.epoch_created
== 0; }
2341 bool is_incomplete() const { return !last_backfill
.is_max(); }
2343 void encode(bufferlist
& bl
) const;
2344 void decode(bufferlist::iterator
& p
);
2345 void dump(Formatter
*f
) const;
2346 bool overlaps_with(const pg_info_t
&oinfo
) const {
2347 return last_update
> oinfo
.log_tail
?
2348 oinfo
.last_update
>= log_tail
:
2349 last_update
>= oinfo
.log_tail
;
2351 static void generate_test_instances(list
<pg_info_t
*>& o
);
2353 WRITE_CLASS_ENCODER(pg_info_t
)
2355 inline ostream
& operator<<(ostream
& out
, const pg_info_t
& pgi
)
2357 out
<< pgi
.pgid
<< "(";
2363 out
<< " v " << pgi
.last_update
;
2364 if (pgi
.last_complete
!= pgi
.last_update
)
2365 out
<< " lc " << pgi
.last_complete
;
2366 out
<< " (" << pgi
.log_tail
<< "," << pgi
.last_update
<< "]";
2368 if (pgi
.is_incomplete())
2369 out
<< " lb " << pgi
.last_backfill
2370 << (pgi
.last_backfill_bitwise
? " (bitwise)" : " (NIBBLEWISE)");
2371 //out << " c " << pgi.epoch_created;
2372 out
<< " local-lis/les=" << pgi
.last_interval_started
2373 << "/" << pgi
.last_epoch_started
;
2374 out
<< " n=" << pgi
.stats
.stats
.sum
.num_objects
;
2375 out
<< " " << pgi
.history
2381 * pg_fast_info_t - common pg_info_t fields
2383 * These are the fields of pg_info_t (and children) that are updated for
2384 * most IO operations.
2387 * Because we rely on these fields to be applied to the normal
2388 * info struct, adding a new field here that is not also new in info
2389 * means that we must set an incompat OSD feature bit!
2391 struct pg_fast_info_t
{
2392 eversion_t last_update
;
2393 eversion_t last_complete
;
2394 version_t last_user_version
;
2395 struct { // pg_stat_t stats
2397 version_t reported_seq
;
2399 utime_t last_active
;
2400 utime_t last_peered
;
2402 utime_t last_unstale
;
2403 utime_t last_undegraded
;
2404 utime_t last_fullsized
;
2405 int64_t log_size
; // (also ondisk_log_size, which has the same value)
2406 struct { // object_stat_collection_t stats;
2407 struct { // objct_stat_sum_t sum
2408 int64_t num_bytes
; // in bytes
2409 int64_t num_objects
;
2410 int64_t num_object_copies
;
2415 int64_t num_objects_dirty
;
2420 void populate_from(const pg_info_t
& info
) {
2421 last_update
= info
.last_update
;
2422 last_complete
= info
.last_complete
;
2423 last_user_version
= info
.last_user_version
;
2424 stats
.version
= info
.stats
.version
;
2425 stats
.reported_seq
= info
.stats
.reported_seq
;
2426 stats
.last_fresh
= info
.stats
.last_fresh
;
2427 stats
.last_active
= info
.stats
.last_active
;
2428 stats
.last_peered
= info
.stats
.last_peered
;
2429 stats
.last_clean
= info
.stats
.last_clean
;
2430 stats
.last_unstale
= info
.stats
.last_unstale
;
2431 stats
.last_undegraded
= info
.stats
.last_undegraded
;
2432 stats
.last_fullsized
= info
.stats
.last_fullsized
;
2433 stats
.log_size
= info
.stats
.log_size
;
2434 stats
.stats
.sum
.num_bytes
= info
.stats
.stats
.sum
.num_bytes
;
2435 stats
.stats
.sum
.num_objects
= info
.stats
.stats
.sum
.num_objects
;
2436 stats
.stats
.sum
.num_object_copies
= info
.stats
.stats
.sum
.num_object_copies
;
2437 stats
.stats
.sum
.num_rd
= info
.stats
.stats
.sum
.num_rd
;
2438 stats
.stats
.sum
.num_rd_kb
= info
.stats
.stats
.sum
.num_rd_kb
;
2439 stats
.stats
.sum
.num_wr
= info
.stats
.stats
.sum
.num_wr
;
2440 stats
.stats
.sum
.num_wr_kb
= info
.stats
.stats
.sum
.num_wr_kb
;
2441 stats
.stats
.sum
.num_objects_dirty
= info
.stats
.stats
.sum
.num_objects_dirty
;
2444 bool try_apply_to(pg_info_t
* info
) {
2445 if (last_update
<= info
->last_update
)
2447 info
->last_update
= last_update
;
2448 info
->last_complete
= last_complete
;
2449 info
->last_user_version
= last_user_version
;
2450 info
->stats
.version
= stats
.version
;
2451 info
->stats
.reported_seq
= stats
.reported_seq
;
2452 info
->stats
.last_fresh
= stats
.last_fresh
;
2453 info
->stats
.last_active
= stats
.last_active
;
2454 info
->stats
.last_peered
= stats
.last_peered
;
2455 info
->stats
.last_clean
= stats
.last_clean
;
2456 info
->stats
.last_unstale
= stats
.last_unstale
;
2457 info
->stats
.last_undegraded
= stats
.last_undegraded
;
2458 info
->stats
.last_fullsized
= stats
.last_fullsized
;
2459 info
->stats
.log_size
= stats
.log_size
;
2460 info
->stats
.ondisk_log_size
= stats
.log_size
;
2461 info
->stats
.stats
.sum
.num_bytes
= stats
.stats
.sum
.num_bytes
;
2462 info
->stats
.stats
.sum
.num_objects
= stats
.stats
.sum
.num_objects
;
2463 info
->stats
.stats
.sum
.num_object_copies
= stats
.stats
.sum
.num_object_copies
;
2464 info
->stats
.stats
.sum
.num_rd
= stats
.stats
.sum
.num_rd
;
2465 info
->stats
.stats
.sum
.num_rd_kb
= stats
.stats
.sum
.num_rd_kb
;
2466 info
->stats
.stats
.sum
.num_wr
= stats
.stats
.sum
.num_wr
;
2467 info
->stats
.stats
.sum
.num_wr_kb
= stats
.stats
.sum
.num_wr_kb
;
2468 info
->stats
.stats
.sum
.num_objects_dirty
= stats
.stats
.sum
.num_objects_dirty
;
2472 void encode(bufferlist
& bl
) const {
2473 ENCODE_START(1, 1, bl
);
2474 ::encode(last_update
, bl
);
2475 ::encode(last_complete
, bl
);
2476 ::encode(last_user_version
, bl
);
2477 ::encode(stats
.version
, bl
);
2478 ::encode(stats
.reported_seq
, bl
);
2479 ::encode(stats
.last_fresh
, bl
);
2480 ::encode(stats
.last_active
, bl
);
2481 ::encode(stats
.last_peered
, bl
);
2482 ::encode(stats
.last_clean
, bl
);
2483 ::encode(stats
.last_unstale
, bl
);
2484 ::encode(stats
.last_undegraded
, bl
);
2485 ::encode(stats
.last_fullsized
, bl
);
2486 ::encode(stats
.log_size
, bl
);
2487 ::encode(stats
.stats
.sum
.num_bytes
, bl
);
2488 ::encode(stats
.stats
.sum
.num_objects
, bl
);
2489 ::encode(stats
.stats
.sum
.num_object_copies
, bl
);
2490 ::encode(stats
.stats
.sum
.num_rd
, bl
);
2491 ::encode(stats
.stats
.sum
.num_rd_kb
, bl
);
2492 ::encode(stats
.stats
.sum
.num_wr
, bl
);
2493 ::encode(stats
.stats
.sum
.num_wr_kb
, bl
);
2494 ::encode(stats
.stats
.sum
.num_objects_dirty
, bl
);
2497 void decode(bufferlist::iterator
& p
) {
2499 ::decode(last_update
, p
);
2500 ::decode(last_complete
, p
);
2501 ::decode(last_user_version
, p
);
2502 ::decode(stats
.version
, p
);
2503 ::decode(stats
.reported_seq
, p
);
2504 ::decode(stats
.last_fresh
, p
);
2505 ::decode(stats
.last_active
, p
);
2506 ::decode(stats
.last_peered
, p
);
2507 ::decode(stats
.last_clean
, p
);
2508 ::decode(stats
.last_unstale
, p
);
2509 ::decode(stats
.last_undegraded
, p
);
2510 ::decode(stats
.last_fullsized
, p
);
2511 ::decode(stats
.log_size
, p
);
2512 ::decode(stats
.stats
.sum
.num_bytes
, p
);
2513 ::decode(stats
.stats
.sum
.num_objects
, p
);
2514 ::decode(stats
.stats
.sum
.num_object_copies
, p
);
2515 ::decode(stats
.stats
.sum
.num_rd
, p
);
2516 ::decode(stats
.stats
.sum
.num_rd_kb
, p
);
2517 ::decode(stats
.stats
.sum
.num_wr
, p
);
2518 ::decode(stats
.stats
.sum
.num_wr_kb
, p
);
2519 ::decode(stats
.stats
.sum
.num_objects_dirty
, p
);
2523 WRITE_CLASS_ENCODER(pg_fast_info_t
)
2526 struct pg_notify_t
{
2527 epoch_t query_epoch
;
2533 query_epoch(0), epoch_sent(0), to(shard_id_t::NO_SHARD
),
2534 from(shard_id_t::NO_SHARD
) {}
2538 epoch_t query_epoch
,
2540 const pg_info_t
&info
)
2541 : query_epoch(query_epoch
),
2542 epoch_sent(epoch_sent
),
2543 info(info
), to(to
), from(from
) {
2544 assert(from
== info
.pgid
.shard
);
2546 void encode(bufferlist
&bl
) const;
2547 void decode(bufferlist::iterator
&p
);
2548 void dump(Formatter
*f
) const;
2549 static void generate_test_instances(list
<pg_notify_t
*> &o
);
2551 WRITE_CLASS_ENCODER(pg_notify_t
)
2552 ostream
&operator<<(ostream
&lhs
, const pg_notify_t
¬ify
);
2557 * PastIntervals -- information needed to determine the PriorSet and
2558 * the might_have_unfound set
2560 class PastIntervals
{
2562 struct pg_interval_t
{
2563 vector
<int32_t> up
, acting
;
2564 epoch_t first
, last
;
2570 : first(0), last(0),
2571 maybe_went_rw(false),
2577 vector
<int32_t> &&up
,
2578 vector
<int32_t> &&acting
,
2584 : up(up
), acting(acting
), first(first
), last(last
),
2585 maybe_went_rw(maybe_went_rw
), primary(primary
), up_primary(up_primary
)
2588 void encode(bufferlist
& bl
) const;
2589 void decode(bufferlist::iterator
& bl
);
2590 void dump(Formatter
*f
) const;
2591 static void generate_test_instances(list
<pg_interval_t
*>& o
);
2594 PastIntervals() = default;
2595 PastIntervals(bool ec_pool
, const OSDMap
&osdmap
) : PastIntervals() {
2596 update_type_from_map(ec_pool
, osdmap
);
2598 PastIntervals(bool ec_pool
, bool compact
) : PastIntervals() {
2599 update_type(ec_pool
, compact
);
2601 PastIntervals(PastIntervals
&&rhs
) = default;
2602 PastIntervals
&operator=(PastIntervals
&&rhs
) = default;
2604 PastIntervals(const PastIntervals
&rhs
);
2605 PastIntervals
&operator=(const PastIntervals
&rhs
);
2607 class interval_rep
{
2609 virtual size_t size() const = 0;
2610 virtual bool empty() const = 0;
2611 virtual void clear() = 0;
2612 virtual pair
<epoch_t
, epoch_t
> get_bounds() const = 0;
2613 virtual set
<pg_shard_t
> get_all_participants(
2614 bool ec_pool
) const = 0;
2615 virtual void add_interval(bool ec_pool
, const pg_interval_t
&interval
) = 0;
2616 virtual unique_ptr
<interval_rep
> clone() const = 0;
2617 virtual ostream
&print(ostream
&out
) const = 0;
2618 virtual void encode(bufferlist
&bl
) const = 0;
2619 virtual void decode(bufferlist::iterator
&bl
) = 0;
2620 virtual void dump(Formatter
*f
) const = 0;
2621 virtual bool is_classic() const = 0;
2622 virtual void iterate_mayberw_back_to(
2625 std::function
<void(epoch_t
, const set
<pg_shard_t
> &)> &&f
) const = 0;
2627 virtual bool has_full_intervals() const { return false; }
2628 virtual void iterate_all_intervals(
2629 std::function
<void(const pg_interval_t
&)> &&f
) const {
2630 assert(!has_full_intervals());
2631 assert(0 == "not valid for this implementation");
2634 virtual ~interval_rep() {}
2636 friend class pi_simple_rep
;
2637 friend class pi_compact_rep
;
2640 unique_ptr
<interval_rep
> past_intervals
;
2642 PastIntervals(interval_rep
*rep
) : past_intervals(rep
) {}
2645 void add_interval(bool ec_pool
, const pg_interval_t
&interval
) {
2646 assert(past_intervals
);
2647 return past_intervals
->add_interval(ec_pool
, interval
);
2650 bool is_classic() const {
2651 assert(past_intervals
);
2652 return past_intervals
->is_classic();
2655 void encode(bufferlist
&bl
) const {
2656 ENCODE_START(1, 1, bl
);
2657 if (past_intervals
) {
2658 __u8 type
= is_classic() ? 1 : 2;
2660 past_intervals
->encode(bl
);
2662 ::encode((__u8
)0, bl
);
2666 void encode_classic(bufferlist
&bl
) const {
2667 if (past_intervals
) {
2668 assert(past_intervals
->is_classic());
2669 past_intervals
->encode(bl
);
2672 ::encode((uint32_t)0, bl
);
2676 void decode(bufferlist::iterator
&bl
);
2677 void decode_classic(bufferlist::iterator
&bl
);
2679 void dump(Formatter
*f
) const {
2680 assert(past_intervals
);
2681 past_intervals
->dump(f
);
2683 static void generate_test_instances(list
<PastIntervals
*> & o
);
2686 * Determines whether there is an interval change
2688 static bool is_new_interval(
2689 int old_acting_primary
,
2690 int new_acting_primary
,
2691 const vector
<int> &old_acting
,
2692 const vector
<int> &new_acting
,
2695 const vector
<int> &old_up
,
2696 const vector
<int> &new_up
,
2701 unsigned old_pg_num
,
2702 unsigned new_pg_num
,
2703 bool old_sort_bitwise
,
2704 bool new_sort_bitwise
,
2705 bool old_recovery_deletes
,
2706 bool new_recovery_deletes
,
2711 * Determines whether there is an interval change
2713 static bool is_new_interval(
2714 int old_acting_primary
, ///< [in] primary as of lastmap
2715 int new_acting_primary
, ///< [in] primary as of lastmap
2716 const vector
<int> &old_acting
, ///< [in] acting as of lastmap
2717 const vector
<int> &new_acting
, ///< [in] acting as of osdmap
2718 int old_up_primary
, ///< [in] up primary of lastmap
2719 int new_up_primary
, ///< [in] up primary of osdmap
2720 const vector
<int> &old_up
, ///< [in] up as of lastmap
2721 const vector
<int> &new_up
, ///< [in] up as of osdmap
2722 ceph::shared_ptr
<const OSDMap
> osdmap
, ///< [in] current map
2723 ceph::shared_ptr
<const OSDMap
> lastmap
, ///< [in] last map
2724 pg_t pgid
///< [in] pgid for pg
2728 * Integrates a new map into *past_intervals, returns true
2729 * if an interval was closed out.
2731 static bool check_new_interval(
2732 int old_acting_primary
, ///< [in] primary as of lastmap
2733 int new_acting_primary
, ///< [in] primary as of osdmap
2734 const vector
<int> &old_acting
, ///< [in] acting as of lastmap
2735 const vector
<int> &new_acting
, ///< [in] acting as of osdmap
2736 int old_up_primary
, ///< [in] up primary of lastmap
2737 int new_up_primary
, ///< [in] up primary of osdmap
2738 const vector
<int> &old_up
, ///< [in] up as of lastmap
2739 const vector
<int> &new_up
, ///< [in] up as of osdmap
2740 epoch_t same_interval_since
, ///< [in] as of osdmap
2741 epoch_t last_epoch_clean
, ///< [in] current
2742 ceph::shared_ptr
<const OSDMap
> osdmap
, ///< [in] current map
2743 ceph::shared_ptr
<const OSDMap
> lastmap
, ///< [in] last map
2744 pg_t pgid
, ///< [in] pgid for pg
2745 IsPGRecoverablePredicate
*could_have_gone_active
, /// [in] predicate whether the pg can be active
2746 PastIntervals
*past_intervals
, ///< [out] intervals
2747 ostream
*out
= 0 ///< [out] debug ostream
2750 friend ostream
& operator<<(ostream
& out
, const PastIntervals
&i
);
2752 template <typename F
>
2753 void iterate_mayberw_back_to(
2757 assert(past_intervals
);
2758 past_intervals
->iterate_mayberw_back_to(ec_pool
, les
, std::forward
<F
>(f
));
2761 assert(past_intervals
);
2762 past_intervals
->clear();
2766 * Should return a value which gives an indication of the amount
2767 * of state contained
2769 size_t size() const {
2770 assert(past_intervals
);
2771 return past_intervals
->size();
2774 bool empty() const {
2775 assert(past_intervals
);
2776 return past_intervals
->empty();
2779 void swap(PastIntervals
&other
) {
2781 swap(other
.past_intervals
, past_intervals
);
2785 * Return all shards which have been in the acting set back to the
2786 * latest epoch to which we have trimmed except for pg_whoami
2788 set
<pg_shard_t
> get_might_have_unfound(
2789 pg_shard_t pg_whoami
,
2790 bool ec_pool
) const {
2791 assert(past_intervals
);
2792 auto ret
= past_intervals
->get_all_participants(ec_pool
);
2793 ret
.erase(pg_whoami
);
2798 * Return all shards which we might want to talk to for peering
2800 set
<pg_shard_t
> get_all_probe(
2801 bool ec_pool
) const {
2802 assert(past_intervals
);
2803 return past_intervals
->get_all_participants(ec_pool
);
2806 /* Return the set of epochs [start, end) represented by the
2807 * past_interval set.
2809 pair
<epoch_t
, epoch_t
> get_bounds() const {
2810 assert(past_intervals
);
2811 return past_intervals
->get_bounds();
2821 bool ec_pool
= false;
2822 set
<pg_shard_t
> probe
; /// current+prior OSDs we need to probe.
2823 set
<int> down
; /// down osds that would normally be in @a probe and might be interesting.
2824 map
<int, epoch_t
> blocked_by
; /// current lost_at values for any OSDs in cur set for which (re)marking them lost would affect cur set
2826 bool pg_down
= false; /// some down osds are included in @a cur; the DOWN pg state bit should be set.
2827 unique_ptr
<IsPGRecoverablePredicate
> pcontdec
;
2829 PriorSet() = default;
2830 PriorSet(PriorSet
&&) = default;
2831 PriorSet
&operator=(PriorSet
&&) = default;
2833 PriorSet
&operator=(const PriorSet
&) = delete;
2834 PriorSet(const PriorSet
&) = delete;
2836 bool operator==(const PriorSet
&rhs
) const {
2837 return (ec_pool
== rhs
.ec_pool
) &&
2838 (probe
== rhs
.probe
) &&
2839 (down
== rhs
.down
) &&
2840 (blocked_by
== rhs
.blocked_by
) &&
2841 (pg_down
== rhs
.pg_down
);
2844 bool affected_by_map(
2845 const OSDMap
&osdmap
,
2846 const DoutPrefixProvider
*dpp
) const;
2848 // For verifying tests
2851 set
<pg_shard_t
> probe
,
2853 map
<int, epoch_t
> blocked_by
,
2855 IsPGRecoverablePredicate
*pcontdec
)
2856 : ec_pool(ec_pool
), probe(probe
), down(down
), blocked_by(blocked_by
),
2857 pg_down(pg_down
), pcontdec(pcontdec
) {}
2860 template <typename F
>
2862 const PastIntervals
&past_intervals
,
2864 epoch_t last_epoch_started
,
2865 IsPGRecoverablePredicate
*c
,
2867 const vector
<int> &up
,
2868 const vector
<int> &acting
,
2869 const DoutPrefixProvider
*dpp
);
2871 friend class PastIntervals
;
2874 void update_type(bool ec_pool
, bool compact
);
2875 void update_type_from_map(bool ec_pool
, const OSDMap
&osdmap
);
2877 template <typename
... Args
>
2878 PriorSet
get_prior_set(Args
&&... args
) const {
2879 return PriorSet(*this, std::forward
<Args
>(args
)...);
2882 WRITE_CLASS_ENCODER(PastIntervals
)
2884 ostream
& operator<<(ostream
& out
, const PastIntervals::pg_interval_t
& i
);
2885 ostream
& operator<<(ostream
& out
, const PastIntervals
&i
);
2886 ostream
& operator<<(ostream
& out
, const PastIntervals::PriorSet
&i
);
2888 template <typename F
>
2889 PastIntervals::PriorSet::PriorSet(
2890 const PastIntervals
&past_intervals
,
2892 epoch_t last_epoch_started
,
2893 IsPGRecoverablePredicate
*c
,
2895 const vector
<int> &up
,
2896 const vector
<int> &acting
,
2897 const DoutPrefixProvider
*dpp
)
2898 : ec_pool(ec_pool
), pg_down(false), pcontdec(c
)
2901 * We have to be careful to gracefully deal with situations like
2902 * so. Say we have a power outage or something that takes out both
2903 * OSDs, but the monitor doesn't mark them down in the same epoch.
2904 * The history may look like
2908 * 3: let's say B dies for good, too (say, from the power spike)
2911 * which makes it look like B may have applied updates to the PG
2912 * that we need in order to proceed. This sucks...
2914 * To minimize the risk of this happening, we CANNOT go active if
2915 * _any_ OSDs in the prior set are down until we send an MOSDAlive
2916 * to the monitor such that the OSDMap sets osd_up_thru to an epoch.
2917 * Then, we have something like
2924 * -> we can ignore B, bc it couldn't have gone active (alive_thru
2935 * -> we must wait for B, bc it was alive through 2, and could have
2936 * written to the pg.
2938 * If B is really dead, then an administrator will need to manually
2939 * intervene by marking the OSD as "lost."
2942 // Include current acting and up nodes... not because they may
2943 // contain old data (this interval hasn't gone active, obviously),
2944 // but because we want their pg_info to inform choose_acting(), and
2945 // so that we know what they do/do not have explicitly before
2946 // sending them any new info/logs/whatever.
2947 for (unsigned i
= 0; i
< acting
.size(); i
++) {
2948 if (acting
[i
] != 0x7fffffff /* CRUSH_ITEM_NONE, can't import crush.h here */)
2949 probe
.insert(pg_shard_t(acting
[i
], ec_pool
? shard_id_t(i
) : shard_id_t::NO_SHARD
));
2951 // It may be possible to exclude the up nodes, but let's keep them in
2953 for (unsigned i
= 0; i
< up
.size(); i
++) {
2954 if (up
[i
] != 0x7fffffff /* CRUSH_ITEM_NONE, can't import crush.h here */)
2955 probe
.insert(pg_shard_t(up
[i
], ec_pool
? shard_id_t(i
) : shard_id_t::NO_SHARD
));
2958 set
<pg_shard_t
> all_probe
= past_intervals
.get_all_probe(ec_pool
);
2959 ldpp_dout(dpp
, 10) << "build_prior all_probe " << all_probe
<< dendl
;
2960 for (auto &&i
: all_probe
) {
2961 switch (f(0, i
.osd
, nullptr)) {
2975 past_intervals
.iterate_mayberw_back_to(
2978 [&](epoch_t start
, const set
<pg_shard_t
> &acting
) {
2979 ldpp_dout(dpp
, 10) << "build_prior maybe_rw interval:" << start
2980 << ", acting: " << acting
<< dendl
;
2982 // look at candidate osds during this interval. each falls into
2983 // one of three categories: up, down (but potentially
2984 // interesting), or lost (down, but we won't wait for it).
2985 set
<pg_shard_t
> up_now
;
2986 map
<int, epoch_t
> candidate_blocked_by
;
2987 // any candidates down now (that might have useful data)
2988 bool any_down_now
= false;
2990 // consider ACTING osds
2991 for (auto &&so
: acting
) {
2992 epoch_t lost_at
= 0;
2993 switch (f(start
, so
.osd
, &lost_at
)) {
2995 // include past acting osds if they are up.
3000 ldpp_dout(dpp
, 10) << "build_prior prior osd." << so
.osd
3001 << " no longer exists" << dendl
;
3005 ldpp_dout(dpp
, 10) << "build_prior prior osd." << so
.osd
3006 << " is down, but lost_at " << lost_at
<< dendl
;
3011 ldpp_dout(dpp
, 10) << "build_prior prior osd." << so
.osd
3012 << " is down" << dendl
;
3013 candidate_blocked_by
[so
.osd
] = lost_at
;
3014 any_down_now
= true;
3020 // if not enough osds survived this interval, and we may have gone rw,
3021 // then we need to wait for one of those osds to recover to
3022 // ensure that we haven't lost any information.
3023 if (!(*pcontdec
)(up_now
) && any_down_now
) {
3024 // fixme: how do we identify a "clean" shutdown anyway?
3025 ldpp_dout(dpp
, 10) << "build_prior possibly went active+rw,"
3026 << " insufficient up; including down osds" << dendl
;
3027 assert(!candidate_blocked_by
.empty());
3030 candidate_blocked_by
.begin(),
3031 candidate_blocked_by
.end());
3035 ldpp_dout(dpp
, 10) << "build_prior final: probe " << probe
3037 << " blocked_by " << blocked_by
3038 << (pg_down
? " pg_down":"")
3043 * pg_query_t - used to ask a peer for information about a pg.
3045 * note: if version=0, type=LOG, then we just provide our full log.
3054 const char *get_type_name() const {
3056 case INFO
: return "info";
3057 case LOG
: return "log";
3058 case MISSING
: return "missing";
3059 case FULLLOG
: return "fulllog";
3060 default: return "???";
3066 pg_history_t history
;
3071 pg_query_t() : type(-1), epoch_sent(0), to(shard_id_t::NO_SHARD
),
3072 from(shard_id_t::NO_SHARD
) {}
3077 const pg_history_t
& h
,
3081 epoch_sent(epoch_sent
),
3082 to(to
), from(from
) {
3090 const pg_history_t
& h
,
3092 : type(t
), since(s
), history(h
),
3093 epoch_sent(epoch_sent
), to(to
), from(from
) {
3097 void encode(bufferlist
&bl
, uint64_t features
) const;
3098 void decode(bufferlist::iterator
&bl
);
3100 void dump(Formatter
*f
) const;
3101 static void generate_test_instances(list
<pg_query_t
*>& o
);
3103 WRITE_CLASS_ENCODER_FEATURES(pg_query_t
)
3105 inline ostream
& operator<<(ostream
& out
, const pg_query_t
& q
) {
3106 out
<< "query(" << q
.get_type_name() << " " << q
.since
;
3107 if (q
.type
== pg_query_t::LOG
)
3108 out
<< " " << q
.history
;
3114 class ObjectModDesc
{
3115 bool can_local_rollback
;
3116 bool rollback_info_completed
;
3118 // version required to decode, reflected in encode/decode version
3119 __u8 max_required_version
= 1;
3123 virtual void append(uint64_t old_offset
) {}
3124 virtual void setattrs(map
<string
, boost::optional
<bufferlist
> > &attrs
) {}
3125 virtual void rmobject(version_t old_version
) {}
3127 * Used to support the unfound_lost_delete log event: if the stashed
3128 * version exists, we unstash it, otherwise, we do nothing. This way
3129 * each replica rolls back to whatever state it had prior to the attempt
3130 * at mark unfound lost delete
3132 virtual void try_rmobject(version_t old_version
) {
3133 rmobject(old_version
);
3135 virtual void create() {}
3136 virtual void update_snaps(const set
<snapid_t
> &old_snaps
) {}
3137 virtual void rollback_extents(
3139 const vector
<pair
<uint64_t, uint64_t> > &extents
) {}
3140 virtual ~Visitor() {}
3142 void visit(Visitor
*visitor
) const;
3143 mutable bufferlist bl
;
3151 ROLLBACK_EXTENTS
= 7
3153 ObjectModDesc() : can_local_rollback(true), rollback_info_completed(false) {
3154 bl
.reassign_to_mempool(mempool::mempool_osd_pglog
);
3156 void claim(ObjectModDesc
&other
) {
3159 can_local_rollback
= other
.can_local_rollback
;
3160 rollback_info_completed
= other
.rollback_info_completed
;
3162 void claim_append(ObjectModDesc
&other
) {
3163 if (!can_local_rollback
|| rollback_info_completed
)
3165 if (!other
.can_local_rollback
) {
3166 mark_unrollbackable();
3169 bl
.claim_append(other
.bl
);
3170 rollback_info_completed
= other
.rollback_info_completed
;
3172 void swap(ObjectModDesc
&other
) {
3176 swap(other
.can_local_rollback
, can_local_rollback
);
3177 swap(other
.rollback_info_completed
, rollback_info_completed
);
3178 swap(other
.max_required_version
, max_required_version
);
3180 void append_id(ModID id
) {
3184 void append(uint64_t old_size
) {
3185 if (!can_local_rollback
|| rollback_info_completed
)
3187 ENCODE_START(1, 1, bl
);
3189 ::encode(old_size
, bl
);
3192 void setattrs(map
<string
, boost::optional
<bufferlist
> > &old_attrs
) {
3193 if (!can_local_rollback
|| rollback_info_completed
)
3195 ENCODE_START(1, 1, bl
);
3196 append_id(SETATTRS
);
3197 ::encode(old_attrs
, bl
);
3200 bool rmobject(version_t deletion_version
) {
3201 if (!can_local_rollback
|| rollback_info_completed
)
3203 ENCODE_START(1, 1, bl
);
3205 ::encode(deletion_version
, bl
);
3207 rollback_info_completed
= true;
3210 bool try_rmobject(version_t deletion_version
) {
3211 if (!can_local_rollback
|| rollback_info_completed
)
3213 ENCODE_START(1, 1, bl
);
3214 append_id(TRY_DELETE
);
3215 ::encode(deletion_version
, bl
);
3217 rollback_info_completed
= true;
3221 if (!can_local_rollback
|| rollback_info_completed
)
3223 rollback_info_completed
= true;
3224 ENCODE_START(1, 1, bl
);
3228 void update_snaps(const set
<snapid_t
> &old_snaps
) {
3229 if (!can_local_rollback
|| rollback_info_completed
)
3231 ENCODE_START(1, 1, bl
);
3232 append_id(UPDATE_SNAPS
);
3233 ::encode(old_snaps
, bl
);
3236 void rollback_extents(
3237 version_t gen
, const vector
<pair
<uint64_t, uint64_t> > &extents
) {
3238 assert(can_local_rollback
);
3239 assert(!rollback_info_completed
);
3240 if (max_required_version
< 2)
3241 max_required_version
= 2;
3242 ENCODE_START(2, 2, bl
);
3243 append_id(ROLLBACK_EXTENTS
);
3245 ::encode(extents
, bl
);
3249 // cannot be rolled back
3250 void mark_unrollbackable() {
3251 can_local_rollback
= false;
3254 bool can_rollback() const {
3255 return can_local_rollback
;
3257 bool empty() const {
3258 return can_local_rollback
&& (bl
.length() == 0);
3261 bool requires_kraken() const {
3262 return max_required_version
>= 2;
3266 * Create fresh copy of bl bytes to avoid keeping large buffers around
3267 * in the case that bl contains ptrs which point into a much larger
3270 void trim_bl() const {
3271 if (bl
.length() > 0)
3274 void encode(bufferlist
&bl
) const;
3275 void decode(bufferlist::iterator
&bl
);
3276 void dump(Formatter
*f
) const;
3277 static void generate_test_instances(list
<ObjectModDesc
*>& o
);
3279 WRITE_CLASS_ENCODER(ObjectModDesc
)
3283 * pg_log_entry_t - single entry/event in pg log
3286 struct pg_log_entry_t
{
3288 MODIFY
= 1, // some unspecified modification (but not *all* modifications)
3289 CLONE
= 2, // cloned object from head
3290 DELETE
= 3, // deleted object
3291 BACKLOG
= 4, // event invented by generate_backlog [deprecated]
3292 LOST_REVERT
= 5, // lost new version, revert to an older version.
3293 LOST_DELETE
= 6, // lost new version, revert to no object (deleted).
3294 LOST_MARK
= 7, // lost new version, now EIO
3295 PROMOTE
= 8, // promoted object from another tier
3296 CLEAN
= 9, // mark an object clean
3297 ERROR
= 10, // write that returned an error
3299 static const char *get_op_name(int op
) {
3325 const char *get_op_name() const {
3326 return get_op_name(op
);
3329 // describes state for a locally-rollbackable entry
3330 ObjectModDesc mod_desc
;
3331 bufferlist snaps
; // only for clone entries
3333 osd_reqid_t reqid
; // caller+tid to uniquely identify request
3334 mempool::osd_pglog::vector
<pair
<osd_reqid_t
, version_t
> > extra_reqids
;
3335 eversion_t version
, prior_version
, reverting_to
;
3336 version_t user_version
; // the user version for this entry
3337 utime_t mtime
; // this is the _user_ mtime, mind you
3338 int32_t return_code
; // only stored for ERRORs for dup detection
3341 bool invalid_hash
; // only when decoding sobject_t based entries
3342 bool invalid_pool
; // only when decoding pool-less hobject based entries
3345 : user_version(0), return_code(0), op(0),
3346 invalid_hash(false), invalid_pool(false) {
3347 snaps
.reassign_to_mempool(mempool::mempool_osd_pglog
);
3349 pg_log_entry_t(int _op
, const hobject_t
& _soid
,
3350 const eversion_t
& v
, const eversion_t
& pv
,
3352 const osd_reqid_t
& rid
, const utime_t
& mt
,
3354 : soid(_soid
), reqid(rid
), version(v
), prior_version(pv
), user_version(uv
),
3355 mtime(mt
), return_code(return_code
), op(_op
),
3356 invalid_hash(false), invalid_pool(false) {
3357 snaps
.reassign_to_mempool(mempool::mempool_osd_pglog
);
3360 bool is_clone() const { return op
== CLONE
; }
3361 bool is_modify() const { return op
== MODIFY
; }
3362 bool is_promote() const { return op
== PROMOTE
; }
3363 bool is_clean() const { return op
== CLEAN
; }
3364 bool is_backlog() const { return op
== BACKLOG
; }
3365 bool is_lost_revert() const { return op
== LOST_REVERT
; }
3366 bool is_lost_delete() const { return op
== LOST_DELETE
; }
3367 bool is_lost_mark() const { return op
== LOST_MARK
; }
3368 bool is_error() const { return op
== ERROR
; }
3370 bool is_update() const {
3372 is_clone() || is_modify() || is_promote() || is_clean() ||
3373 is_backlog() || is_lost_revert() || is_lost_mark();
3375 bool is_delete() const {
3376 return op
== DELETE
|| op
== LOST_DELETE
;
3379 bool can_rollback() const {
3380 return mod_desc
.can_rollback();
3383 void mark_unrollbackable() {
3384 mod_desc
.mark_unrollbackable();
3387 bool requires_kraken() const {
3388 return mod_desc
.requires_kraken();
3391 // Errors are only used for dup detection, whereas
3392 // the index by objects is used by recovery, copy_get,
3393 // and other facilities that don't expect or need to
3394 // be aware of error entries.
3395 bool object_is_indexed() const {
3399 bool reqid_is_indexed() const {
3400 return reqid
!= osd_reqid_t() &&
3401 (op
== MODIFY
|| op
== DELETE
|| op
== ERROR
);
3404 string
get_key_name() const;
3405 void encode_with_checksum(bufferlist
& bl
) const;
3406 void decode_with_checksum(bufferlist::iterator
& p
);
3408 void encode(bufferlist
&bl
) const;
3409 void decode(bufferlist::iterator
&bl
);
3410 void dump(Formatter
*f
) const;
3411 static void generate_test_instances(list
<pg_log_entry_t
*>& o
);
3414 WRITE_CLASS_ENCODER(pg_log_entry_t
)
3416 ostream
& operator<<(ostream
& out
, const pg_log_entry_t
& e
);
3418 struct pg_log_dup_t
{
3419 osd_reqid_t reqid
; // caller+tid to uniquely identify request
3421 version_t user_version
; // the user version for this entry
3422 int32_t return_code
; // only stored for ERRORs for dup detection
3425 : user_version(0), return_code(0)
3427 explicit pg_log_dup_t(const pg_log_entry_t
& entry
)
3428 : reqid(entry
.reqid
), version(entry
.version
),
3429 user_version(entry
.user_version
), return_code(entry
.return_code
)
3431 pg_log_dup_t(const eversion_t
& v
, version_t uv
,
3432 const osd_reqid_t
& rid
, int return_code
)
3433 : reqid(rid
), version(v
), user_version(uv
),
3434 return_code(return_code
)
3437 string
get_key_name() const;
3438 void encode(bufferlist
&bl
) const;
3439 void decode(bufferlist::iterator
&bl
);
3440 void dump(Formatter
*f
) const;
3441 static void generate_test_instances(list
<pg_log_dup_t
*>& o
);
3443 bool operator==(const pg_log_dup_t
&rhs
) const {
3444 return reqid
== rhs
.reqid
&&
3445 version
== rhs
.version
&&
3446 user_version
== rhs
.user_version
&&
3447 return_code
== rhs
.return_code
;
3449 bool operator!=(const pg_log_dup_t
&rhs
) const {
3450 return !(*this == rhs
);
3453 friend std::ostream
& operator<<(std::ostream
& out
, const pg_log_dup_t
& e
);
3455 WRITE_CLASS_ENCODER(pg_log_dup_t
)
3457 std::ostream
& operator<<(std::ostream
& out
, const pg_log_dup_t
& e
);
3460 * pg_log_t - incremental log of recent pg changes.
3462 * serves as a recovery queue for recent changes.
3466 * head - newest entry (update|delete)
3467 * tail - entry previous to oldest (update|delete) for which we have
3468 * complete negative information.
3469 * i.e. we can infer pg contents for any store whose last_update >= tail.
3471 eversion_t head
; // newest entry
3472 eversion_t tail
; // version prior to oldest
3475 // We can rollback rollback-able entries > can_rollback_to
3476 eversion_t can_rollback_to
;
3478 // always <= can_rollback_to, indicates how far stashed rollback
3479 // data can be found
3480 eversion_t rollback_info_trimmed_to
;
3484 mempool::osd_pglog::list
<pg_log_entry_t
> log
;
3486 // entries just for dup op detection ordered oldest to newest
3487 mempool::osd_pglog::list
<pg_log_dup_t
> dups
;
3489 pg_log_t() = default;
3490 pg_log_t(const eversion_t
&last_update
,
3491 const eversion_t
&log_tail
,
3492 const eversion_t
&can_rollback_to
,
3493 const eversion_t
&rollback_info_trimmed_to
,
3494 mempool::osd_pglog::list
<pg_log_entry_t
> &&entries
,
3495 mempool::osd_pglog::list
<pg_log_dup_t
> &&dup_entries
)
3496 : head(last_update
), tail(log_tail
), can_rollback_to(can_rollback_to
),
3497 rollback_info_trimmed_to(rollback_info_trimmed_to
),
3498 log(std::move(entries
)), dups(std::move(dup_entries
)) {}
3499 pg_log_t(const eversion_t
&last_update
,
3500 const eversion_t
&log_tail
,
3501 const eversion_t
&can_rollback_to
,
3502 const eversion_t
&rollback_info_trimmed_to
,
3503 const std::list
<pg_log_entry_t
> &entries
,
3504 const std::list
<pg_log_dup_t
> &dup_entries
)
3505 : head(last_update
), tail(log_tail
), can_rollback_to(can_rollback_to
),
3506 rollback_info_trimmed_to(rollback_info_trimmed_to
) {
3507 for (auto &&entry
: entries
) {
3508 log
.push_back(entry
);
3510 for (auto &&entry
: dup_entries
) {
3511 dups
.push_back(entry
);
3517 rollback_info_trimmed_to
= can_rollback_to
= head
= tail
= z
;
3522 eversion_t
get_rollback_info_trimmed_to() const {
3523 return rollback_info_trimmed_to
;
3525 eversion_t
get_can_rollback_to() const {
3526 return can_rollback_to
;
3530 pg_log_t
split_out_child(pg_t child_pgid
, unsigned split_bits
) {
3531 mempool::osd_pglog::list
<pg_log_entry_t
> oldlog
, childlog
;
3534 eversion_t old_tail
;
3535 unsigned mask
= ~((~0)<<split_bits
);
3536 for (auto i
= oldlog
.begin();
3539 if ((i
->soid
.get_hash() & mask
) == child_pgid
.m_seed
) {
3540 childlog
.push_back(*i
);
3547 // osd_reqid is unique, so it doesn't matter if there are extra
3548 // dup entries in each pg. To avoid storing oid with the dup
3549 // entries, just copy the whole list.
3550 auto childdups(dups
);
3556 rollback_info_trimmed_to
,
3557 std::move(childlog
),
3558 std::move(childdups
));
3561 mempool::osd_pglog::list
<pg_log_entry_t
> rewind_from_head(eversion_t newhead
) {
3562 assert(newhead
>= tail
);
3564 mempool::osd_pglog::list
<pg_log_entry_t
>::iterator p
= log
.end();
3565 mempool::osd_pglog::list
<pg_log_entry_t
> divergent
;
3567 if (p
== log
.begin()) {
3568 // yikes, the whole thing is divergent!
3570 swap(divergent
, log
);
3574 if (p
->version
.version
<= newhead
.version
) {
3576 * look at eversion.version here. we want to avoid a situation like:
3577 * our log: 100'10 (0'0) m 10000004d3a.00000000/head by client4225.1:18529
3578 * new log: 122'10 (0'0) m 10000004d3a.00000000/head by client4225.1:18529
3579 * lower_bound = 100'9
3580 * i.e, same request, different version. If the eversion.version is > the
3581 * lower_bound, we it is divergent.
3584 divergent
.splice(divergent
.begin(), log
, p
, log
.end());
3587 assert(p
->version
> newhead
);
3591 if (can_rollback_to
> newhead
)
3592 can_rollback_to
= newhead
;
3594 if (rollback_info_trimmed_to
> newhead
)
3595 rollback_info_trimmed_to
= newhead
;
3600 bool empty() const {
3605 return head
.version
== 0 && head
.epoch
== 0;
3608 size_t approx_size() const {
3609 return head
.version
- tail
.version
;
3612 static void filter_log(spg_t import_pgid
, const OSDMap
&curmap
,
3613 const string
&hit_set_namespace
, const pg_log_t
&in
,
3614 pg_log_t
&out
, pg_log_t
&reject
);
3617 * copy entries from the tail of another pg_log_t
3619 * @param other pg_log_t to copy from
3620 * @param from copy entries after this version
3622 void copy_after(const pg_log_t
&other
, eversion_t from
);
3625 * copy a range of entries from another pg_log_t
3627 * @param other pg_log_t to copy from
3628 * @param from copy entries after this version
3629 * @param to up to and including this version
3631 void copy_range(const pg_log_t
&other
, eversion_t from
, eversion_t to
);
3634 * copy up to N entries
3636 * @param other source log
3637 * @param max max number of entries to copy
3639 void copy_up_to(const pg_log_t
&other
, int max
);
3641 ostream
& print(ostream
& out
) const;
3643 void encode(bufferlist
&bl
) const;
3644 void decode(bufferlist::iterator
&bl
, int64_t pool
= -1);
3645 void dump(Formatter
*f
) const;
3646 static void generate_test_instances(list
<pg_log_t
*>& o
);
3648 WRITE_CLASS_ENCODER(pg_log_t
)
3650 inline ostream
& operator<<(ostream
& out
, const pg_log_t
& log
)
3652 out
<< "log((" << log
.tail
<< "," << log
.head
<< "], crt="
3653 << log
.get_can_rollback_to() << ")";
3659 * pg_missing_t - summary of missing objects.
3661 * kept in memory, as a supplement to pg_log_t
3662 * also used to pass missing info in messages.
3664 struct pg_missing_item
{
3665 eversion_t need
, have
;
3666 enum missing_flags_t
{
3670 pg_missing_item() : flags(FLAG_NONE
) {}
3671 explicit pg_missing_item(eversion_t n
) : need(n
), flags(FLAG_NONE
) {} // have no old version
3672 pg_missing_item(eversion_t n
, eversion_t h
, bool is_delete
=false) : need(n
), have(h
) {
3673 set_delete(is_delete
);
3676 void encode(bufferlist
& bl
, uint64_t features
) const {
3677 if (HAVE_FEATURE(features
, OSD_RECOVERY_DELETES
)) {
3678 // encoding a zeroed eversion_t to differentiate between this and
3679 // legacy unversioned encoding - a need value of 0'0 is not
3680 // possible. This can be replaced with the legacy encoding
3681 // macros post-luminous.
3686 ::encode(static_cast<uint8_t>(flags
), bl
);
3688 // legacy unversioned encoding
3693 void decode(bufferlist::iterator
& bl
) {
3696 if (e
!= eversion_t()) {
3697 // legacy encoding, this is the need value
3705 flags
= static_cast<missing_flags_t
>(f
);
3709 void set_delete(bool is_delete
) {
3710 flags
= is_delete
? FLAG_DELETE
: FLAG_NONE
;
3713 bool is_delete() const {
3714 return (flags
& FLAG_DELETE
) == FLAG_DELETE
;
3717 string
flag_str() const {
3718 if (flags
== FLAG_NONE
) {
3725 void dump(Formatter
*f
) const {
3726 f
->dump_stream("need") << need
;
3727 f
->dump_stream("have") << have
;
3728 f
->dump_stream("flags") << flag_str();
3730 static void generate_test_instances(list
<pg_missing_item
*>& o
) {
3731 o
.push_back(new pg_missing_item
);
3732 o
.push_back(new pg_missing_item
);
3733 o
.back()->need
= eversion_t(1, 2);
3734 o
.back()->have
= eversion_t(1, 1);
3735 o
.push_back(new pg_missing_item
);
3736 o
.back()->need
= eversion_t(3, 5);
3737 o
.back()->have
= eversion_t(3, 4);
3738 o
.back()->flags
= FLAG_DELETE
;
3740 bool operator==(const pg_missing_item
&rhs
) const {
3741 return need
== rhs
.need
&& have
== rhs
.have
&& flags
== rhs
.flags
;
3743 bool operator!=(const pg_missing_item
&rhs
) const {
3744 return !(*this == rhs
);
3747 WRITE_CLASS_ENCODER_FEATURES(pg_missing_item
)
3748 ostream
& operator<<(ostream
& out
, const pg_missing_item
&item
);
3750 class pg_missing_const_i
{
3752 virtual const map
<hobject_t
, pg_missing_item
> &
3753 get_items() const = 0;
3754 virtual const map
<version_t
, hobject_t
> &get_rmissing() const = 0;
3755 virtual bool get_may_include_deletes() const = 0;
3756 virtual unsigned int num_missing() const = 0;
3757 virtual bool have_missing() const = 0;
3758 virtual bool is_missing(const hobject_t
& oid
, pg_missing_item
*out
= nullptr) const = 0;
3759 virtual bool is_missing(const hobject_t
& oid
, eversion_t v
) const = 0;
3760 virtual eversion_t
have_old(const hobject_t
& oid
) const = 0;
3761 virtual ~pg_missing_const_i() {}
3765 template <bool Track
>
3766 class ChangeTracker
{
3768 void changed(const hobject_t
&obj
) {}
3769 template <typename F
>
3770 void get_changed(F
&&f
) const {}
3772 bool is_clean() const {
3777 class ChangeTracker
<true> {
3778 set
<hobject_t
> _changed
;
3780 void changed(const hobject_t
&obj
) {
3781 _changed
.insert(obj
);
3783 template <typename F
>
3784 void get_changed(F
&&f
) const {
3785 for (auto const &i
: _changed
) {
3792 bool is_clean() const {
3793 return _changed
.empty();
3797 template <bool TrackChanges
>
3798 class pg_missing_set
: public pg_missing_const_i
{
3799 using item
= pg_missing_item
;
3800 map
<hobject_t
, item
> missing
; // oid -> (need v, have v)
3801 map
<version_t
, hobject_t
> rmissing
; // v -> oid
3802 ChangeTracker
<TrackChanges
> tracker
;
3805 pg_missing_set() = default;
3807 template <typename missing_type
>
3808 pg_missing_set(const missing_type
&m
) {
3809 missing
= m
.get_items();
3810 rmissing
= m
.get_rmissing();
3811 may_include_deletes
= m
.get_may_include_deletes();
3812 for (auto &&i
: missing
)
3813 tracker
.changed(i
.first
);
3816 bool may_include_deletes
= false;
3818 const map
<hobject_t
, item
> &get_items() const override
{
3821 const map
<version_t
, hobject_t
> &get_rmissing() const override
{
3824 bool get_may_include_deletes() const override
{
3825 return may_include_deletes
;
3827 unsigned int num_missing() const override
{
3828 return missing
.size();
3830 bool have_missing() const override
{
3831 return !missing
.empty();
3833 bool is_missing(const hobject_t
& oid
, pg_missing_item
*out
= nullptr) const override
{
3834 auto iter
= missing
.find(oid
);
3835 if (iter
== missing
.end())
3838 *out
= iter
->second
;
3841 bool is_missing(const hobject_t
& oid
, eversion_t v
) const override
{
3842 map
<hobject_t
, item
>::const_iterator m
=
3844 if (m
== missing
.end())
3846 const item
&item(m
->second
);
3851 eversion_t
have_old(const hobject_t
& oid
) const override
{
3852 map
<hobject_t
, item
>::const_iterator m
=
3854 if (m
== missing
.end())
3855 return eversion_t();
3856 const item
&item(m
->second
);
3860 void claim(pg_missing_set
& o
) {
3861 static_assert(!TrackChanges
, "Can't use claim with TrackChanges");
3862 missing
.swap(o
.missing
);
3863 rmissing
.swap(o
.rmissing
);
3867 * this needs to be called in log order as we extend the log. it
3868 * assumes missing is accurate up through the previous log entry.
3870 void add_next_event(const pg_log_entry_t
& e
) {
3871 map
<hobject_t
, item
>::iterator missing_it
;
3872 missing_it
= missing
.find(e
.soid
);
3873 bool is_missing_divergent_item
= missing_it
!= missing
.end();
3874 if (e
.prior_version
== eversion_t() || e
.is_clone()) {
3876 if (is_missing_divergent_item
) { // use iterator
3877 rmissing
.erase((missing_it
->second
).need
.version
);
3878 missing_it
->second
= item(e
.version
, eversion_t(), e
.is_delete()); // .have = nil
3879 } else // create new element in missing map
3880 missing
[e
.soid
] = item(e
.version
, eversion_t(), e
.is_delete()); // .have = nil
3881 } else if (is_missing_divergent_item
) {
3882 // already missing (prior).
3883 rmissing
.erase((missing_it
->second
).need
.version
);
3884 (missing_it
->second
).need
= e
.version
; // leave .have unchanged.
3885 missing_it
->second
.set_delete(e
.is_delete());
3886 } else if (e
.is_backlog()) {
3887 // May not have prior version
3888 assert(0 == "these don't exist anymore");
3890 // not missing, we must have prior_version (if any)
3891 assert(!is_missing_divergent_item
);
3892 missing
[e
.soid
] = item(e
.version
, e
.prior_version
, e
.is_delete());
3894 rmissing
[e
.version
.version
] = e
.soid
;
3895 tracker
.changed(e
.soid
);
3898 void revise_need(hobject_t oid
, eversion_t need
, bool is_delete
) {
3899 if (missing
.count(oid
)) {
3900 rmissing
.erase(missing
[oid
].need
.version
);
3901 missing
[oid
].need
= need
; // no not adjust .have
3902 missing
[oid
].set_delete(is_delete
);
3904 missing
[oid
] = item(need
, eversion_t(), is_delete
);
3906 rmissing
[need
.version
] = oid
;
3908 tracker
.changed(oid
);
3911 void revise_have(hobject_t oid
, eversion_t have
) {
3912 if (missing
.count(oid
)) {
3913 tracker
.changed(oid
);
3914 missing
[oid
].have
= have
;
3918 void add(const hobject_t
& oid
, eversion_t need
, eversion_t have
,
3920 missing
[oid
] = item(need
, have
, is_delete
);
3921 rmissing
[need
.version
] = oid
;
3922 tracker
.changed(oid
);
3925 void rm(const hobject_t
& oid
, eversion_t v
) {
3926 std::map
<hobject_t
, item
>::iterator p
= missing
.find(oid
);
3927 if (p
!= missing
.end() && p
->second
.need
<= v
)
3931 void rm(std::map
<hobject_t
, item
>::const_iterator m
) {
3932 tracker
.changed(m
->first
);
3933 rmissing
.erase(m
->second
.need
.version
);
3937 void got(const hobject_t
& oid
, eversion_t v
) {
3938 std::map
<hobject_t
, item
>::iterator p
= missing
.find(oid
);
3939 assert(p
!= missing
.end());
3940 assert(p
->second
.need
<= v
|| p
->second
.is_delete());
3944 void got(std::map
<hobject_t
, item
>::const_iterator m
) {
3945 tracker
.changed(m
->first
);
3946 rmissing
.erase(m
->second
.need
.version
);
3952 unsigned split_bits
,
3953 pg_missing_set
*omissing
) {
3954 omissing
->may_include_deletes
= may_include_deletes
;
3955 unsigned mask
= ~((~0)<<split_bits
);
3956 for (map
<hobject_t
, item
>::iterator i
= missing
.begin();
3959 if ((i
->first
.get_hash() & mask
) == child_pgid
.m_seed
) {
3960 omissing
->add(i
->first
, i
->second
.need
, i
->second
.have
,
3961 i
->second
.is_delete());
3970 for (auto const &i
: missing
)
3971 tracker
.changed(i
.first
);
3976 void encode(bufferlist
&bl
) const {
3977 ENCODE_START(4, 2, bl
);
3978 ::encode(missing
, bl
, may_include_deletes
? CEPH_FEATURE_OSD_RECOVERY_DELETES
: 0);
3979 ::encode(may_include_deletes
, bl
);
3982 void decode(bufferlist::iterator
&bl
, int64_t pool
= -1) {
3983 for (auto const &i
: missing
)
3984 tracker
.changed(i
.first
);
3985 DECODE_START_LEGACY_COMPAT_LEN(4, 2, 2, bl
);
3986 ::decode(missing
, bl
);
3987 if (struct_v
>= 4) {
3988 ::decode(may_include_deletes
, bl
);
3993 // Handle hobject_t upgrade
3994 map
<hobject_t
, item
> tmp
;
3995 for (map
<hobject_t
, item
>::iterator i
=
3999 if (!i
->first
.is_max() && i
->first
.pool
== -1) {
4000 hobject_t
to_insert(i
->first
);
4001 to_insert
.pool
= pool
;
4002 tmp
[to_insert
] = i
->second
;
4008 missing
.insert(tmp
.begin(), tmp
.end());
4011 for (map
<hobject_t
,item
>::iterator it
=
4013 it
!= missing
.end();
4015 rmissing
[it
->second
.need
.version
] = it
->first
;
4016 for (auto const &i
: missing
)
4017 tracker
.changed(i
.first
);
4019 void dump(Formatter
*f
) const {
4020 f
->open_array_section("missing");
4021 for (map
<hobject_t
,item
>::const_iterator p
=
4022 missing
.begin(); p
!= missing
.end(); ++p
) {
4023 f
->open_object_section("item");
4024 f
->dump_stream("object") << p
->first
;
4029 f
->dump_bool("may_include_deletes", may_include_deletes
);
4031 template <typename F
>
4032 void filter_objects(F
&&f
) {
4033 for (auto i
= missing
.begin(); i
!= missing
.end();) {
4041 static void generate_test_instances(list
<pg_missing_set
*>& o
) {
4042 o
.push_back(new pg_missing_set
);
4043 o
.push_back(new pg_missing_set
);
4045 hobject_t(object_t("foo"), "foo", 123, 456, 0, ""),
4046 eversion_t(5, 6), eversion_t(5, 1), false);
4047 o
.push_back(new pg_missing_set
);
4049 hobject_t(object_t("foo"), "foo", 123, 456, 0, ""),
4050 eversion_t(5, 6), eversion_t(5, 1), true);
4051 o
.back()->may_include_deletes
= true;
4053 template <typename F
>
4054 void get_changed(F
&&f
) const {
4055 tracker
.get_changed(f
);
4060 bool is_clean() const {
4061 return tracker
.is_clean();
4063 template <typename missing_t
>
4064 bool debug_verify_from_init(
4065 const missing_t
&init_missing
,
4066 ostream
*oss
) const {
4069 auto check_missing(init_missing
.get_items());
4070 tracker
.get_changed([&](const hobject_t
&hoid
) {
4071 check_missing
.erase(hoid
);
4072 if (missing
.count(hoid
)) {
4073 check_missing
.insert(*(missing
.find(hoid
)));
4077 if (check_missing
.size() != missing
.size()) {
4079 *oss
<< "Size mismatch, check: " << check_missing
.size()
4080 << ", actual: " << missing
.size() << "\n";
4084 for (auto &i
: missing
) {
4085 if (!check_missing
.count(i
.first
)) {
4087 *oss
<< "check_missing missing " << i
.first
<< "\n";
4089 } else if (check_missing
[i
.first
] != i
.second
) {
4091 *oss
<< "check_missing missing item mismatch on " << i
.first
4092 << ", check: " << check_missing
[i
.first
]
4093 << ", actual: " << i
.second
<< "\n";
4098 *oss
<< "check_missing: " << check_missing
<< "\n";
4099 set
<hobject_t
> changed
;
4100 tracker
.get_changed([&](const hobject_t
&hoid
) { changed
.insert(hoid
); });
4101 *oss
<< "changed: " << changed
<< "\n";
4106 template <bool TrackChanges
>
4108 const pg_missing_set
<TrackChanges
> &c
, bufferlist
&bl
, uint64_t features
=0) {
4111 ENCODE_DUMP_POST(cl
);
4113 template <bool TrackChanges
>
4114 void decode(pg_missing_set
<TrackChanges
> &c
, bufferlist::iterator
&p
) {
4117 template <bool TrackChanges
>
4118 ostream
& operator<<(ostream
& out
, const pg_missing_set
<TrackChanges
> &missing
)
4120 out
<< "missing(" << missing
.num_missing()
4121 << " may_include_deletes = " << missing
.may_include_deletes
;
4122 //if (missing.num_lost()) out << ", " << missing.num_lost() << " lost";
4127 using pg_missing_t
= pg_missing_set
<false>;
4128 using pg_missing_tracker_t
= pg_missing_set
<true>;
4132 * pg list objects response format
4135 struct pg_nls_response_t
{
4136 collection_list_handle_t handle
;
4137 list
<librados::ListObjectImpl
> entries
;
4139 void encode(bufferlist
& bl
) const {
4140 ENCODE_START(1, 1, bl
);
4141 ::encode(handle
, bl
);
4142 __u32 n
= (__u32
)entries
.size();
4144 for (list
<librados::ListObjectImpl
>::const_iterator i
= entries
.begin(); i
!= entries
.end(); ++i
) {
4145 ::encode(i
->nspace
, bl
);
4146 ::encode(i
->oid
, bl
);
4147 ::encode(i
->locator
, bl
);
4151 void decode(bufferlist::iterator
& bl
) {
4152 DECODE_START(1, bl
);
4153 ::decode(handle
, bl
);
4158 librados::ListObjectImpl i
;
4159 ::decode(i
.nspace
, bl
);
4160 ::decode(i
.oid
, bl
);
4161 ::decode(i
.locator
, bl
);
4162 entries
.push_back(i
);
4166 void dump(Formatter
*f
) const {
4167 f
->dump_stream("handle") << handle
;
4168 f
->open_array_section("entries");
4169 for (list
<librados::ListObjectImpl
>::const_iterator p
= entries
.begin(); p
!= entries
.end(); ++p
) {
4170 f
->open_object_section("object");
4171 f
->dump_string("namespace", p
->nspace
);
4172 f
->dump_string("object", p
->oid
);
4173 f
->dump_string("key", p
->locator
);
4178 static void generate_test_instances(list
<pg_nls_response_t
*>& o
) {
4179 o
.push_back(new pg_nls_response_t
);
4180 o
.push_back(new pg_nls_response_t
);
4181 o
.back()->handle
= hobject_t(object_t("hi"), "key", 1, 2, -1, "");
4182 o
.back()->entries
.push_back(librados::ListObjectImpl("", "one", ""));
4183 o
.back()->entries
.push_back(librados::ListObjectImpl("", "two", "twokey"));
4184 o
.back()->entries
.push_back(librados::ListObjectImpl("", "three", ""));
4185 o
.push_back(new pg_nls_response_t
);
4186 o
.back()->handle
= hobject_t(object_t("hi"), "key", 3, 4, -1, "");
4187 o
.back()->entries
.push_back(librados::ListObjectImpl("n1", "n1one", ""));
4188 o
.back()->entries
.push_back(librados::ListObjectImpl("n1", "n1two", "n1twokey"));
4189 o
.back()->entries
.push_back(librados::ListObjectImpl("n1", "n1three", ""));
4190 o
.push_back(new pg_nls_response_t
);
4191 o
.back()->handle
= hobject_t(object_t("hi"), "key", 5, 6, -1, "");
4192 o
.back()->entries
.push_back(librados::ListObjectImpl("", "one", ""));
4193 o
.back()->entries
.push_back(librados::ListObjectImpl("", "two", "twokey"));
4194 o
.back()->entries
.push_back(librados::ListObjectImpl("", "three", ""));
4195 o
.back()->entries
.push_back(librados::ListObjectImpl("n1", "n1one", ""));
4196 o
.back()->entries
.push_back(librados::ListObjectImpl("n1", "n1two", "n1twokey"));
4197 o
.back()->entries
.push_back(librados::ListObjectImpl("n1", "n1three", ""));
4201 WRITE_CLASS_ENCODER(pg_nls_response_t
)
4203 // For backwards compatibility with older OSD requests
4204 struct pg_ls_response_t
{
4205 collection_list_handle_t handle
;
4206 list
<pair
<object_t
, string
> > entries
;
4208 void encode(bufferlist
& bl
) const {
4211 ::encode(handle
, bl
);
4212 ::encode(entries
, bl
);
4214 void decode(bufferlist::iterator
& bl
) {
4218 ::decode(handle
, bl
);
4219 ::decode(entries
, bl
);
4221 void dump(Formatter
*f
) const {
4222 f
->dump_stream("handle") << handle
;
4223 f
->open_array_section("entries");
4224 for (list
<pair
<object_t
, string
> >::const_iterator p
= entries
.begin(); p
!= entries
.end(); ++p
) {
4225 f
->open_object_section("object");
4226 f
->dump_stream("object") << p
->first
;
4227 f
->dump_string("key", p
->second
);
4232 static void generate_test_instances(list
<pg_ls_response_t
*>& o
) {
4233 o
.push_back(new pg_ls_response_t
);
4234 o
.push_back(new pg_ls_response_t
);
4235 o
.back()->handle
= hobject_t(object_t("hi"), "key", 1, 2, -1, "");
4236 o
.back()->entries
.push_back(make_pair(object_t("one"), string()));
4237 o
.back()->entries
.push_back(make_pair(object_t("two"), string("twokey")));
4241 WRITE_CLASS_ENCODER(pg_ls_response_t
)
4244 * object_copy_cursor_t
4246 struct object_copy_cursor_t
{
4247 uint64_t data_offset
;
4253 object_copy_cursor_t()
4255 attr_complete(false),
4256 data_complete(false),
4257 omap_complete(false)
4260 bool is_initial() const {
4261 return !attr_complete
&& data_offset
== 0 && omap_offset
.empty();
4263 bool is_complete() const {
4264 return attr_complete
&& data_complete
&& omap_complete
;
4267 static void generate_test_instances(list
<object_copy_cursor_t
*>& o
);
4268 void encode(bufferlist
& bl
) const;
4269 void decode(bufferlist::iterator
&bl
);
4270 void dump(Formatter
*f
) const;
4272 WRITE_CLASS_ENCODER(object_copy_cursor_t
)
4275 * object_copy_data_t
4277 * Return data from a copy request. The semantics are a little strange
4278 * as a result of the encoding's heritage.
4280 * In particular, the sender unconditionally fills in the cursor (from what
4281 * it receives and sends), the size, and the mtime, but is responsible for
4282 * figuring out whether it should put any data in the attrs, data, or
4283 * omap members (corresponding to xattrs, object data, and the omap entries)
4284 * based on external data (the client includes a max amount to return with
4285 * the copy request). The client then looks into the attrs, data, and/or omap
4286 * based on the contents of the cursor.
4288 struct object_copy_data_t
{
4290 FLAG_DATA_DIGEST
= 1<<0,
4291 FLAG_OMAP_DIGEST
= 1<<1,
4293 object_copy_cursor_t cursor
;
4296 uint32_t data_digest
, omap_digest
;
4298 map
<string
, bufferlist
> attrs
;
4300 bufferlist omap_header
;
4301 bufferlist omap_data
;
4303 /// which snaps we are defined for (if a snap and not the head)
4304 vector
<snapid_t
> snaps
;
4305 ///< latest snap seq for the object (if head)
4308 ///< recent reqids on this object
4309 mempool::osd_pglog::vector
<pair
<osd_reqid_t
, version_t
> > reqids
;
4311 uint64_t truncate_seq
;
4312 uint64_t truncate_size
;
4315 object_copy_data_t() :
4316 size((uint64_t)-1), data_digest(-1),
4317 omap_digest(-1), flags(0),
4321 static void generate_test_instances(list
<object_copy_data_t
*>& o
);
4322 void encode(bufferlist
& bl
, uint64_t features
) const;
4323 void decode(bufferlist::iterator
& bl
);
4324 void dump(Formatter
*f
) const;
4326 WRITE_CLASS_ENCODER_FEATURES(object_copy_data_t
)
4331 struct pg_create_t
{
4332 epoch_t created
; // epoch pg created
4333 pg_t parent
; // split from parent (if != pg_t())
4337 : created(0), split_bits(0) {}
4338 pg_create_t(unsigned c
, pg_t p
, int s
)
4339 : created(c
), parent(p
), split_bits(s
) {}
4341 void encode(bufferlist
&bl
) const;
4342 void decode(bufferlist::iterator
&bl
);
4343 void dump(Formatter
*f
) const;
4344 static void generate_test_instances(list
<pg_create_t
*>& o
);
4346 WRITE_CLASS_ENCODER(pg_create_t
)
4348 // -----------------------------------------
4350 struct osd_peer_stat_t
{
4353 osd_peer_stat_t() { }
4355 void encode(bufferlist
&bl
) const;
4356 void decode(bufferlist::iterator
&bl
);
4357 void dump(Formatter
*f
) const;
4358 static void generate_test_instances(list
<osd_peer_stat_t
*>& o
);
4360 WRITE_CLASS_ENCODER(osd_peer_stat_t
)
4362 ostream
& operator<<(ostream
& out
, const osd_peer_stat_t
&stat
);
4365 // -----------------------------------------
4367 class ObjectExtent
{
4369 * ObjectExtents are used for specifying IO behavior against RADOS
4370 * objects when one is using the ObjectCacher.
4372 * To use this in a real system, *every member* must be filled
4373 * out correctly. In particular, make sure to initialize the
4374 * oloc correctly, as its default values are deliberate poison
4375 * and will cause internal ObjectCacher asserts.
4377 * Similarly, your buffer_extents vector *must* specify a total
4378 * size equal to your length. If the buffer_extents inadvertently
4379 * contain less space than the length member specifies, you
4380 * will get unintelligible asserts deep in the ObjectCacher.
4382 * If you are trying to do testing and don't care about actual
4383 * RADOS function, the simplest thing to do is to initialize
4384 * the ObjectExtent (truncate_size can be 0), create a single entry
4385 * in buffer_extents matching the length, and set oloc.pool to 0.
4388 object_t oid
; // object id
4390 uint64_t offset
; // in object
4391 uint64_t length
; // in object
4392 uint64_t truncate_size
; // in object
4394 object_locator_t oloc
; // object locator (pool etc)
4396 vector
<pair
<uint64_t,uint64_t> > buffer_extents
; // off -> len. extents in buffer being mapped (may be fragmented bc of striping!)
4398 ObjectExtent() : objectno(0), offset(0), length(0), truncate_size(0) {}
4399 ObjectExtent(object_t o
, uint64_t ono
, uint64_t off
, uint64_t l
, uint64_t ts
) :
4400 oid(o
), objectno(ono
), offset(off
), length(l
), truncate_size(ts
) { }
4403 inline ostream
& operator<<(ostream
& out
, const ObjectExtent
&ex
)
4405 return out
<< "extent("
4406 << ex
.oid
<< " (" << ex
.objectno
<< ") in " << ex
.oloc
4407 << " " << ex
.offset
<< "~" << ex
.length
4408 << " -> " << ex
.buffer_extents
4413 // ---------------------------------------
4415 class OSDSuperblock
{
4417 uuid_d cluster_fsid
, osd_fsid
;
4418 int32_t whoami
; // my role in this fs.
4419 epoch_t current_epoch
; // most recent epoch
4420 epoch_t oldest_map
, newest_map
; // oldest/newest maps we have.
4423 CompatSet compat_features
;
4425 // last interval over which i mounted and was then active
4426 epoch_t mounted
; // last epoch i mounted
4427 epoch_t clean_thru
; // epoch i was active and clean thru
4431 current_epoch(0), oldest_map(0), newest_map(0), weight(0),
4432 mounted(0), clean_thru(0) {
4435 void encode(bufferlist
&bl
) const;
4436 void decode(bufferlist::iterator
&bl
);
4437 void dump(Formatter
*f
) const;
4438 static void generate_test_instances(list
<OSDSuperblock
*>& o
);
4440 WRITE_CLASS_ENCODER(OSDSuperblock
)
4442 inline ostream
& operator<<(ostream
& out
, const OSDSuperblock
& sb
)
4444 return out
<< "sb(" << sb
.cluster_fsid
4445 << " osd." << sb
.whoami
4446 << " " << sb
.osd_fsid
4447 << " e" << sb
.current_epoch
4448 << " [" << sb
.oldest_map
<< "," << sb
.newest_map
<< "]"
4449 << " lci=[" << sb
.mounted
<< "," << sb
.clean_thru
<< "]"
4462 * attached to object head. describes most recent snap context, and
4463 * set of existing clones.
4468 vector
<snapid_t
> snaps
; // descending
4469 vector
<snapid_t
> clones
; // ascending
4470 map
<snapid_t
, interval_set
<uint64_t> > clone_overlap
; // overlap w/ next newest
4471 map
<snapid_t
, uint64_t> clone_size
;
4472 map
<snapid_t
, vector
<snapid_t
>> clone_snaps
; // descending
4474 SnapSet() : seq(0), head_exists(false) {}
4475 explicit SnapSet(bufferlist
& bl
) {
4476 bufferlist::iterator p
= bl
.begin();
4480 bool is_legacy() const {
4481 return clone_snaps
.size() < clones
.size() || !head_exists
;
4484 /// populate SnapSet from a librados::snap_set_t
4485 void from_snap_set(const librados::snap_set_t
& ss
, bool legacy
);
4487 /// get space accounted to clone
4488 uint64_t get_clone_bytes(snapid_t clone
) const;
4490 void encode(bufferlist
& bl
) const;
4491 void decode(bufferlist::iterator
& bl
);
4492 void dump(Formatter
*f
) const;
4493 static void generate_test_instances(list
<SnapSet
*>& o
);
4495 SnapContext
get_ssc_as_of(snapid_t as_of
) const {
4498 for (vector
<snapid_t
>::const_iterator i
= snaps
.begin();
4502 out
.snaps
.push_back(*i
);
4507 // return min element of snaps > after, return max if no such element
4508 snapid_t
get_first_snap_after(snapid_t after
, snapid_t max
) const {
4509 for (vector
<snapid_t
>::const_reverse_iterator i
= snaps
.rbegin();
4518 SnapSet
get_filtered(const pg_pool_t
&pinfo
) const;
4519 void filter(const pg_pool_t
&pinfo
);
4521 WRITE_CLASS_ENCODER(SnapSet
)
4523 ostream
& operator<<(ostream
& out
, const SnapSet
& cs
);
4528 #define SS_ATTR "snapset"
4530 struct watch_info_t
{
4532 uint32_t timeout_seconds
;
4535 watch_info_t() : cookie(0), timeout_seconds(0) { }
4536 watch_info_t(uint64_t c
, uint32_t t
, const entity_addr_t
& a
) : cookie(c
), timeout_seconds(t
), addr(a
) {}
4538 void encode(bufferlist
& bl
, uint64_t features
) const;
4539 void decode(bufferlist::iterator
& bl
);
4540 void dump(Formatter
*f
) const;
4541 static void generate_test_instances(list
<watch_info_t
*>& o
);
4543 WRITE_CLASS_ENCODER_FEATURES(watch_info_t
)
4545 static inline bool operator==(const watch_info_t
& l
, const watch_info_t
& r
) {
4546 return l
.cookie
== r
.cookie
&& l
.timeout_seconds
== r
.timeout_seconds
4547 && l
.addr
== r
.addr
;
4550 static inline ostream
& operator<<(ostream
& out
, const watch_info_t
& w
) {
4551 return out
<< "watch(cookie " << w
.cookie
<< " " << w
.timeout_seconds
<< "s"
4552 << " " << w
.addr
<< ")";
4555 struct notify_info_t
{
4562 static inline ostream
& operator<<(ostream
& out
, const notify_info_t
& n
) {
4563 return out
<< "notify(cookie " << n
.cookie
4564 << " notify" << n
.notify_id
4565 << " " << n
.timeout
<< "s)";
4568 struct object_info_t
;
4569 struct object_manifest_t
{
4572 TYPE_REDIRECT
= 1, // start with this
4573 TYPE_CHUNKED
= 2, // do this later
4575 uint8_t type
; // redirect, chunked, ...
4576 hobject_t redirect_target
;
4578 object_manifest_t() : type(0) { }
4579 object_manifest_t(uint8_t type
, const hobject_t
& redirect_target
)
4580 : type(type
), redirect_target(redirect_target
) { }
4582 bool is_empty() const {
4583 return type
== TYPE_NONE
;
4585 bool is_redirect() const {
4586 return type
== TYPE_REDIRECT
;
4588 bool is_chunked() const {
4589 return type
== TYPE_CHUNKED
;
4591 static const char *get_type_name(uint8_t m
) {
4593 case TYPE_NONE
: return "none";
4594 case TYPE_REDIRECT
: return "redirect";
4595 case TYPE_CHUNKED
: return "chunked";
4596 default: return "unknown";
4599 const char *get_type_name() const {
4600 return get_type_name(type
);
4602 static void generate_test_instances(list
<object_manifest_t
*>& o
);
4603 void encode(bufferlist
&bl
) const;
4604 void decode(bufferlist::iterator
&bl
);
4605 void dump(Formatter
*f
) const;
4606 friend ostream
& operator<<(ostream
& out
, const object_info_t
& oi
);
4608 WRITE_CLASS_ENCODER(object_manifest_t
)
4609 ostream
& operator<<(ostream
& out
, const object_manifest_t
& oi
);
4611 struct object_info_t
{
4613 eversion_t version
, prior_version
;
4614 version_t user_version
;
4615 osd_reqid_t last_reqid
;
4619 utime_t local_mtime
; // local mtime
4621 // note: these are currently encoded into a total 16 bits; see
4622 // encode()/decode() for the weirdness.
4625 FLAG_WHITEOUT
= 1<<1, // object logically does not exist
4626 FLAG_DIRTY
= 1<<2, // object has been modified since last flushed or undirtied
4627 FLAG_OMAP
= 1 << 3, // has (or may have) some/any omap data
4628 FLAG_DATA_DIGEST
= 1 << 4, // has data crc
4629 FLAG_OMAP_DIGEST
= 1 << 5, // has omap crc
4630 FLAG_CACHE_PIN
= 1 << 6, // pin the object in cache tier
4631 FLAG_MANIFEST
= 1 << 7, // has manifest
4633 FLAG_USES_TMAP
= 1<<8, // deprecated; no longer used.
4638 static string
get_flag_string(flag_t flags
) {
4640 if (flags
& FLAG_LOST
)
4642 if (flags
& FLAG_WHITEOUT
)
4644 if (flags
& FLAG_DIRTY
)
4646 if (flags
& FLAG_USES_TMAP
)
4648 if (flags
& FLAG_OMAP
)
4650 if (flags
& FLAG_DATA_DIGEST
)
4651 s
+= "|data_digest";
4652 if (flags
& FLAG_OMAP_DIGEST
)
4653 s
+= "|omap_digest";
4654 if (flags
& FLAG_CACHE_PIN
)
4656 if (flags
& FLAG_MANIFEST
)
4662 string
get_flag_string() const {
4663 return get_flag_string(flags
);
4666 /// [clone] descending. pre-luminous; moved to SnapSet
4667 vector
<snapid_t
> legacy_snaps
;
4669 uint64_t truncate_seq
, truncate_size
;
4671 map
<pair
<uint64_t, entity_name_t
>, watch_info_t
> watchers
;
4673 // opportunistic checksums; may or may not be present
4674 __u32 data_digest
; ///< data crc32c
4675 __u32 omap_digest
; ///< omap crc32c
4677 // alloc hint attribute
4678 uint64_t expected_object_size
, expected_write_size
;
4679 uint32_t alloc_hint_flags
;
4681 struct object_manifest_t manifest
;
4683 void copy_user_bits(const object_info_t
& other
);
4685 static ps_t
legacy_object_locator_to_ps(const object_t
&oid
,
4686 const object_locator_t
&loc
);
4688 bool test_flag(flag_t f
) const {
4689 return (flags
& f
) == f
;
4691 void set_flag(flag_t f
) {
4692 flags
= (flag_t
)(flags
| f
);
4694 void clear_flag(flag_t f
) {
4695 flags
= (flag_t
)(flags
& ~f
);
4697 bool is_lost() const {
4698 return test_flag(FLAG_LOST
);
4700 bool is_whiteout() const {
4701 return test_flag(FLAG_WHITEOUT
);
4703 bool is_dirty() const {
4704 return test_flag(FLAG_DIRTY
);
4706 bool is_omap() const {
4707 return test_flag(FLAG_OMAP
);
4709 bool is_data_digest() const {
4710 return test_flag(FLAG_DATA_DIGEST
);
4712 bool is_omap_digest() const {
4713 return test_flag(FLAG_OMAP_DIGEST
);
4715 bool is_cache_pinned() const {
4716 return test_flag(FLAG_CACHE_PIN
);
4718 bool has_manifest() const {
4719 return test_flag(FLAG_MANIFEST
);
4722 void set_data_digest(__u32 d
) {
4723 set_flag(FLAG_DATA_DIGEST
);
4726 void set_omap_digest(__u32 d
) {
4727 set_flag(FLAG_OMAP_DIGEST
);
4730 void clear_data_digest() {
4731 clear_flag(FLAG_DATA_DIGEST
);
4734 void clear_omap_digest() {
4735 clear_flag(FLAG_OMAP_DIGEST
);
4739 set_data_digest(-1);
4740 set_omap_digest(-1);
4743 void encode(bufferlist
& bl
, uint64_t features
) const;
4744 void decode(bufferlist::iterator
& bl
);
4745 void decode(bufferlist
& bl
) {
4746 bufferlist::iterator p
= bl
.begin();
4749 void dump(Formatter
*f
) const;
4750 static void generate_test_instances(list
<object_info_t
*>& o
);
4752 explicit object_info_t()
4753 : user_version(0), size(0), flags((flag_t
)0),
4754 truncate_seq(0), truncate_size(0),
4755 data_digest(-1), omap_digest(-1),
4756 expected_object_size(0), expected_write_size(0),
4760 explicit object_info_t(const hobject_t
& s
)
4762 user_version(0), size(0), flags((flag_t
)0),
4763 truncate_seq(0), truncate_size(0),
4764 data_digest(-1), omap_digest(-1),
4765 expected_object_size(0), expected_write_size(0),
4769 explicit object_info_t(bufferlist
& bl
) {
4773 WRITE_CLASS_ENCODER_FEATURES(object_info_t
)
4775 ostream
& operator<<(ostream
& out
, const object_info_t
& oi
);
4780 struct ObjectRecoveryInfo
{
4785 SnapSet ss
; // only populated if soid is_snap()
4786 interval_set
<uint64_t> copy_subset
;
4787 map
<hobject_t
, interval_set
<uint64_t>> clone_subset
;
4789 ObjectRecoveryInfo() : size(0) { }
4791 static void generate_test_instances(list
<ObjectRecoveryInfo
*>& o
);
4792 void encode(bufferlist
&bl
, uint64_t features
) const;
4793 void decode(bufferlist::iterator
&bl
, int64_t pool
= -1);
4794 ostream
&print(ostream
&out
) const;
4795 void dump(Formatter
*f
) const;
4797 WRITE_CLASS_ENCODER_FEATURES(ObjectRecoveryInfo
)
4798 ostream
& operator<<(ostream
& out
, const ObjectRecoveryInfo
&inf
);
4800 struct ObjectRecoveryProgress
{
4801 uint64_t data_recovered_to
;
4802 string omap_recovered_to
;
4808 ObjectRecoveryProgress()
4809 : data_recovered_to(0),
4811 data_complete(false), omap_complete(false) { }
4813 bool is_complete(const ObjectRecoveryInfo
& info
) const {
4814 return (data_recovered_to
>= (
4815 info
.copy_subset
.empty() ?
4816 0 : info
.copy_subset
.range_end())) &&
4820 static void generate_test_instances(list
<ObjectRecoveryProgress
*>& o
);
4821 void encode(bufferlist
&bl
) const;
4822 void decode(bufferlist::iterator
&bl
);
4823 ostream
&print(ostream
&out
) const;
4824 void dump(Formatter
*f
) const;
4826 WRITE_CLASS_ENCODER(ObjectRecoveryProgress
)
4827 ostream
& operator<<(ostream
& out
, const ObjectRecoveryProgress
&prog
);
4829 struct PushReplyOp
{
4832 static void generate_test_instances(list
<PushReplyOp
*>& o
);
4833 void encode(bufferlist
&bl
) const;
4834 void decode(bufferlist::iterator
&bl
);
4835 ostream
&print(ostream
&out
) const;
4836 void dump(Formatter
*f
) const;
4838 uint64_t cost(CephContext
*cct
) const;
4840 WRITE_CLASS_ENCODER(PushReplyOp
)
4841 ostream
& operator<<(ostream
& out
, const PushReplyOp
&op
);
4846 ObjectRecoveryInfo recovery_info
;
4847 ObjectRecoveryProgress recovery_progress
;
4849 static void generate_test_instances(list
<PullOp
*>& o
);
4850 void encode(bufferlist
&bl
, uint64_t features
) const;
4851 void decode(bufferlist::iterator
&bl
);
4852 ostream
&print(ostream
&out
) const;
4853 void dump(Formatter
*f
) const;
4855 uint64_t cost(CephContext
*cct
) const;
4857 WRITE_CLASS_ENCODER_FEATURES(PullOp
)
4858 ostream
& operator<<(ostream
& out
, const PullOp
&op
);
4864 interval_set
<uint64_t> data_included
;
4865 bufferlist omap_header
;
4866 map
<string
, bufferlist
> omap_entries
;
4867 map
<string
, bufferlist
> attrset
;
4869 ObjectRecoveryInfo recovery_info
;
4870 ObjectRecoveryProgress before_progress
;
4871 ObjectRecoveryProgress after_progress
;
4873 static void generate_test_instances(list
<PushOp
*>& o
);
4874 void encode(bufferlist
&bl
, uint64_t features
) const;
4875 void decode(bufferlist::iterator
&bl
);
4876 ostream
&print(ostream
&out
) const;
4877 void dump(Formatter
*f
) const;
4879 uint64_t cost(CephContext
*cct
) const;
4881 WRITE_CLASS_ENCODER_FEATURES(PushOp
)
4882 ostream
& operator<<(ostream
& out
, const PushOp
&op
);
4886 * summarize pg contents for purposes of a scrub
4890 map
<string
,bufferptr
> attrs
;
4892 __u32 omap_digest
; ///< omap crc32c
4893 __u32 digest
; ///< data crc32c
4895 bool digest_present
:1;
4896 bool omap_digest_present
:1;
4899 bool ec_hash_mismatch
:1;
4900 bool ec_size_mismatch
:1;
4903 // Init invalid size so it won't match if we get a stat EIO error
4904 size(-1), omap_digest(0), digest(0),
4905 negative(false), digest_present(false), omap_digest_present(false),
4906 read_error(false), stat_error(false), ec_hash_mismatch(false), ec_size_mismatch(false) {}
4908 void encode(bufferlist
& bl
) const;
4909 void decode(bufferlist::iterator
& bl
);
4910 void dump(Formatter
*f
) const;
4911 static void generate_test_instances(list
<object
*>& o
);
4913 WRITE_CLASS_ENCODER(object
)
4915 map
<hobject_t
,object
> objects
;
4916 eversion_t valid_through
;
4917 eversion_t incr_since
;
4919 void merge_incr(const ScrubMap
&l
);
4920 void insert(const ScrubMap
&r
) {
4921 objects
.insert(r
.objects
.begin(), r
.objects
.end());
4923 void swap(ScrubMap
&r
) {
4925 swap(objects
, r
.objects
);
4926 swap(valid_through
, r
.valid_through
);
4927 swap(incr_since
, r
.incr_since
);
4930 void encode(bufferlist
& bl
) const;
4931 void decode(bufferlist::iterator
& bl
, int64_t pool
=-1);
4932 void dump(Formatter
*f
) const;
4933 static void generate_test_instances(list
<ScrubMap
*>& o
);
4935 WRITE_CLASS_ENCODER(ScrubMap::object
)
4936 WRITE_CLASS_ENCODER(ScrubMap
)
4942 bufferlist indata
, outdata
;
4946 memset(&op
, 0, sizeof(ceph_osd_op
));
4950 * split a bufferlist into constituent indata members of a vector of OSDOps
4952 * @param ops [out] vector of OSDOps
4953 * @param in [in] combined data buffer
4955 static void split_osd_op_vector_in_data(vector
<OSDOp
>& ops
, bufferlist
& in
);
4958 * merge indata members of a vector of OSDOp into a single bufferlist
4960 * Notably this also encodes certain other OSDOp data into the data
4961 * buffer, including the sobject_t soid.
4963 * @param ops [in] vector of OSDOps
4964 * @param out [out] combined data buffer
4966 static void merge_osd_op_vector_in_data(vector
<OSDOp
>& ops
, bufferlist
& out
);
4969 * split a bufferlist into constituent outdata members of a vector of OSDOps
4971 * @param ops [out] vector of OSDOps
4972 * @param in [in] combined data buffer
4974 static void split_osd_op_vector_out_data(vector
<OSDOp
>& ops
, bufferlist
& in
);
4977 * merge outdata members of a vector of OSDOps into a single bufferlist
4979 * @param ops [in] vector of OSDOps
4980 * @param out [out] combined data buffer
4982 static void merge_osd_op_vector_out_data(vector
<OSDOp
>& ops
, bufferlist
& out
);
4985 * Clear data as much as possible, leave minimal data for historical op dump
4987 * @param ops [in] vector of OSDOps
4989 static void clear_data(vector
<OSDOp
>& ops
);
4992 ostream
& operator<<(ostream
& out
, const OSDOp
& op
);
4994 struct watch_item_t
{
4997 uint32_t timeout_seconds
;
5000 watch_item_t() : cookie(0), timeout_seconds(0) { }
5001 watch_item_t(entity_name_t name
, uint64_t cookie
, uint32_t timeout
,
5002 const entity_addr_t
& addr
)
5003 : name(name
), cookie(cookie
), timeout_seconds(timeout
),
5006 void encode(bufferlist
&bl
, uint64_t features
) const {
5007 ENCODE_START(2, 1, bl
);
5009 ::encode(cookie
, bl
);
5010 ::encode(timeout_seconds
, bl
);
5011 ::encode(addr
, bl
, features
);
5014 void decode(bufferlist::iterator
&bl
) {
5015 DECODE_START(2, bl
);
5017 ::decode(cookie
, bl
);
5018 ::decode(timeout_seconds
, bl
);
5019 if (struct_v
>= 2) {
5025 WRITE_CLASS_ENCODER_FEATURES(watch_item_t
)
5027 struct obj_watch_item_t
{
5033 * obj list watch response format
5036 struct obj_list_watch_response_t
{
5037 list
<watch_item_t
> entries
;
5039 void encode(bufferlist
& bl
, uint64_t features
) const {
5040 ENCODE_START(1, 1, bl
);
5041 ::encode(entries
, bl
, features
);
5044 void decode(bufferlist::iterator
& bl
) {
5045 DECODE_START(1, bl
);
5046 ::decode(entries
, bl
);
5049 void dump(Formatter
*f
) const {
5050 f
->open_array_section("entries");
5051 for (list
<watch_item_t
>::const_iterator p
= entries
.begin(); p
!= entries
.end(); ++p
) {
5052 f
->open_object_section("watch");
5053 f
->dump_stream("watcher") << p
->name
;
5054 f
->dump_int("cookie", p
->cookie
);
5055 f
->dump_int("timeout", p
->timeout_seconds
);
5056 f
->open_object_section("addr");
5063 static void generate_test_instances(list
<obj_list_watch_response_t
*>& o
) {
5065 o
.push_back(new obj_list_watch_response_t
);
5066 o
.push_back(new obj_list_watch_response_t
);
5067 ea
.set_type(entity_addr_t::TYPE_LEGACY
);
5069 ea
.set_family(AF_INET
);
5070 ea
.set_in4_quad(0, 127);
5071 ea
.set_in4_quad(1, 0);
5072 ea
.set_in4_quad(2, 0);
5073 ea
.set_in4_quad(3, 1);
5075 o
.back()->entries
.push_back(watch_item_t(entity_name_t(entity_name_t::TYPE_CLIENT
, 1), 10, 30, ea
));
5077 ea
.set_in4_quad(3, 2);
5079 o
.back()->entries
.push_back(watch_item_t(entity_name_t(entity_name_t::TYPE_CLIENT
, 2), 20, 60, ea
));
5082 WRITE_CLASS_ENCODER_FEATURES(obj_list_watch_response_t
)
5086 vector
<snapid_t
> snaps
; // ascending
5087 vector
< pair
<uint64_t,uint64_t> > overlap
;
5090 clone_info() : cloneid(CEPH_NOSNAP
), size(0) {}
5092 void encode(bufferlist
& bl
) const {
5093 ENCODE_START(1, 1, bl
);
5094 ::encode(cloneid
, bl
);
5095 ::encode(snaps
, bl
);
5096 ::encode(overlap
, bl
);
5100 void decode(bufferlist::iterator
& bl
) {
5101 DECODE_START(1, bl
);
5102 ::decode(cloneid
, bl
);
5103 ::decode(snaps
, bl
);
5104 ::decode(overlap
, bl
);
5108 void dump(Formatter
*f
) const {
5109 if (cloneid
== CEPH_NOSNAP
)
5110 f
->dump_string("cloneid", "HEAD");
5112 f
->dump_unsigned("cloneid", cloneid
.val
);
5113 f
->open_array_section("snapshots");
5114 for (vector
<snapid_t
>::const_iterator p
= snaps
.begin(); p
!= snaps
.end(); ++p
) {
5115 f
->open_object_section("snap");
5116 f
->dump_unsigned("id", p
->val
);
5120 f
->open_array_section("overlaps");
5121 for (vector
< pair
<uint64_t,uint64_t> >::const_iterator q
= overlap
.begin();
5122 q
!= overlap
.end(); ++q
) {
5123 f
->open_object_section("overlap");
5124 f
->dump_unsigned("offset", q
->first
);
5125 f
->dump_unsigned("length", q
->second
);
5129 f
->dump_unsigned("size", size
);
5131 static void generate_test_instances(list
<clone_info
*>& o
) {
5132 o
.push_back(new clone_info
);
5133 o
.push_back(new clone_info
);
5134 o
.back()->cloneid
= 1;
5135 o
.back()->snaps
.push_back(1);
5136 o
.back()->overlap
.push_back(pair
<uint64_t,uint64_t>(0,4096));
5137 o
.back()->overlap
.push_back(pair
<uint64_t,uint64_t>(8192,4096));
5138 o
.back()->size
= 16384;
5139 o
.push_back(new clone_info
);
5140 o
.back()->cloneid
= CEPH_NOSNAP
;
5141 o
.back()->size
= 32768;
5144 WRITE_CLASS_ENCODER(clone_info
)
5147 * obj list snaps response format
5150 struct obj_list_snap_response_t
{
5151 vector
<clone_info
> clones
; // ascending
5154 void encode(bufferlist
& bl
) const {
5155 ENCODE_START(2, 1, bl
);
5156 ::encode(clones
, bl
);
5160 void decode(bufferlist::iterator
& bl
) {
5161 DECODE_START(2, bl
);
5162 ::decode(clones
, bl
);
5169 void dump(Formatter
*f
) const {
5170 f
->open_array_section("clones");
5171 for (vector
<clone_info
>::const_iterator p
= clones
.begin(); p
!= clones
.end(); ++p
) {
5172 f
->open_object_section("clone");
5176 f
->dump_unsigned("seq", seq
);
5179 static void generate_test_instances(list
<obj_list_snap_response_t
*>& o
) {
5180 o
.push_back(new obj_list_snap_response_t
);
5181 o
.push_back(new obj_list_snap_response_t
);
5184 cl
.snaps
.push_back(1);
5185 cl
.overlap
.push_back(pair
<uint64_t,uint64_t>(0,4096));
5186 cl
.overlap
.push_back(pair
<uint64_t,uint64_t>(8192,4096));
5188 o
.back()->clones
.push_back(cl
);
5189 cl
.cloneid
= CEPH_NOSNAP
;
5193 o
.back()->clones
.push_back(cl
);
5194 o
.back()->seq
= 123;
5198 WRITE_CLASS_ENCODER(obj_list_snap_response_t
)
5202 struct PromoteCounter
{
5203 std::atomic_ullong attempts
{0};
5204 std::atomic_ullong objects
{0};
5205 std::atomic_ullong bytes
{0};
5211 void finish(uint64_t size
) {
5216 void sample_and_attenuate(uint64_t *a
, uint64_t *o
, uint64_t *b
) {
5227 * ObjectStore full statfs information
5229 struct store_statfs_t
5231 uint64_t total
= 0; // Total bytes
5232 uint64_t available
= 0; // Free bytes available
5234 int64_t allocated
= 0; // Bytes allocated by the store
5235 int64_t stored
= 0; // Bytes actually stored by the user
5236 int64_t compressed
= 0; // Bytes stored after compression
5237 int64_t compressed_allocated
= 0; // Bytes allocated for compressed data
5238 int64_t compressed_original
= 0; // Bytes that were successfully compressed
5241 *this = store_statfs_t();
5243 bool operator ==(const store_statfs_t
& other
) const;
5244 void dump(Formatter
*f
) const;
5246 ostream
&operator<<(ostream
&lhs
, const store_statfs_t
&rhs
);