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 osd_stat_t() : kb(0), kb_used(0), kb_avail(0),
908 snap_trim_queue_len(0), num_snap_trimming(0) {}
910 void add(const osd_stat_t
& o
) {
912 kb_used
+= o
.kb_used
;
913 kb_avail
+= o
.kb_avail
;
914 snap_trim_queue_len
+= o
.snap_trim_queue_len
;
915 num_snap_trimming
+= o
.num_snap_trimming
;
916 op_queue_age_hist
.add(o
.op_queue_age_hist
);
917 os_perf_stat
.add(o
.os_perf_stat
);
919 void sub(const osd_stat_t
& o
) {
921 kb_used
-= o
.kb_used
;
922 kb_avail
-= o
.kb_avail
;
923 snap_trim_queue_len
-= o
.snap_trim_queue_len
;
924 num_snap_trimming
-= o
.num_snap_trimming
;
925 op_queue_age_hist
.sub(o
.op_queue_age_hist
);
926 os_perf_stat
.sub(o
.os_perf_stat
);
929 void dump(Formatter
*f
) const;
930 void encode(bufferlist
&bl
) const;
931 void decode(bufferlist::iterator
&bl
);
932 static void generate_test_instances(std::list
<osd_stat_t
*>& o
);
934 WRITE_CLASS_ENCODER(osd_stat_t
)
936 inline bool operator==(const osd_stat_t
& l
, const osd_stat_t
& r
) {
937 return l
.kb
== r
.kb
&&
938 l
.kb_used
== r
.kb_used
&&
939 l
.kb_avail
== r
.kb_avail
&&
940 l
.snap_trim_queue_len
== r
.snap_trim_queue_len
&&
941 l
.num_snap_trimming
== r
.num_snap_trimming
&&
942 l
.hb_peers
== r
.hb_peers
&&
943 l
.op_queue_age_hist
== r
.op_queue_age_hist
&&
944 l
.os_perf_stat
== r
.os_perf_stat
;
946 inline bool operator!=(const osd_stat_t
& l
, const osd_stat_t
& r
) {
952 inline ostream
& operator<<(ostream
& out
, const osd_stat_t
& s
) {
953 return out
<< "osd_stat(" << kb_t(s
.kb_used
) << " used, "
954 << kb_t(s
.kb_avail
) << " avail, "
955 << kb_t(s
.kb
) << " total, "
956 << "peers " << s
.hb_peers
957 << " op hist " << s
.op_queue_age_hist
.h
965 #define PG_STATE_CREATING (1<<0) // creating
966 #define PG_STATE_ACTIVE (1<<1) // i am active. (primary: replicas too)
967 #define PG_STATE_CLEAN (1<<2) // peers are complete, clean of stray replicas.
968 #define PG_STATE_DOWN (1<<4) // a needed replica is down, PG offline
969 //#define PG_STATE_REPLAY (1<<5) // crashed, waiting for replay
970 //#define PG_STATE_STRAY (1<<6) // i must notify the primary i exist.
971 //#define PG_STATE_SPLITTING (1<<7) // i am splitting
972 #define PG_STATE_SCRUBBING (1<<8) // scrubbing
973 //#define PG_STATE_SCRUBQ (1<<9) // queued for scrub
974 #define PG_STATE_DEGRADED (1<<10) // pg contains objects with reduced redundancy
975 #define PG_STATE_INCONSISTENT (1<<11) // pg replicas are inconsistent (but shouldn't be)
976 #define PG_STATE_PEERING (1<<12) // pg is (re)peering
977 #define PG_STATE_REPAIR (1<<13) // pg should repair on next scrub
978 #define PG_STATE_RECOVERING (1<<14) // pg is recovering/migrating objects
979 #define PG_STATE_BACKFILL_WAIT (1<<15) // [active] reserving backfill
980 #define PG_STATE_INCOMPLETE (1<<16) // incomplete content, peering failed.
981 #define PG_STATE_STALE (1<<17) // our state for this pg is stale, unknown.
982 #define PG_STATE_REMAPPED (1<<18) // pg is explicitly remapped to different OSDs than CRUSH
983 #define PG_STATE_DEEP_SCRUB (1<<19) // deep scrub: check CRC32 on files
984 #define PG_STATE_BACKFILL (1<<20) // [active] backfilling pg content
985 #define PG_STATE_BACKFILL_TOOFULL (1<<21) // backfill can't proceed: too full
986 #define PG_STATE_RECOVERY_WAIT (1<<22) // waiting for recovery reservations
987 #define PG_STATE_UNDERSIZED (1<<23) // pg acting < pool size
988 #define PG_STATE_ACTIVATING (1<<24) // pg is peered but not yet active
989 #define PG_STATE_PEERED (1<<25) // peered, cannot go active, can recover
990 #define PG_STATE_SNAPTRIM (1<<26) // trimming snaps
991 #define PG_STATE_SNAPTRIM_WAIT (1<<27) // queued to trim snaps
992 #define PG_STATE_RECOVERY_TOOFULL (1<<28) // recovery can't proceed: too full
993 #define PG_STATE_SNAPTRIM_ERROR (1<<29) // error stopped trimming snaps
994 #define PG_STATE_FORCED_RECOVERY (1<<30) // force recovery of this pg before any other
995 #define PG_STATE_FORCED_BACKFILL (1<<31) // force backfill of this pg before any other
997 std::string
pg_state_string(int state
);
998 std::string
pg_vector_string(const vector
<int32_t> &a
);
999 int pg_string_state(const std::string
& state
);
1005 * attributes for a single pool snapshot.
1007 struct pool_snap_info_t
{
1012 void dump(Formatter
*f
) const;
1013 void encode(bufferlist
& bl
, uint64_t features
) const;
1014 void decode(bufferlist::iterator
& bl
);
1015 static void generate_test_instances(list
<pool_snap_info_t
*>& o
);
1017 WRITE_CLASS_ENCODER_FEATURES(pool_snap_info_t
)
1019 inline ostream
& operator<<(ostream
& out
, const pool_snap_info_t
& si
) {
1020 return out
<< si
.snapid
<< '(' << si
.name
<< ' ' << si
.stamp
<< ')';
1035 DEEP_SCRUB_INTERVAL
,
1037 RECOVERY_OP_PRIORITY
,
1040 COMPRESSION_ALGORITHM
,
1041 COMPRESSION_REQUIRED_RATIO
,
1042 COMPRESSION_MAX_BLOB_SIZE
,
1043 COMPRESSION_MIN_BLOB_SIZE
,
1059 opt_desc_t(key_t k
, type_t t
) : key(k
), type(t
) {}
1061 bool operator==(const opt_desc_t
& rhs
) const {
1062 return key
== rhs
.key
&& type
== rhs
.type
;
1066 typedef boost::variant
<std::string
,int,double> value_t
;
1068 static bool is_opt_name(const std::string
& name
);
1069 static opt_desc_t
get_opt_desc(const std::string
& name
);
1071 pool_opts_t() : opts() {}
1073 bool is_set(key_t key
) const;
1075 template<typename T
>
1076 void set(key_t key
, const T
&val
) {
1077 value_t value
= val
;
1081 template<typename T
>
1082 bool get(key_t key
, T
*val
) const {
1083 opts_t::const_iterator i
= opts
.find(key
);
1084 if (i
== opts
.end()) {
1087 *val
= boost::get
<T
>(i
->second
);
1091 const value_t
& get(key_t key
) const;
1093 bool unset(key_t key
);
1095 void dump(const std::string
& name
, Formatter
*f
) const;
1097 void dump(Formatter
*f
) const;
1098 void encode(bufferlist
&bl
) const;
1099 void decode(bufferlist::iterator
&bl
);
1102 typedef std::map
<key_t
, value_t
> opts_t
;
1105 friend ostream
& operator<<(ostream
& out
, const pool_opts_t
& opts
);
1107 WRITE_CLASS_ENCODER(pool_opts_t
)
1113 static const char *APPLICATION_NAME_CEPHFS
;
1114 static const char *APPLICATION_NAME_RBD
;
1115 static const char *APPLICATION_NAME_RGW
;
1118 TYPE_REPLICATED
= 1, // replication
1119 //TYPE_RAID4 = 2, // raid4 (never implemented)
1120 TYPE_ERASURE
= 3, // erasure-coded
1122 static const char *get_type_name(int t
) {
1124 case TYPE_REPLICATED
: return "replicated";
1125 //case TYPE_RAID4: return "raid4";
1126 case TYPE_ERASURE
: return "erasure";
1127 default: return "???";
1130 const char *get_type_name() const {
1131 return get_type_name(type
);
1135 FLAG_HASHPSPOOL
= 1<<0, // hash pg seed and pool together (instead of adding)
1136 FLAG_FULL
= 1<<1, // pool is full
1137 FLAG_EC_OVERWRITES
= 1<<2, // enables overwrites, once enabled, cannot be disabled
1138 FLAG_INCOMPLETE_CLONES
= 1<<3, // may have incomplete clones (bc we are/were an overlay)
1139 FLAG_NODELETE
= 1<<4, // pool can't be deleted
1140 FLAG_NOPGCHANGE
= 1<<5, // pool's pg and pgp num can't be changed
1141 FLAG_NOSIZECHANGE
= 1<<6, // pool's size and min size can't be changed
1142 FLAG_WRITE_FADVISE_DONTNEED
= 1<<7, // write mode with LIBRADOS_OP_FLAG_FADVISE_DONTNEED
1143 FLAG_NOSCRUB
= 1<<8, // block periodic scrub
1144 FLAG_NODEEP_SCRUB
= 1<<9, // block periodic deep-scrub
1147 static const char *get_flag_name(int f
) {
1149 case FLAG_HASHPSPOOL
: return "hashpspool";
1150 case FLAG_FULL
: return "full";
1151 case FLAG_EC_OVERWRITES
: return "ec_overwrites";
1152 case FLAG_INCOMPLETE_CLONES
: return "incomplete_clones";
1153 case FLAG_NODELETE
: return "nodelete";
1154 case FLAG_NOPGCHANGE
: return "nopgchange";
1155 case FLAG_NOSIZECHANGE
: return "nosizechange";
1156 case FLAG_WRITE_FADVISE_DONTNEED
: return "write_fadvise_dontneed";
1157 case FLAG_NOSCRUB
: return "noscrub";
1158 case FLAG_NODEEP_SCRUB
: return "nodeep-scrub";
1159 default: return "???";
1162 static string
get_flags_string(uint64_t f
) {
1164 for (unsigned n
=0; f
&& n
<64; ++n
) {
1165 if (f
& (1ull << n
)) {
1168 s
+= get_flag_name(1ull << n
);
1173 string
get_flags_string() const {
1174 return get_flags_string(flags
);
1176 static uint64_t get_flag_by_name(const string
& name
) {
1177 if (name
== "hashpspool")
1178 return FLAG_HASHPSPOOL
;
1181 if (name
== "ec_overwrites")
1182 return FLAG_EC_OVERWRITES
;
1183 if (name
== "incomplete_clones")
1184 return FLAG_INCOMPLETE_CLONES
;
1185 if (name
== "nodelete")
1186 return FLAG_NODELETE
;
1187 if (name
== "nopgchange")
1188 return FLAG_NOPGCHANGE
;
1189 if (name
== "nosizechange")
1190 return FLAG_NOSIZECHANGE
;
1191 if (name
== "write_fadvise_dontneed")
1192 return FLAG_WRITE_FADVISE_DONTNEED
;
1193 if (name
== "noscrub")
1194 return FLAG_NOSCRUB
;
1195 if (name
== "nodeep-scrub")
1196 return FLAG_NODEEP_SCRUB
;
1200 /// converts the acting/up vector to a set of pg shards
1201 void convert_to_pg_shards(const vector
<int> &from
, set
<pg_shard_t
>* to
) const;
1204 CACHEMODE_NONE
= 0, ///< no caching
1205 CACHEMODE_WRITEBACK
= 1, ///< write to cache, flush later
1206 CACHEMODE_FORWARD
= 2, ///< forward if not in cache
1207 CACHEMODE_READONLY
= 3, ///< handle reads, forward writes [not strongly consistent]
1208 CACHEMODE_READFORWARD
= 4, ///< forward reads, write to cache flush later
1209 CACHEMODE_READPROXY
= 5, ///< proxy reads, write to cache flush later
1210 CACHEMODE_PROXY
= 6, ///< proxy if not in cache
1212 static const char *get_cache_mode_name(cache_mode_t m
) {
1214 case CACHEMODE_NONE
: return "none";
1215 case CACHEMODE_WRITEBACK
: return "writeback";
1216 case CACHEMODE_FORWARD
: return "forward";
1217 case CACHEMODE_READONLY
: return "readonly";
1218 case CACHEMODE_READFORWARD
: return "readforward";
1219 case CACHEMODE_READPROXY
: return "readproxy";
1220 case CACHEMODE_PROXY
: return "proxy";
1221 default: return "unknown";
1224 static cache_mode_t
get_cache_mode_from_str(const string
& s
) {
1226 return CACHEMODE_NONE
;
1227 if (s
== "writeback")
1228 return CACHEMODE_WRITEBACK
;
1230 return CACHEMODE_FORWARD
;
1231 if (s
== "readonly")
1232 return CACHEMODE_READONLY
;
1233 if (s
== "readforward")
1234 return CACHEMODE_READFORWARD
;
1235 if (s
== "readproxy")
1236 return CACHEMODE_READPROXY
;
1238 return CACHEMODE_PROXY
;
1239 return (cache_mode_t
)-1;
1241 const char *get_cache_mode_name() const {
1242 return get_cache_mode_name(cache_mode
);
1244 bool cache_mode_requires_hit_set() const {
1245 switch (cache_mode
) {
1246 case CACHEMODE_NONE
:
1247 case CACHEMODE_FORWARD
:
1248 case CACHEMODE_READONLY
:
1249 case CACHEMODE_PROXY
:
1251 case CACHEMODE_WRITEBACK
:
1252 case CACHEMODE_READFORWARD
:
1253 case CACHEMODE_READPROXY
:
1256 assert(0 == "implement me");
1260 uint64_t flags
; ///< FLAG_*
1261 __u8 type
; ///< TYPE_*
1262 __u8 size
, min_size
; ///< number of osds in each pg
1263 __u8 crush_rule
; ///< crush placement rule
1264 __u8 object_hash
; ///< hash mapping object name to ps
1266 __u32 pg_num
, pgp_num
; ///< number of pgs
1270 map
<string
,string
> properties
; ///< OBSOLETE
1271 string erasure_code_profile
; ///< name of the erasure code profile in OSDMap
1272 epoch_t last_change
; ///< most recent epoch changed, exclusing snapshot changes
1273 epoch_t last_force_op_resend
; ///< last epoch that forced clients to resend
1274 /// last epoch that forced clients to resend (pre-luminous clients only)
1275 epoch_t last_force_op_resend_preluminous
;
1276 snapid_t snap_seq
; ///< seq for per-pool snapshot
1277 epoch_t snap_epoch
; ///< osdmap epoch of last snap
1278 uint64_t auid
; ///< who owns the pg
1279 __u32 crash_replay_interval
; ///< seconds to allow clients to replay ACKed but unCOMMITted requests
1281 uint64_t quota_max_bytes
; ///< maximum number of bytes for this pool
1282 uint64_t quota_max_objects
; ///< maximum number of objects for this pool
1285 * Pool snaps (global to this pool). These define a SnapContext for
1286 * the pool, unless the client manually specifies an alternate
1289 map
<snapid_t
, pool_snap_info_t
> snaps
;
1291 * Alternatively, if we are defining non-pool snaps (e.g. via the
1292 * Ceph MDS), we must track @removed_snaps (since @snaps is not
1293 * used). Snaps and removed_snaps are to be used exclusive of each
1296 interval_set
<snapid_t
> removed_snaps
;
1298 unsigned pg_num_mask
, pgp_num_mask
;
1300 set
<uint64_t> tiers
; ///< pools that are tiers of us
1301 int64_t tier_of
; ///< pool for which we are a tier
1302 // Note that write wins for read+write ops
1303 int64_t read_tier
; ///< pool/tier for objecter to direct reads to
1304 int64_t write_tier
; ///< pool/tier for objecter to direct writes to
1305 cache_mode_t cache_mode
; ///< cache pool mode
1307 bool is_tier() const { return tier_of
>= 0; }
1308 bool has_tiers() const { return !tiers
.empty(); }
1313 clear_tier_tunables();
1315 bool has_read_tier() const { return read_tier
>= 0; }
1316 void clear_read_tier() { read_tier
= -1; }
1317 bool has_write_tier() const { return write_tier
>= 0; }
1318 void clear_write_tier() { write_tier
= -1; }
1319 void clear_tier_tunables() {
1320 if (cache_mode
!= CACHEMODE_NONE
)
1321 flags
|= FLAG_INCOMPLETE_CLONES
;
1322 cache_mode
= CACHEMODE_NONE
;
1324 target_max_bytes
= 0;
1325 target_max_objects
= 0;
1326 cache_target_dirty_ratio_micro
= 0;
1327 cache_target_dirty_high_ratio_micro
= 0;
1328 cache_target_full_ratio_micro
= 0;
1329 hit_set_params
= HitSet::Params();
1332 hit_set_grade_decay_rate
= 0;
1333 hit_set_search_last_n
= 0;
1334 grade_table
.resize(0);
1337 uint64_t target_max_bytes
; ///< tiering: target max pool size
1338 uint64_t target_max_objects
; ///< tiering: target max pool size
1340 uint32_t cache_target_dirty_ratio_micro
; ///< cache: fraction of target to leave dirty
1341 uint32_t cache_target_dirty_high_ratio_micro
; ///<cache: fraction of target to flush with high speed
1342 uint32_t cache_target_full_ratio_micro
; ///< cache: fraction of target to fill before we evict in earnest
1344 uint32_t cache_min_flush_age
; ///< minimum age (seconds) before we can flush
1345 uint32_t cache_min_evict_age
; ///< minimum age (seconds) before we can evict
1347 HitSet::Params hit_set_params
; ///< The HitSet params to use on this pool
1348 uint32_t hit_set_period
; ///< periodicity of HitSet segments (seconds)
1349 uint32_t hit_set_count
; ///< number of periods to retain
1350 bool use_gmt_hitset
; ///< use gmt to name the hitset archive object
1351 uint32_t min_read_recency_for_promote
; ///< minimum number of HitSet to check before promote on read
1352 uint32_t min_write_recency_for_promote
; ///< minimum number of HitSet to check before promote on write
1353 uint32_t hit_set_grade_decay_rate
; ///< current hit_set has highest priority on objects
1354 ///temperature count,the follow hit_set's priority decay
1355 ///by this params than pre hit_set
1356 uint32_t hit_set_search_last_n
; ///<accumulate atmost N hit_sets for temperature
1358 uint32_t stripe_width
; ///< erasure coded stripe size in bytes
1360 uint64_t expected_num_objects
; ///< expected number of objects on this pool, a value of 0 indicates
1361 ///< user does not specify any expected value
1362 bool fast_read
; ///< whether turn on fast read on the pool or not
1364 pool_opts_t opts
; ///< options
1366 /// application -> key/value metadata
1367 map
<string
, std::map
<string
, string
>> application_metadata
;
1370 vector
<uint32_t> grade_table
;
1373 uint32_t get_grade(unsigned i
) const {
1374 if (grade_table
.size() <= i
)
1376 return grade_table
[i
];
1378 void calc_grade_table() {
1379 unsigned v
= 1000000;
1380 grade_table
.resize(hit_set_count
);
1381 for (unsigned i
= 0; i
< hit_set_count
; i
++) {
1382 v
= v
* (1 - (hit_set_grade_decay_rate
/ 100.0));
1388 : flags(0), type(0), size(0), min_size(0),
1389 crush_rule(0), object_hash(0),
1390 pg_num(0), pgp_num(0),
1392 last_force_op_resend(0),
1393 last_force_op_resend_preluminous(0),
1394 snap_seq(0), snap_epoch(0),
1396 crash_replay_interval(0),
1397 quota_max_bytes(0), quota_max_objects(0),
1398 pg_num_mask(0), pgp_num_mask(0),
1399 tier_of(-1), read_tier(-1), write_tier(-1),
1400 cache_mode(CACHEMODE_NONE
),
1401 target_max_bytes(0), target_max_objects(0),
1402 cache_target_dirty_ratio_micro(0),
1403 cache_target_dirty_high_ratio_micro(0),
1404 cache_target_full_ratio_micro(0),
1405 cache_min_flush_age(0),
1406 cache_min_evict_age(0),
1410 use_gmt_hitset(true),
1411 min_read_recency_for_promote(0),
1412 min_write_recency_for_promote(0),
1413 hit_set_grade_decay_rate(0),
1414 hit_set_search_last_n(0),
1416 expected_num_objects(0),
1421 void dump(Formatter
*f
) const;
1423 uint64_t get_flags() const { return flags
; }
1424 bool has_flag(uint64_t f
) const { return flags
& f
; }
1425 void set_flag(uint64_t f
) { flags
|= f
; }
1426 void unset_flag(uint64_t f
) { flags
&= ~f
; }
1428 bool ec_pool() const {
1429 return type
== TYPE_ERASURE
;
1431 bool require_rollback() const {
1435 /// true if incomplete clones may be present
1436 bool allow_incomplete_clones() const {
1437 return cache_mode
!= CACHEMODE_NONE
|| has_flag(FLAG_INCOMPLETE_CLONES
);
1440 unsigned get_type() const { return type
; }
1441 unsigned get_size() const { return size
; }
1442 unsigned get_min_size() const { return min_size
; }
1443 int get_crush_rule() const { return crush_rule
; }
1444 int get_object_hash() const { return object_hash
; }
1445 const char *get_object_hash_name() const {
1446 return ceph_str_hash_name(get_object_hash());
1448 epoch_t
get_last_change() const { return last_change
; }
1449 epoch_t
get_last_force_op_resend() const { return last_force_op_resend
; }
1450 epoch_t
get_last_force_op_resend_preluminous() const {
1451 return last_force_op_resend_preluminous
;
1453 epoch_t
get_snap_epoch() const { return snap_epoch
; }
1454 snapid_t
get_snap_seq() const { return snap_seq
; }
1455 uint64_t get_auid() const { return auid
; }
1456 unsigned get_crash_replay_interval() const { return crash_replay_interval
; }
1458 void set_snap_seq(snapid_t s
) { snap_seq
= s
; }
1459 void set_snap_epoch(epoch_t e
) { snap_epoch
= e
; }
1461 void set_stripe_width(uint32_t s
) { stripe_width
= s
; }
1462 uint32_t get_stripe_width() const { return stripe_width
; }
1464 bool is_replicated() const { return get_type() == TYPE_REPLICATED
; }
1465 bool is_erasure() const { return get_type() == TYPE_ERASURE
; }
1467 bool supports_omap() const {
1468 return !(get_type() == TYPE_ERASURE
);
1471 bool requires_aligned_append() const {
1472 return is_erasure() && !has_flag(FLAG_EC_OVERWRITES
);
1474 uint64_t required_alignment() const { return stripe_width
; }
1476 bool allows_ecoverwrites() const {
1477 return has_flag(FLAG_EC_OVERWRITES
);
1480 bool can_shift_osds() const {
1481 switch (get_type()) {
1482 case TYPE_REPLICATED
:
1487 assert(0 == "unhandled pool type");
1491 unsigned get_pg_num() const { return pg_num
; }
1492 unsigned get_pgp_num() const { return pgp_num
; }
1494 unsigned get_pg_num_mask() const { return pg_num_mask
; }
1495 unsigned get_pgp_num_mask() const { return pgp_num_mask
; }
1497 // if pg_num is not a multiple of two, pgs are not equally sized.
1498 // return, for a given pg, the fraction (denominator) of the total
1499 // pool size that it represents.
1500 unsigned get_pg_num_divisor(pg_t pgid
) const;
1502 void set_pg_num(int p
) {
1506 void set_pgp_num(int p
) {
1511 void set_quota_max_bytes(uint64_t m
) {
1512 quota_max_bytes
= m
;
1514 uint64_t get_quota_max_bytes() {
1515 return quota_max_bytes
;
1518 void set_quota_max_objects(uint64_t m
) {
1519 quota_max_objects
= m
;
1521 uint64_t get_quota_max_objects() {
1522 return quota_max_objects
;
1525 void set_last_force_op_resend(uint64_t t
) {
1526 last_force_op_resend
= t
;
1527 last_force_op_resend_preluminous
= t
;
1530 void calc_pg_masks();
1533 * we have two snap modes:
1534 * - pool global snaps
1535 * - snap existence/non-existence defined by snaps[] and snap_seq
1536 * - user managed snaps
1537 * - removal governed by removed_snaps
1539 * we know which mode we're using based on whether removed_snaps is empty.
1540 * If nothing has been created, both functions report false.
1542 bool is_pool_snaps_mode() const;
1543 bool is_unmanaged_snaps_mode() const;
1544 bool is_removed_snap(snapid_t s
) const;
1547 * build set of known-removed sets from either pool snaps or
1548 * explicit removed_snaps set.
1550 void build_removed_snaps(interval_set
<snapid_t
>& rs
) const;
1551 snapid_t
snap_exists(const char *s
) const;
1552 void add_snap(const char *n
, utime_t stamp
);
1553 void add_unmanaged_snap(uint64_t& snapid
);
1554 void remove_snap(snapid_t s
);
1555 void remove_unmanaged_snap(snapid_t s
);
1557 SnapContext
get_snap_context() const;
1559 /// hash a object name+namespace key to a hash position
1560 uint32_t hash_key(const string
& key
, const string
& ns
) const;
1562 /// round a hash position down to a pg num
1563 uint32_t raw_hash_to_pg(uint32_t v
) const;
1566 * map a raw pg (with full precision ps) into an actual pg, for storage
1568 pg_t
raw_pg_to_pg(pg_t pg
) const;
1571 * map raw pg (full precision ps) into a placement seed. include
1572 * pool id in that value so that different pools don't use the same
1575 ps_t
raw_pg_to_pps(pg_t pg
) const;
1577 /// choose a random hash position within a pg
1578 uint32_t get_random_pg_position(pg_t pgid
, uint32_t seed
) const;
1580 void encode(bufferlist
& bl
, uint64_t features
) const;
1581 void decode(bufferlist::iterator
& bl
);
1583 static void generate_test_instances(list
<pg_pool_t
*>& o
);
1585 WRITE_CLASS_ENCODER_FEATURES(pg_pool_t
)
1587 ostream
& operator<<(ostream
& out
, const pg_pool_t
& p
);
1591 * a summation of object stats
1593 * This is just a container for object stats; we don't know what for.
1595 * If you add members in object_stat_sum_t, you should make sure there are
1596 * not padding among these members.
1597 * You should also modify the padding_check function.
1600 struct object_stat_sum_t
{
1601 /**************************************************************************
1602 * WARNING: be sure to update operator==, floor, and split when
1603 * adding/removing fields!
1604 **************************************************************************/
1605 int64_t num_bytes
; // in bytes
1606 int64_t num_objects
;
1607 int64_t num_object_clones
;
1608 int64_t num_object_copies
; // num_objects * num_replicas
1609 int64_t num_objects_missing_on_primary
;
1610 int64_t num_objects_degraded
;
1611 int64_t num_objects_unfound
;
1616 int64_t num_scrub_errors
; // total deep and shallow scrub errors
1617 int64_t num_objects_recovered
;
1618 int64_t num_bytes_recovered
;
1619 int64_t num_keys_recovered
;
1620 int64_t num_shallow_scrub_errors
;
1621 int64_t num_deep_scrub_errors
;
1622 int64_t num_objects_dirty
;
1623 int64_t num_whiteouts
;
1624 int64_t num_objects_omap
;
1625 int64_t num_objects_hit_set_archive
;
1626 int64_t num_objects_misplaced
;
1627 int64_t num_bytes_hit_set_archive
;
1629 int64_t num_flush_kb
;
1631 int64_t num_evict_kb
;
1632 int64_t num_promote
;
1633 int32_t num_flush_mode_high
; // 1 when in high flush mode, otherwise 0
1634 int32_t num_flush_mode_low
; // 1 when in low flush mode, otherwise 0
1635 int32_t num_evict_mode_some
; // 1 when in evict some mode, otherwise 0
1636 int32_t num_evict_mode_full
; // 1 when in evict full mode, otherwise 0
1637 int64_t num_objects_pinned
;
1638 int64_t num_objects_missing
;
1639 int64_t num_legacy_snapsets
; ///< upper bound on pre-luminous-style SnapSets
1643 num_objects(0), num_object_clones(0), num_object_copies(0),
1644 num_objects_missing_on_primary(0), num_objects_degraded(0),
1645 num_objects_unfound(0),
1646 num_rd(0), num_rd_kb(0), num_wr(0), num_wr_kb(0),
1647 num_scrub_errors(0),
1648 num_objects_recovered(0),
1649 num_bytes_recovered(0),
1650 num_keys_recovered(0),
1651 num_shallow_scrub_errors(0),
1652 num_deep_scrub_errors(0),
1653 num_objects_dirty(0),
1655 num_objects_omap(0),
1656 num_objects_hit_set_archive(0),
1657 num_objects_misplaced(0),
1658 num_bytes_hit_set_archive(0),
1664 num_flush_mode_high(0), num_flush_mode_low(0),
1665 num_evict_mode_some(0), num_evict_mode_full(0),
1666 num_objects_pinned(0),
1667 num_objects_missing(0),
1668 num_legacy_snapsets(0)
1671 void floor(int64_t f
) {
1672 #define FLOOR(x) if (x < f) x = f
1675 FLOOR(num_object_clones
);
1676 FLOOR(num_object_copies
);
1677 FLOOR(num_objects_missing_on_primary
);
1678 FLOOR(num_objects_missing
);
1679 FLOOR(num_objects_degraded
);
1680 FLOOR(num_objects_misplaced
);
1681 FLOOR(num_objects_unfound
);
1686 FLOOR(num_scrub_errors
);
1687 FLOOR(num_shallow_scrub_errors
);
1688 FLOOR(num_deep_scrub_errors
);
1689 FLOOR(num_objects_recovered
);
1690 FLOOR(num_bytes_recovered
);
1691 FLOOR(num_keys_recovered
);
1692 FLOOR(num_objects_dirty
);
1693 FLOOR(num_whiteouts
);
1694 FLOOR(num_objects_omap
);
1695 FLOOR(num_objects_hit_set_archive
);
1696 FLOOR(num_bytes_hit_set_archive
);
1698 FLOOR(num_flush_kb
);
1700 FLOOR(num_evict_kb
);
1702 FLOOR(num_flush_mode_high
);
1703 FLOOR(num_flush_mode_low
);
1704 FLOOR(num_evict_mode_some
);
1705 FLOOR(num_evict_mode_full
);
1706 FLOOR(num_objects_pinned
);
1707 FLOOR(num_legacy_snapsets
);
1711 void split(vector
<object_stat_sum_t
> &out
) const {
1712 #define SPLIT(PARAM) \
1713 for (unsigned i = 0; i < out.size(); ++i) { \
1714 out[i].PARAM = PARAM / out.size(); \
1715 if (i < (PARAM % out.size())) { \
1719 #define SPLIT_PRESERVE_NONZERO(PARAM) \
1720 for (unsigned i = 0; i < out.size(); ++i) { \
1722 out[i].PARAM = 1 + PARAM / out.size(); \
1729 SPLIT(num_object_clones
);
1730 SPLIT(num_object_copies
);
1731 SPLIT(num_objects_missing_on_primary
);
1732 SPLIT(num_objects_missing
);
1733 SPLIT(num_objects_degraded
);
1734 SPLIT(num_objects_misplaced
);
1735 SPLIT(num_objects_unfound
);
1740 SPLIT(num_scrub_errors
);
1741 SPLIT(num_shallow_scrub_errors
);
1742 SPLIT(num_deep_scrub_errors
);
1743 SPLIT(num_objects_recovered
);
1744 SPLIT(num_bytes_recovered
);
1745 SPLIT(num_keys_recovered
);
1746 SPLIT(num_objects_dirty
);
1747 SPLIT(num_whiteouts
);
1748 SPLIT(num_objects_omap
);
1749 SPLIT(num_objects_hit_set_archive
);
1750 SPLIT(num_bytes_hit_set_archive
);
1752 SPLIT(num_flush_kb
);
1754 SPLIT(num_evict_kb
);
1756 SPLIT(num_flush_mode_high
);
1757 SPLIT(num_flush_mode_low
);
1758 SPLIT(num_evict_mode_some
);
1759 SPLIT(num_evict_mode_full
);
1760 SPLIT(num_objects_pinned
);
1761 SPLIT_PRESERVE_NONZERO(num_legacy_snapsets
);
1763 #undef SPLIT_PRESERVE_NONZERO
1767 memset(this, 0, sizeof(*this));
1770 void calc_copies(int nrep
) {
1771 num_object_copies
= nrep
* num_objects
;
1774 bool is_zero() const {
1775 return mem_is_zero((char*)this, sizeof(*this));
1778 void add(const object_stat_sum_t
& o
);
1779 void sub(const object_stat_sum_t
& o
);
1781 void dump(Formatter
*f
) const;
1782 void padding_check() {
1784 sizeof(object_stat_sum_t
) ==
1786 sizeof(num_objects
) +
1787 sizeof(num_object_clones
) +
1788 sizeof(num_object_copies
) +
1789 sizeof(num_objects_missing_on_primary
) +
1790 sizeof(num_objects_degraded
) +
1791 sizeof(num_objects_unfound
) +
1796 sizeof(num_scrub_errors
) +
1797 sizeof(num_objects_recovered
) +
1798 sizeof(num_bytes_recovered
) +
1799 sizeof(num_keys_recovered
) +
1800 sizeof(num_shallow_scrub_errors
) +
1801 sizeof(num_deep_scrub_errors
) +
1802 sizeof(num_objects_dirty
) +
1803 sizeof(num_whiteouts
) +
1804 sizeof(num_objects_omap
) +
1805 sizeof(num_objects_hit_set_archive
) +
1806 sizeof(num_objects_misplaced
) +
1807 sizeof(num_bytes_hit_set_archive
) +
1809 sizeof(num_flush_kb
) +
1811 sizeof(num_evict_kb
) +
1812 sizeof(num_promote
) +
1813 sizeof(num_flush_mode_high
) +
1814 sizeof(num_flush_mode_low
) +
1815 sizeof(num_evict_mode_some
) +
1816 sizeof(num_evict_mode_full
) +
1817 sizeof(num_objects_pinned
) +
1818 sizeof(num_objects_missing
) +
1819 sizeof(num_legacy_snapsets
)
1821 "object_stat_sum_t have padding");
1823 void encode(bufferlist
& bl
) const;
1824 void decode(bufferlist::iterator
& bl
);
1825 static void generate_test_instances(list
<object_stat_sum_t
*>& o
);
1827 WRITE_CLASS_ENCODER(object_stat_sum_t
)
1829 bool operator==(const object_stat_sum_t
& l
, const object_stat_sum_t
& r
);
1832 * a collection of object stat sums
1834 * This is a collection of stat sums over different categories.
1836 struct object_stat_collection_t
{
1837 /**************************************************************************
1838 * WARNING: be sure to update the operator== when adding/removing fields! *
1839 **************************************************************************/
1840 object_stat_sum_t sum
;
1842 void calc_copies(int nrep
) {
1843 sum
.calc_copies(nrep
);
1846 void dump(Formatter
*f
) const;
1847 void encode(bufferlist
& bl
) const;
1848 void decode(bufferlist::iterator
& bl
);
1849 static void generate_test_instances(list
<object_stat_collection_t
*>& o
);
1851 bool is_zero() const {
1852 return sum
.is_zero();
1859 void floor(int64_t f
) {
1863 void add(const object_stat_sum_t
& o
) {
1867 void add(const object_stat_collection_t
& o
) {
1870 void sub(const object_stat_collection_t
& o
) {
1874 WRITE_CLASS_ENCODER(object_stat_collection_t
)
1876 inline bool operator==(const object_stat_collection_t
& l
,
1877 const object_stat_collection_t
& r
) {
1878 return l
.sum
== r
.sum
;
1883 * aggregate stats for a single PG.
1886 /**************************************************************************
1887 * WARNING: be sure to update the operator== when adding/removing fields! *
1888 **************************************************************************/
1890 version_t reported_seq
; // sequence number
1891 epoch_t reported_epoch
; // epoch of this report
1893 utime_t last_fresh
; // last reported
1894 utime_t last_change
; // new state != previous state
1895 utime_t last_active
; // state & PG_STATE_ACTIVE
1896 utime_t last_peered
; // state & PG_STATE_ACTIVE || state & PG_STATE_PEERED
1897 utime_t last_clean
; // state & PG_STATE_CLEAN
1898 utime_t last_unstale
; // (state & PG_STATE_STALE) == 0
1899 utime_t last_undegraded
; // (state & PG_STATE_DEGRADED) == 0
1900 utime_t last_fullsized
; // (state & PG_STATE_UNDERSIZED) == 0
1902 eversion_t log_start
; // (log_start,version]
1903 eversion_t ondisk_log_start
; // there may be more on disk
1906 epoch_t last_epoch_clean
;
1908 __u32 parent_split_bits
;
1910 eversion_t last_scrub
;
1911 eversion_t last_deep_scrub
;
1912 utime_t last_scrub_stamp
;
1913 utime_t last_deep_scrub_stamp
;
1914 utime_t last_clean_scrub_stamp
;
1916 object_stat_collection_t stats
;
1919 int64_t ondisk_log_size
; // >= active_log_size
1921 vector
<int32_t> up
, acting
;
1922 epoch_t mapping_epoch
;
1924 vector
<int32_t> blocked_by
; ///< osds on which the pg is blocked
1926 utime_t last_became_active
;
1927 utime_t last_became_peered
;
1929 /// up, acting primaries
1931 int32_t acting_primary
;
1933 bool stats_invalid
:1;
1934 /// true if num_objects_dirty is not accurate (because it was not
1935 /// maintained starting from pool creation)
1936 bool dirty_stats_invalid
:1;
1937 bool omap_stats_invalid
:1;
1938 bool hitset_stats_invalid
:1;
1939 bool hitset_bytes_stats_invalid
:1;
1940 bool pin_stats_invalid
:1;
1946 created(0), last_epoch_clean(0),
1947 parent_split_bits(0),
1948 log_size(0), ondisk_log_size(0),
1952 stats_invalid(false),
1953 dirty_stats_invalid(false),
1954 omap_stats_invalid(false),
1955 hitset_stats_invalid(false),
1956 hitset_bytes_stats_invalid(false),
1957 pin_stats_invalid(false)
1960 epoch_t
get_effective_last_epoch_clean() const {
1961 if (state
& PG_STATE_CLEAN
) {
1962 // we are clean as of this report, and should thus take the
1964 return reported_epoch
;
1966 return last_epoch_clean
;
1970 pair
<epoch_t
, version_t
> get_version_pair() const {
1971 return make_pair(reported_epoch
, reported_seq
);
1974 void floor(int64_t f
) {
1978 if (ondisk_log_size
< f
)
1979 ondisk_log_size
= f
;
1982 void add(const pg_stat_t
& o
) {
1984 log_size
+= o
.log_size
;
1985 ondisk_log_size
+= o
.ondisk_log_size
;
1987 void sub(const pg_stat_t
& o
) {
1989 log_size
-= o
.log_size
;
1990 ondisk_log_size
-= o
.ondisk_log_size
;
1993 bool is_acting_osd(int32_t osd
, bool primary
) const;
1994 void dump(Formatter
*f
) const;
1995 void dump_brief(Formatter
*f
) const;
1996 void encode(bufferlist
&bl
) const;
1997 void decode(bufferlist::iterator
&bl
);
1998 static void generate_test_instances(list
<pg_stat_t
*>& o
);
2000 WRITE_CLASS_ENCODER(pg_stat_t
)
2002 bool operator==(const pg_stat_t
& l
, const pg_stat_t
& r
);
2005 * summation over an entire pool
2007 struct pool_stat_t
{
2008 object_stat_collection_t stats
;
2010 int64_t ondisk_log_size
; // >= active_log_size
2011 int32_t up
; ///< number of up replicas or shards
2012 int32_t acting
; ///< number of acting replicas or shards
2014 pool_stat_t() : log_size(0), ondisk_log_size(0), up(0), acting(0)
2017 void floor(int64_t f
) {
2021 if (ondisk_log_size
< f
)
2022 ondisk_log_size
= f
;
2029 void add(const pg_stat_t
& o
) {
2031 log_size
+= o
.log_size
;
2032 ondisk_log_size
+= o
.ondisk_log_size
;
2034 acting
+= o
.acting
.size();
2036 void sub(const pg_stat_t
& o
) {
2038 log_size
-= o
.log_size
;
2039 ondisk_log_size
-= o
.ondisk_log_size
;
2041 acting
-= o
.acting
.size();
2044 bool is_zero() const {
2045 return (stats
.is_zero() &&
2047 ondisk_log_size
== 0 &&
2052 void dump(Formatter
*f
) const;
2053 void encode(bufferlist
&bl
, uint64_t features
) const;
2054 void decode(bufferlist::iterator
&bl
);
2055 static void generate_test_instances(list
<pool_stat_t
*>& o
);
2057 WRITE_CLASS_ENCODER_FEATURES(pool_stat_t
)
2060 // -----------------------------------------
2063 * pg_hit_set_info_t - information about a single recorded HitSet
2065 * Track basic metadata about a HitSet, like the nubmer of insertions
2066 * and the time range it covers.
2068 struct pg_hit_set_info_t
{
2069 utime_t begin
, end
; ///< time interval
2070 eversion_t version
; ///< version this HitSet object was written
2071 bool using_gmt
; ///< use gmt for creating the hit_set archive object name
2073 friend bool operator==(const pg_hit_set_info_t
& l
,
2074 const pg_hit_set_info_t
& r
) {
2076 l
.begin
== r
.begin
&&
2078 l
.version
== r
.version
&&
2079 l
.using_gmt
== r
.using_gmt
;
2082 explicit pg_hit_set_info_t(bool using_gmt
= true)
2083 : using_gmt(using_gmt
) {}
2085 void encode(bufferlist
&bl
) const;
2086 void decode(bufferlist::iterator
&bl
);
2087 void dump(Formatter
*f
) const;
2088 static void generate_test_instances(list
<pg_hit_set_info_t
*>& o
);
2090 WRITE_CLASS_ENCODER(pg_hit_set_info_t
)
2093 * pg_hit_set_history_t - information about a history of hitsets
2095 * Include information about the currently accumulating hit set as well
2096 * as archived/historical ones.
2098 struct pg_hit_set_history_t
{
2099 eversion_t current_last_update
; ///< last version inserted into current set
2100 list
<pg_hit_set_info_t
> history
; ///< archived sets, sorted oldest -> newest
2102 friend bool operator==(const pg_hit_set_history_t
& l
,
2103 const pg_hit_set_history_t
& r
) {
2105 l
.current_last_update
== r
.current_last_update
&&
2106 l
.history
== r
.history
;
2109 void encode(bufferlist
&bl
) const;
2110 void decode(bufferlist::iterator
&bl
);
2111 void dump(Formatter
*f
) const;
2112 static void generate_test_instances(list
<pg_hit_set_history_t
*>& o
);
2114 WRITE_CLASS_ENCODER(pg_hit_set_history_t
)
2117 // -----------------------------------------
2120 * pg_history_t - information about recent pg peering/mapping history
2122 * This is aggressively shared between OSDs to bound the amount of past
2123 * history they need to worry about.
2125 struct pg_history_t
{
2126 epoch_t epoch_created
; // epoch in which *pg* was created (pool or pg)
2127 epoch_t epoch_pool_created
; // epoch in which *pool* was created
2128 // (note: may be pg creation epoch for
2129 // pre-luminous clusters)
2130 epoch_t last_epoch_started
; // lower bound on last epoch started (anywhere, not necessarily locally)
2131 epoch_t last_interval_started
; // first epoch of last_epoch_started interval
2132 epoch_t last_epoch_clean
; // lower bound on last epoch the PG was completely clean.
2133 epoch_t last_interval_clean
; // first epoch of last_epoch_clean interval
2134 epoch_t last_epoch_split
; // as parent or child
2135 epoch_t last_epoch_marked_full
; // pool or cluster
2138 * In the event of a map discontinuity, same_*_since may reflect the first
2139 * map the osd has seen in the new map sequence rather than the actual start
2140 * of the interval. This is ok since a discontinuity at epoch e means there
2141 * must have been a clean interval between e and now and that we cannot be
2142 * in the active set during the interval containing e.
2144 epoch_t same_up_since
; // same acting set since
2145 epoch_t same_interval_since
; // same acting AND up set since
2146 epoch_t same_primary_since
; // same primary at least back through this epoch.
2148 eversion_t last_scrub
;
2149 eversion_t last_deep_scrub
;
2150 utime_t last_scrub_stamp
;
2151 utime_t last_deep_scrub_stamp
;
2152 utime_t last_clean_scrub_stamp
;
2154 friend bool operator==(const pg_history_t
& l
, const pg_history_t
& r
) {
2156 l
.epoch_created
== r
.epoch_created
&&
2157 l
.epoch_pool_created
== r
.epoch_pool_created
&&
2158 l
.last_epoch_started
== r
.last_epoch_started
&&
2159 l
.last_interval_started
== r
.last_interval_started
&&
2160 l
.last_epoch_clean
== r
.last_epoch_clean
&&
2161 l
.last_interval_clean
== r
.last_interval_clean
&&
2162 l
.last_epoch_split
== r
.last_epoch_split
&&
2163 l
.last_epoch_marked_full
== r
.last_epoch_marked_full
&&
2164 l
.same_up_since
== r
.same_up_since
&&
2165 l
.same_interval_since
== r
.same_interval_since
&&
2166 l
.same_primary_since
== r
.same_primary_since
&&
2167 l
.last_scrub
== r
.last_scrub
&&
2168 l
.last_deep_scrub
== r
.last_deep_scrub
&&
2169 l
.last_scrub_stamp
== r
.last_scrub_stamp
&&
2170 l
.last_deep_scrub_stamp
== r
.last_deep_scrub_stamp
&&
2171 l
.last_clean_scrub_stamp
== r
.last_clean_scrub_stamp
;
2176 epoch_pool_created(0),
2177 last_epoch_started(0),
2178 last_interval_started(0),
2179 last_epoch_clean(0),
2180 last_interval_clean(0),
2181 last_epoch_split(0),
2182 last_epoch_marked_full(0),
2183 same_up_since(0), same_interval_since(0), same_primary_since(0) {}
2185 bool merge(const pg_history_t
&other
) {
2186 // Here, we only update the fields which cannot be calculated from the OSDmap.
2187 bool modified
= false;
2188 if (epoch_created
< other
.epoch_created
) {
2189 epoch_created
= other
.epoch_created
;
2192 if (epoch_pool_created
< other
.epoch_pool_created
) {
2193 // FIXME: for jewel compat only; this should either be 0 or always the
2194 // same value across all pg instances.
2195 epoch_pool_created
= other
.epoch_pool_created
;
2198 if (last_epoch_started
< other
.last_epoch_started
) {
2199 last_epoch_started
= other
.last_epoch_started
;
2202 if (last_interval_started
< other
.last_interval_started
) {
2203 last_interval_started
= other
.last_interval_started
;
2206 if (last_epoch_clean
< other
.last_epoch_clean
) {
2207 last_epoch_clean
= other
.last_epoch_clean
;
2210 if (last_interval_clean
< other
.last_interval_clean
) {
2211 last_interval_clean
= other
.last_interval_clean
;
2214 if (last_epoch_split
< other
.last_epoch_split
) {
2215 last_epoch_split
= other
.last_epoch_split
;
2218 if (last_epoch_marked_full
< other
.last_epoch_marked_full
) {
2219 last_epoch_marked_full
= other
.last_epoch_marked_full
;
2222 if (other
.last_scrub
> last_scrub
) {
2223 last_scrub
= other
.last_scrub
;
2226 if (other
.last_scrub_stamp
> last_scrub_stamp
) {
2227 last_scrub_stamp
= other
.last_scrub_stamp
;
2230 if (other
.last_deep_scrub
> last_deep_scrub
) {
2231 last_deep_scrub
= other
.last_deep_scrub
;
2234 if (other
.last_deep_scrub_stamp
> last_deep_scrub_stamp
) {
2235 last_deep_scrub_stamp
= other
.last_deep_scrub_stamp
;
2238 if (other
.last_clean_scrub_stamp
> last_clean_scrub_stamp
) {
2239 last_clean_scrub_stamp
= other
.last_clean_scrub_stamp
;
2245 void encode(bufferlist
& bl
) const;
2246 void decode(bufferlist::iterator
& p
);
2247 void dump(Formatter
*f
) const;
2248 static void generate_test_instances(list
<pg_history_t
*>& o
);
2250 WRITE_CLASS_ENCODER(pg_history_t
)
2252 inline ostream
& operator<<(ostream
& out
, const pg_history_t
& h
) {
2253 return out
<< "ec=" << h
.epoch_created
<< "/" << h
.epoch_pool_created
2254 << " lis/c " << h
.last_interval_started
2255 << "/" << h
.last_interval_clean
2256 << " les/c/f " << h
.last_epoch_started
<< "/" << h
.last_epoch_clean
2257 << "/" << h
.last_epoch_marked_full
2258 << " " << h
.same_up_since
2259 << "/" << h
.same_interval_since
2260 << "/" << h
.same_primary_since
;
2265 * pg_info_t - summary of PG statistics.
2268 * - last_complete implies we have all objects that existed as of that
2269 * stamp, OR a newer object, OR have already applied a later delete.
2270 * - if last_complete >= log.bottom, then we know pg contents thru log.head.
2271 * otherwise, we have no idea what the pg is supposed to contain.
2275 eversion_t last_update
; ///< last object version applied to store.
2276 eversion_t last_complete
; ///< last version pg was complete through.
2277 epoch_t last_epoch_started
; ///< last epoch at which this pg started on this osd
2278 epoch_t last_interval_started
; ///< first epoch of last_epoch_started interval
2280 version_t last_user_version
; ///< last user object version applied to store
2282 eversion_t log_tail
; ///< oldest log entry.
2284 hobject_t last_backfill
; ///< objects >= this and < last_complete may be missing
2285 bool last_backfill_bitwise
; ///< true if last_backfill reflects a bitwise (vs nibblewise) sort
2287 interval_set
<snapid_t
> purged_snaps
;
2291 pg_history_t history
;
2292 pg_hit_set_history_t hit_set
;
2294 friend bool operator==(const pg_info_t
& l
, const pg_info_t
& r
) {
2297 l
.last_update
== r
.last_update
&&
2298 l
.last_complete
== r
.last_complete
&&
2299 l
.last_epoch_started
== r
.last_epoch_started
&&
2300 l
.last_interval_started
== r
.last_interval_started
&&
2301 l
.last_user_version
== r
.last_user_version
&&
2302 l
.log_tail
== r
.log_tail
&&
2303 l
.last_backfill
== r
.last_backfill
&&
2304 l
.last_backfill_bitwise
== r
.last_backfill_bitwise
&&
2305 l
.purged_snaps
== r
.purged_snaps
&&
2306 l
.stats
== r
.stats
&&
2307 l
.history
== r
.history
&&
2308 l
.hit_set
== r
.hit_set
;
2312 : last_epoch_started(0),
2313 last_interval_started(0),
2314 last_user_version(0),
2315 last_backfill(hobject_t::get_max()),
2316 last_backfill_bitwise(false)
2318 // cppcheck-suppress noExplicitConstructor
2321 last_epoch_started(0),
2322 last_interval_started(0),
2323 last_user_version(0),
2324 last_backfill(hobject_t::get_max()),
2325 last_backfill_bitwise(false)
2328 void set_last_backfill(hobject_t pos
) {
2329 last_backfill
= pos
;
2330 last_backfill_bitwise
= true;
2333 bool is_empty() const { return last_update
.version
== 0; }
2334 bool dne() const { return history
.epoch_created
== 0; }
2336 bool is_incomplete() const { return !last_backfill
.is_max(); }
2338 void encode(bufferlist
& bl
) const;
2339 void decode(bufferlist::iterator
& p
);
2340 void dump(Formatter
*f
) const;
2341 bool overlaps_with(const pg_info_t
&oinfo
) const {
2342 return last_update
> oinfo
.log_tail
?
2343 oinfo
.last_update
>= log_tail
:
2344 last_update
>= oinfo
.log_tail
;
2346 static void generate_test_instances(list
<pg_info_t
*>& o
);
2348 WRITE_CLASS_ENCODER(pg_info_t
)
2350 inline ostream
& operator<<(ostream
& out
, const pg_info_t
& pgi
)
2352 out
<< pgi
.pgid
<< "(";
2358 out
<< " v " << pgi
.last_update
;
2359 if (pgi
.last_complete
!= pgi
.last_update
)
2360 out
<< " lc " << pgi
.last_complete
;
2361 out
<< " (" << pgi
.log_tail
<< "," << pgi
.last_update
<< "]";
2363 if (pgi
.is_incomplete())
2364 out
<< " lb " << pgi
.last_backfill
2365 << (pgi
.last_backfill_bitwise
? " (bitwise)" : " (NIBBLEWISE)");
2366 //out << " c " << pgi.epoch_created;
2367 out
<< " local-lis/les=" << pgi
.last_interval_started
2368 << "/" << pgi
.last_epoch_started
;
2369 out
<< " n=" << pgi
.stats
.stats
.sum
.num_objects
;
2370 out
<< " " << pgi
.history
2376 * pg_fast_info_t - common pg_info_t fields
2378 * These are the fields of pg_info_t (and children) that are updated for
2379 * most IO operations.
2382 * Because we rely on these fields to be applied to the normal
2383 * info struct, adding a new field here that is not also new in info
2384 * means that we must set an incompat OSD feature bit!
2386 struct pg_fast_info_t
{
2387 eversion_t last_update
;
2388 eversion_t last_complete
;
2389 version_t last_user_version
;
2390 struct { // pg_stat_t stats
2392 version_t reported_seq
;
2394 utime_t last_active
;
2395 utime_t last_peered
;
2397 utime_t last_unstale
;
2398 utime_t last_undegraded
;
2399 utime_t last_fullsized
;
2400 int64_t log_size
; // (also ondisk_log_size, which has the same value)
2401 struct { // object_stat_collection_t stats;
2402 struct { // objct_stat_sum_t sum
2403 int64_t num_bytes
; // in bytes
2404 int64_t num_objects
;
2405 int64_t num_object_copies
;
2410 int64_t num_objects_dirty
;
2415 void populate_from(const pg_info_t
& info
) {
2416 last_update
= info
.last_update
;
2417 last_complete
= info
.last_complete
;
2418 last_user_version
= info
.last_user_version
;
2419 stats
.version
= info
.stats
.version
;
2420 stats
.reported_seq
= info
.stats
.reported_seq
;
2421 stats
.last_fresh
= info
.stats
.last_fresh
;
2422 stats
.last_active
= info
.stats
.last_active
;
2423 stats
.last_peered
= info
.stats
.last_peered
;
2424 stats
.last_clean
= info
.stats
.last_clean
;
2425 stats
.last_unstale
= info
.stats
.last_unstale
;
2426 stats
.last_undegraded
= info
.stats
.last_undegraded
;
2427 stats
.last_fullsized
= info
.stats
.last_fullsized
;
2428 stats
.log_size
= info
.stats
.log_size
;
2429 stats
.stats
.sum
.num_bytes
= info
.stats
.stats
.sum
.num_bytes
;
2430 stats
.stats
.sum
.num_objects
= info
.stats
.stats
.sum
.num_objects
;
2431 stats
.stats
.sum
.num_object_copies
= info
.stats
.stats
.sum
.num_object_copies
;
2432 stats
.stats
.sum
.num_rd
= info
.stats
.stats
.sum
.num_rd
;
2433 stats
.stats
.sum
.num_rd_kb
= info
.stats
.stats
.sum
.num_rd_kb
;
2434 stats
.stats
.sum
.num_wr
= info
.stats
.stats
.sum
.num_wr
;
2435 stats
.stats
.sum
.num_wr_kb
= info
.stats
.stats
.sum
.num_wr_kb
;
2436 stats
.stats
.sum
.num_objects_dirty
= info
.stats
.stats
.sum
.num_objects_dirty
;
2439 bool try_apply_to(pg_info_t
* info
) {
2440 if (last_update
<= info
->last_update
)
2442 info
->last_update
= last_update
;
2443 info
->last_complete
= last_complete
;
2444 info
->last_user_version
= last_user_version
;
2445 info
->stats
.version
= stats
.version
;
2446 info
->stats
.reported_seq
= stats
.reported_seq
;
2447 info
->stats
.last_fresh
= stats
.last_fresh
;
2448 info
->stats
.last_active
= stats
.last_active
;
2449 info
->stats
.last_peered
= stats
.last_peered
;
2450 info
->stats
.last_clean
= stats
.last_clean
;
2451 info
->stats
.last_unstale
= stats
.last_unstale
;
2452 info
->stats
.last_undegraded
= stats
.last_undegraded
;
2453 info
->stats
.last_fullsized
= stats
.last_fullsized
;
2454 info
->stats
.log_size
= stats
.log_size
;
2455 info
->stats
.ondisk_log_size
= stats
.log_size
;
2456 info
->stats
.stats
.sum
.num_bytes
= stats
.stats
.sum
.num_bytes
;
2457 info
->stats
.stats
.sum
.num_objects
= stats
.stats
.sum
.num_objects
;
2458 info
->stats
.stats
.sum
.num_object_copies
= stats
.stats
.sum
.num_object_copies
;
2459 info
->stats
.stats
.sum
.num_rd
= stats
.stats
.sum
.num_rd
;
2460 info
->stats
.stats
.sum
.num_rd_kb
= stats
.stats
.sum
.num_rd_kb
;
2461 info
->stats
.stats
.sum
.num_wr
= stats
.stats
.sum
.num_wr
;
2462 info
->stats
.stats
.sum
.num_wr_kb
= stats
.stats
.sum
.num_wr_kb
;
2463 info
->stats
.stats
.sum
.num_objects_dirty
= stats
.stats
.sum
.num_objects_dirty
;
2467 void encode(bufferlist
& bl
) const {
2468 ENCODE_START(1, 1, bl
);
2469 ::encode(last_update
, bl
);
2470 ::encode(last_complete
, bl
);
2471 ::encode(last_user_version
, bl
);
2472 ::encode(stats
.version
, bl
);
2473 ::encode(stats
.reported_seq
, bl
);
2474 ::encode(stats
.last_fresh
, bl
);
2475 ::encode(stats
.last_active
, bl
);
2476 ::encode(stats
.last_peered
, bl
);
2477 ::encode(stats
.last_clean
, bl
);
2478 ::encode(stats
.last_unstale
, bl
);
2479 ::encode(stats
.last_undegraded
, bl
);
2480 ::encode(stats
.last_fullsized
, bl
);
2481 ::encode(stats
.log_size
, bl
);
2482 ::encode(stats
.stats
.sum
.num_bytes
, bl
);
2483 ::encode(stats
.stats
.sum
.num_objects
, bl
);
2484 ::encode(stats
.stats
.sum
.num_object_copies
, bl
);
2485 ::encode(stats
.stats
.sum
.num_rd
, bl
);
2486 ::encode(stats
.stats
.sum
.num_rd_kb
, bl
);
2487 ::encode(stats
.stats
.sum
.num_wr
, bl
);
2488 ::encode(stats
.stats
.sum
.num_wr_kb
, bl
);
2489 ::encode(stats
.stats
.sum
.num_objects_dirty
, bl
);
2492 void decode(bufferlist::iterator
& p
) {
2494 ::decode(last_update
, p
);
2495 ::decode(last_complete
, p
);
2496 ::decode(last_user_version
, p
);
2497 ::decode(stats
.version
, p
);
2498 ::decode(stats
.reported_seq
, p
);
2499 ::decode(stats
.last_fresh
, p
);
2500 ::decode(stats
.last_active
, p
);
2501 ::decode(stats
.last_peered
, p
);
2502 ::decode(stats
.last_clean
, p
);
2503 ::decode(stats
.last_unstale
, p
);
2504 ::decode(stats
.last_undegraded
, p
);
2505 ::decode(stats
.last_fullsized
, p
);
2506 ::decode(stats
.log_size
, p
);
2507 ::decode(stats
.stats
.sum
.num_bytes
, p
);
2508 ::decode(stats
.stats
.sum
.num_objects
, p
);
2509 ::decode(stats
.stats
.sum
.num_object_copies
, p
);
2510 ::decode(stats
.stats
.sum
.num_rd
, p
);
2511 ::decode(stats
.stats
.sum
.num_rd_kb
, p
);
2512 ::decode(stats
.stats
.sum
.num_wr
, p
);
2513 ::decode(stats
.stats
.sum
.num_wr_kb
, p
);
2514 ::decode(stats
.stats
.sum
.num_objects_dirty
, p
);
2518 WRITE_CLASS_ENCODER(pg_fast_info_t
)
2521 struct pg_notify_t
{
2522 epoch_t query_epoch
;
2528 query_epoch(0), epoch_sent(0), to(shard_id_t::NO_SHARD
),
2529 from(shard_id_t::NO_SHARD
) {}
2533 epoch_t query_epoch
,
2535 const pg_info_t
&info
)
2536 : query_epoch(query_epoch
),
2537 epoch_sent(epoch_sent
),
2538 info(info
), to(to
), from(from
) {
2539 assert(from
== info
.pgid
.shard
);
2541 void encode(bufferlist
&bl
) const;
2542 void decode(bufferlist::iterator
&p
);
2543 void dump(Formatter
*f
) const;
2544 static void generate_test_instances(list
<pg_notify_t
*> &o
);
2546 WRITE_CLASS_ENCODER(pg_notify_t
)
2547 ostream
&operator<<(ostream
&lhs
, const pg_notify_t
¬ify
);
2552 * PastIntervals -- information needed to determine the PriorSet and
2553 * the might_have_unfound set
2555 class PastIntervals
{
2557 struct pg_interval_t
{
2558 vector
<int32_t> up
, acting
;
2559 epoch_t first
, last
;
2565 : first(0), last(0),
2566 maybe_went_rw(false),
2572 vector
<int32_t> &&up
,
2573 vector
<int32_t> &&acting
,
2579 : up(up
), acting(acting
), first(first
), last(last
),
2580 maybe_went_rw(maybe_went_rw
), primary(primary
), up_primary(up_primary
)
2583 void encode(bufferlist
& bl
) const;
2584 void decode(bufferlist::iterator
& bl
);
2585 void dump(Formatter
*f
) const;
2586 static void generate_test_instances(list
<pg_interval_t
*>& o
);
2589 PastIntervals() = default;
2590 PastIntervals(bool ec_pool
, const OSDMap
&osdmap
) : PastIntervals() {
2591 update_type_from_map(ec_pool
, osdmap
);
2593 PastIntervals(bool ec_pool
, bool compact
) : PastIntervals() {
2594 update_type(ec_pool
, compact
);
2596 PastIntervals(PastIntervals
&&rhs
) = default;
2597 PastIntervals
&operator=(PastIntervals
&&rhs
) = default;
2599 PastIntervals(const PastIntervals
&rhs
);
2600 PastIntervals
&operator=(const PastIntervals
&rhs
);
2602 class interval_rep
{
2604 virtual size_t size() const = 0;
2605 virtual bool empty() const = 0;
2606 virtual void clear() = 0;
2607 virtual pair
<epoch_t
, epoch_t
> get_bounds() const = 0;
2608 virtual set
<pg_shard_t
> get_all_participants(
2609 bool ec_pool
) const = 0;
2610 virtual void add_interval(bool ec_pool
, const pg_interval_t
&interval
) = 0;
2611 virtual unique_ptr
<interval_rep
> clone() const = 0;
2612 virtual ostream
&print(ostream
&out
) const = 0;
2613 virtual void encode(bufferlist
&bl
) const = 0;
2614 virtual void decode(bufferlist::iterator
&bl
) = 0;
2615 virtual void dump(Formatter
*f
) const = 0;
2616 virtual bool is_classic() const = 0;
2617 virtual void iterate_mayberw_back_to(
2620 std::function
<void(epoch_t
, const set
<pg_shard_t
> &)> &&f
) const = 0;
2622 virtual bool has_full_intervals() const { return false; }
2623 virtual void iterate_all_intervals(
2624 std::function
<void(const pg_interval_t
&)> &&f
) const {
2625 assert(!has_full_intervals());
2626 assert(0 == "not valid for this implementation");
2629 virtual ~interval_rep() {}
2631 friend class pi_simple_rep
;
2632 friend class pi_compact_rep
;
2635 unique_ptr
<interval_rep
> past_intervals
;
2637 PastIntervals(interval_rep
*rep
) : past_intervals(rep
) {}
2640 void add_interval(bool ec_pool
, const pg_interval_t
&interval
) {
2641 assert(past_intervals
);
2642 return past_intervals
->add_interval(ec_pool
, interval
);
2645 bool is_classic() const {
2646 assert(past_intervals
);
2647 return past_intervals
->is_classic();
2650 void encode(bufferlist
&bl
) const {
2651 ENCODE_START(1, 1, bl
);
2652 if (past_intervals
) {
2653 __u8 type
= is_classic() ? 1 : 2;
2655 past_intervals
->encode(bl
);
2657 ::encode((__u8
)0, bl
);
2661 void encode_classic(bufferlist
&bl
) const {
2662 if (past_intervals
) {
2663 assert(past_intervals
->is_classic());
2664 past_intervals
->encode(bl
);
2667 ::encode((uint32_t)0, bl
);
2671 void decode(bufferlist::iterator
&bl
);
2672 void decode_classic(bufferlist::iterator
&bl
);
2674 void dump(Formatter
*f
) const {
2675 assert(past_intervals
);
2676 past_intervals
->dump(f
);
2678 static void generate_test_instances(list
<PastIntervals
*> & o
);
2681 * Determines whether there is an interval change
2683 static bool is_new_interval(
2684 int old_acting_primary
,
2685 int new_acting_primary
,
2686 const vector
<int> &old_acting
,
2687 const vector
<int> &new_acting
,
2690 const vector
<int> &old_up
,
2691 const vector
<int> &new_up
,
2696 unsigned old_pg_num
,
2697 unsigned new_pg_num
,
2698 bool old_sort_bitwise
,
2699 bool new_sort_bitwise
,
2700 bool old_recovery_deletes
,
2701 bool new_recovery_deletes
,
2706 * Determines whether there is an interval change
2708 static bool is_new_interval(
2709 int old_acting_primary
, ///< [in] primary as of lastmap
2710 int new_acting_primary
, ///< [in] primary as of lastmap
2711 const vector
<int> &old_acting
, ///< [in] acting as of lastmap
2712 const vector
<int> &new_acting
, ///< [in] acting as of osdmap
2713 int old_up_primary
, ///< [in] up primary of lastmap
2714 int new_up_primary
, ///< [in] up primary of osdmap
2715 const vector
<int> &old_up
, ///< [in] up as of lastmap
2716 const vector
<int> &new_up
, ///< [in] up as of osdmap
2717 ceph::shared_ptr
<const OSDMap
> osdmap
, ///< [in] current map
2718 ceph::shared_ptr
<const OSDMap
> lastmap
, ///< [in] last map
2719 pg_t pgid
///< [in] pgid for pg
2723 * Integrates a new map into *past_intervals, returns true
2724 * if an interval was closed out.
2726 static bool check_new_interval(
2727 int old_acting_primary
, ///< [in] primary as of lastmap
2728 int new_acting_primary
, ///< [in] primary as of osdmap
2729 const vector
<int> &old_acting
, ///< [in] acting as of lastmap
2730 const vector
<int> &new_acting
, ///< [in] acting as of osdmap
2731 int old_up_primary
, ///< [in] up primary of lastmap
2732 int new_up_primary
, ///< [in] up primary of osdmap
2733 const vector
<int> &old_up
, ///< [in] up as of lastmap
2734 const vector
<int> &new_up
, ///< [in] up as of osdmap
2735 epoch_t same_interval_since
, ///< [in] as of osdmap
2736 epoch_t last_epoch_clean
, ///< [in] current
2737 ceph::shared_ptr
<const OSDMap
> osdmap
, ///< [in] current map
2738 ceph::shared_ptr
<const OSDMap
> lastmap
, ///< [in] last map
2739 pg_t pgid
, ///< [in] pgid for pg
2740 IsPGRecoverablePredicate
*could_have_gone_active
, /// [in] predicate whether the pg can be active
2741 PastIntervals
*past_intervals
, ///< [out] intervals
2742 ostream
*out
= 0 ///< [out] debug ostream
2745 friend ostream
& operator<<(ostream
& out
, const PastIntervals
&i
);
2747 template <typename F
>
2748 void iterate_mayberw_back_to(
2752 assert(past_intervals
);
2753 past_intervals
->iterate_mayberw_back_to(ec_pool
, les
, std::forward
<F
>(f
));
2756 assert(past_intervals
);
2757 past_intervals
->clear();
2761 * Should return a value which gives an indication of the amount
2762 * of state contained
2764 size_t size() const {
2765 assert(past_intervals
);
2766 return past_intervals
->size();
2769 bool empty() const {
2770 assert(past_intervals
);
2771 return past_intervals
->empty();
2774 void swap(PastIntervals
&other
) {
2776 swap(other
.past_intervals
, past_intervals
);
2780 * Return all shards which have been in the acting set back to the
2781 * latest epoch to which we have trimmed except for pg_whoami
2783 set
<pg_shard_t
> get_might_have_unfound(
2784 pg_shard_t pg_whoami
,
2785 bool ec_pool
) const {
2786 assert(past_intervals
);
2787 auto ret
= past_intervals
->get_all_participants(ec_pool
);
2788 ret
.erase(pg_whoami
);
2793 * Return all shards which we might want to talk to for peering
2795 set
<pg_shard_t
> get_all_probe(
2796 bool ec_pool
) const {
2797 assert(past_intervals
);
2798 return past_intervals
->get_all_participants(ec_pool
);
2801 /* Return the set of epochs [start, end) represented by the
2802 * past_interval set.
2804 pair
<epoch_t
, epoch_t
> get_bounds() const {
2805 assert(past_intervals
);
2806 return past_intervals
->get_bounds();
2816 bool ec_pool
= false;
2817 set
<pg_shard_t
> probe
; /// current+prior OSDs we need to probe.
2818 set
<int> down
; /// down osds that would normally be in @a probe and might be interesting.
2819 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
2821 bool pg_down
= false; /// some down osds are included in @a cur; the DOWN pg state bit should be set.
2822 unique_ptr
<IsPGRecoverablePredicate
> pcontdec
;
2824 PriorSet() = default;
2825 PriorSet(PriorSet
&&) = default;
2826 PriorSet
&operator=(PriorSet
&&) = default;
2828 PriorSet
&operator=(const PriorSet
&) = delete;
2829 PriorSet(const PriorSet
&) = delete;
2831 bool operator==(const PriorSet
&rhs
) const {
2832 return (ec_pool
== rhs
.ec_pool
) &&
2833 (probe
== rhs
.probe
) &&
2834 (down
== rhs
.down
) &&
2835 (blocked_by
== rhs
.blocked_by
) &&
2836 (pg_down
== rhs
.pg_down
);
2839 bool affected_by_map(
2840 const OSDMap
&osdmap
,
2841 const DoutPrefixProvider
*dpp
) const;
2843 // For verifying tests
2846 set
<pg_shard_t
> probe
,
2848 map
<int, epoch_t
> blocked_by
,
2850 IsPGRecoverablePredicate
*pcontdec
)
2851 : ec_pool(ec_pool
), probe(probe
), down(down
), blocked_by(blocked_by
),
2852 pg_down(pg_down
), pcontdec(pcontdec
) {}
2855 template <typename F
>
2857 const PastIntervals
&past_intervals
,
2859 epoch_t last_epoch_started
,
2860 IsPGRecoverablePredicate
*c
,
2862 const vector
<int> &up
,
2863 const vector
<int> &acting
,
2864 const DoutPrefixProvider
*dpp
);
2866 friend class PastIntervals
;
2869 void update_type(bool ec_pool
, bool compact
);
2870 void update_type_from_map(bool ec_pool
, const OSDMap
&osdmap
);
2872 template <typename
... Args
>
2873 PriorSet
get_prior_set(Args
&&... args
) const {
2874 return PriorSet(*this, std::forward
<Args
>(args
)...);
2877 WRITE_CLASS_ENCODER(PastIntervals
)
2879 ostream
& operator<<(ostream
& out
, const PastIntervals::pg_interval_t
& i
);
2880 ostream
& operator<<(ostream
& out
, const PastIntervals
&i
);
2881 ostream
& operator<<(ostream
& out
, const PastIntervals::PriorSet
&i
);
2883 template <typename F
>
2884 PastIntervals::PriorSet::PriorSet(
2885 const PastIntervals
&past_intervals
,
2887 epoch_t last_epoch_started
,
2888 IsPGRecoverablePredicate
*c
,
2890 const vector
<int> &up
,
2891 const vector
<int> &acting
,
2892 const DoutPrefixProvider
*dpp
)
2893 : ec_pool(ec_pool
), pg_down(false), pcontdec(c
)
2896 * We have to be careful to gracefully deal with situations like
2897 * so. Say we have a power outage or something that takes out both
2898 * OSDs, but the monitor doesn't mark them down in the same epoch.
2899 * The history may look like
2903 * 3: let's say B dies for good, too (say, from the power spike)
2906 * which makes it look like B may have applied updates to the PG
2907 * that we need in order to proceed. This sucks...
2909 * To minimize the risk of this happening, we CANNOT go active if
2910 * _any_ OSDs in the prior set are down until we send an MOSDAlive
2911 * to the monitor such that the OSDMap sets osd_up_thru to an epoch.
2912 * Then, we have something like
2919 * -> we can ignore B, bc it couldn't have gone active (alive_thru
2930 * -> we must wait for B, bc it was alive through 2, and could have
2931 * written to the pg.
2933 * If B is really dead, then an administrator will need to manually
2934 * intervene by marking the OSD as "lost."
2937 // Include current acting and up nodes... not because they may
2938 // contain old data (this interval hasn't gone active, obviously),
2939 // but because we want their pg_info to inform choose_acting(), and
2940 // so that we know what they do/do not have explicitly before
2941 // sending them any new info/logs/whatever.
2942 for (unsigned i
= 0; i
< acting
.size(); i
++) {
2943 if (acting
[i
] != 0x7fffffff /* CRUSH_ITEM_NONE, can't import crush.h here */)
2944 probe
.insert(pg_shard_t(acting
[i
], ec_pool
? shard_id_t(i
) : shard_id_t::NO_SHARD
));
2946 // It may be possible to exclude the up nodes, but let's keep them in
2948 for (unsigned i
= 0; i
< up
.size(); i
++) {
2949 if (up
[i
] != 0x7fffffff /* CRUSH_ITEM_NONE, can't import crush.h here */)
2950 probe
.insert(pg_shard_t(up
[i
], ec_pool
? shard_id_t(i
) : shard_id_t::NO_SHARD
));
2953 set
<pg_shard_t
> all_probe
= past_intervals
.get_all_probe(ec_pool
);
2954 ldpp_dout(dpp
, 10) << "build_prior all_probe " << all_probe
<< dendl
;
2955 for (auto &&i
: all_probe
) {
2956 switch (f(0, i
.osd
, nullptr)) {
2970 past_intervals
.iterate_mayberw_back_to(
2973 [&](epoch_t start
, const set
<pg_shard_t
> &acting
) {
2974 ldpp_dout(dpp
, 10) << "build_prior maybe_rw interval:" << start
2975 << ", acting: " << acting
<< dendl
;
2977 // look at candidate osds during this interval. each falls into
2978 // one of three categories: up, down (but potentially
2979 // interesting), or lost (down, but we won't wait for it).
2980 set
<pg_shard_t
> up_now
;
2981 map
<int, epoch_t
> candidate_blocked_by
;
2982 // any candidates down now (that might have useful data)
2983 bool any_down_now
= false;
2985 // consider ACTING osds
2986 for (auto &&so
: acting
) {
2987 epoch_t lost_at
= 0;
2988 switch (f(start
, so
.osd
, &lost_at
)) {
2990 // include past acting osds if they are up.
2995 ldpp_dout(dpp
, 10) << "build_prior prior osd." << so
.osd
2996 << " no longer exists" << dendl
;
3000 ldpp_dout(dpp
, 10) << "build_prior prior osd." << so
.osd
3001 << " is down, but lost_at " << lost_at
<< dendl
;
3006 ldpp_dout(dpp
, 10) << "build_prior prior osd." << so
.osd
3007 << " is down" << dendl
;
3008 candidate_blocked_by
[so
.osd
] = lost_at
;
3009 any_down_now
= true;
3015 // if not enough osds survived this interval, and we may have gone rw,
3016 // then we need to wait for one of those osds to recover to
3017 // ensure that we haven't lost any information.
3018 if (!(*pcontdec
)(up_now
) && any_down_now
) {
3019 // fixme: how do we identify a "clean" shutdown anyway?
3020 ldpp_dout(dpp
, 10) << "build_prior possibly went active+rw,"
3021 << " insufficient up; including down osds" << dendl
;
3022 assert(!candidate_blocked_by
.empty());
3025 candidate_blocked_by
.begin(),
3026 candidate_blocked_by
.end());
3030 ldpp_dout(dpp
, 10) << "build_prior final: probe " << probe
3032 << " blocked_by " << blocked_by
3033 << (pg_down
? " pg_down":"")
3038 * pg_query_t - used to ask a peer for information about a pg.
3040 * note: if version=0, type=LOG, then we just provide our full log.
3049 const char *get_type_name() const {
3051 case INFO
: return "info";
3052 case LOG
: return "log";
3053 case MISSING
: return "missing";
3054 case FULLLOG
: return "fulllog";
3055 default: return "???";
3061 pg_history_t history
;
3066 pg_query_t() : type(-1), epoch_sent(0), to(shard_id_t::NO_SHARD
),
3067 from(shard_id_t::NO_SHARD
) {}
3072 const pg_history_t
& h
,
3076 epoch_sent(epoch_sent
),
3077 to(to
), from(from
) {
3085 const pg_history_t
& h
,
3087 : type(t
), since(s
), history(h
),
3088 epoch_sent(epoch_sent
), to(to
), from(from
) {
3092 void encode(bufferlist
&bl
, uint64_t features
) const;
3093 void decode(bufferlist::iterator
&bl
);
3095 void dump(Formatter
*f
) const;
3096 static void generate_test_instances(list
<pg_query_t
*>& o
);
3098 WRITE_CLASS_ENCODER_FEATURES(pg_query_t
)
3100 inline ostream
& operator<<(ostream
& out
, const pg_query_t
& q
) {
3101 out
<< "query(" << q
.get_type_name() << " " << q
.since
;
3102 if (q
.type
== pg_query_t::LOG
)
3103 out
<< " " << q
.history
;
3109 class ObjectModDesc
{
3110 bool can_local_rollback
;
3111 bool rollback_info_completed
;
3113 // version required to decode, reflected in encode/decode version
3114 __u8 max_required_version
= 1;
3118 virtual void append(uint64_t old_offset
) {}
3119 virtual void setattrs(map
<string
, boost::optional
<bufferlist
> > &attrs
) {}
3120 virtual void rmobject(version_t old_version
) {}
3122 * Used to support the unfound_lost_delete log event: if the stashed
3123 * version exists, we unstash it, otherwise, we do nothing. This way
3124 * each replica rolls back to whatever state it had prior to the attempt
3125 * at mark unfound lost delete
3127 virtual void try_rmobject(version_t old_version
) {
3128 rmobject(old_version
);
3130 virtual void create() {}
3131 virtual void update_snaps(const set
<snapid_t
> &old_snaps
) {}
3132 virtual void rollback_extents(
3134 const vector
<pair
<uint64_t, uint64_t> > &extents
) {}
3135 virtual ~Visitor() {}
3137 void visit(Visitor
*visitor
) const;
3138 mutable bufferlist bl
;
3146 ROLLBACK_EXTENTS
= 7
3148 ObjectModDesc() : can_local_rollback(true), rollback_info_completed(false) {
3149 bl
.reassign_to_mempool(mempool::mempool_osd_pglog
);
3151 void claim(ObjectModDesc
&other
) {
3154 can_local_rollback
= other
.can_local_rollback
;
3155 rollback_info_completed
= other
.rollback_info_completed
;
3157 void claim_append(ObjectModDesc
&other
) {
3158 if (!can_local_rollback
|| rollback_info_completed
)
3160 if (!other
.can_local_rollback
) {
3161 mark_unrollbackable();
3164 bl
.claim_append(other
.bl
);
3165 rollback_info_completed
= other
.rollback_info_completed
;
3167 void swap(ObjectModDesc
&other
) {
3171 swap(other
.can_local_rollback
, can_local_rollback
);
3172 swap(other
.rollback_info_completed
, rollback_info_completed
);
3173 swap(other
.max_required_version
, max_required_version
);
3175 void append_id(ModID id
) {
3179 void append(uint64_t old_size
) {
3180 if (!can_local_rollback
|| rollback_info_completed
)
3182 ENCODE_START(1, 1, bl
);
3184 ::encode(old_size
, bl
);
3187 void setattrs(map
<string
, boost::optional
<bufferlist
> > &old_attrs
) {
3188 if (!can_local_rollback
|| rollback_info_completed
)
3190 ENCODE_START(1, 1, bl
);
3191 append_id(SETATTRS
);
3192 ::encode(old_attrs
, bl
);
3195 bool rmobject(version_t deletion_version
) {
3196 if (!can_local_rollback
|| rollback_info_completed
)
3198 ENCODE_START(1, 1, bl
);
3200 ::encode(deletion_version
, bl
);
3202 rollback_info_completed
= true;
3205 bool try_rmobject(version_t deletion_version
) {
3206 if (!can_local_rollback
|| rollback_info_completed
)
3208 ENCODE_START(1, 1, bl
);
3209 append_id(TRY_DELETE
);
3210 ::encode(deletion_version
, bl
);
3212 rollback_info_completed
= true;
3216 if (!can_local_rollback
|| rollback_info_completed
)
3218 rollback_info_completed
= true;
3219 ENCODE_START(1, 1, bl
);
3223 void update_snaps(const set
<snapid_t
> &old_snaps
) {
3224 if (!can_local_rollback
|| rollback_info_completed
)
3226 ENCODE_START(1, 1, bl
);
3227 append_id(UPDATE_SNAPS
);
3228 ::encode(old_snaps
, bl
);
3231 void rollback_extents(
3232 version_t gen
, const vector
<pair
<uint64_t, uint64_t> > &extents
) {
3233 assert(can_local_rollback
);
3234 assert(!rollback_info_completed
);
3235 if (max_required_version
< 2)
3236 max_required_version
= 2;
3237 ENCODE_START(2, 2, bl
);
3238 append_id(ROLLBACK_EXTENTS
);
3240 ::encode(extents
, bl
);
3244 // cannot be rolled back
3245 void mark_unrollbackable() {
3246 can_local_rollback
= false;
3249 bool can_rollback() const {
3250 return can_local_rollback
;
3252 bool empty() const {
3253 return can_local_rollback
&& (bl
.length() == 0);
3256 bool requires_kraken() const {
3257 return max_required_version
>= 2;
3261 * Create fresh copy of bl bytes to avoid keeping large buffers around
3262 * in the case that bl contains ptrs which point into a much larger
3265 void trim_bl() const {
3266 if (bl
.length() > 0)
3269 void encode(bufferlist
&bl
) const;
3270 void decode(bufferlist::iterator
&bl
);
3271 void dump(Formatter
*f
) const;
3272 static void generate_test_instances(list
<ObjectModDesc
*>& o
);
3274 WRITE_CLASS_ENCODER(ObjectModDesc
)
3278 * pg_log_entry_t - single entry/event in pg log
3281 struct pg_log_entry_t
{
3283 MODIFY
= 1, // some unspecified modification (but not *all* modifications)
3284 CLONE
= 2, // cloned object from head
3285 DELETE
= 3, // deleted object
3286 BACKLOG
= 4, // event invented by generate_backlog [deprecated]
3287 LOST_REVERT
= 5, // lost new version, revert to an older version.
3288 LOST_DELETE
= 6, // lost new version, revert to no object (deleted).
3289 LOST_MARK
= 7, // lost new version, now EIO
3290 PROMOTE
= 8, // promoted object from another tier
3291 CLEAN
= 9, // mark an object clean
3292 ERROR
= 10, // write that returned an error
3294 static const char *get_op_name(int op
) {
3320 const char *get_op_name() const {
3321 return get_op_name(op
);
3324 // describes state for a locally-rollbackable entry
3325 ObjectModDesc mod_desc
;
3326 bufferlist snaps
; // only for clone entries
3328 osd_reqid_t reqid
; // caller+tid to uniquely identify request
3329 mempool::osd_pglog::vector
<pair
<osd_reqid_t
, version_t
> > extra_reqids
;
3330 eversion_t version
, prior_version
, reverting_to
;
3331 version_t user_version
; // the user version for this entry
3332 utime_t mtime
; // this is the _user_ mtime, mind you
3333 int32_t return_code
; // only stored for ERRORs for dup detection
3336 bool invalid_hash
; // only when decoding sobject_t based entries
3337 bool invalid_pool
; // only when decoding pool-less hobject based entries
3340 : user_version(0), return_code(0), op(0),
3341 invalid_hash(false), invalid_pool(false) {
3342 snaps
.reassign_to_mempool(mempool::mempool_osd_pglog
);
3344 pg_log_entry_t(int _op
, const hobject_t
& _soid
,
3345 const eversion_t
& v
, const eversion_t
& pv
,
3347 const osd_reqid_t
& rid
, const utime_t
& mt
,
3349 : soid(_soid
), reqid(rid
), version(v
), prior_version(pv
), user_version(uv
),
3350 mtime(mt
), return_code(return_code
), op(_op
),
3351 invalid_hash(false), invalid_pool(false) {
3352 snaps
.reassign_to_mempool(mempool::mempool_osd_pglog
);
3355 bool is_clone() const { return op
== CLONE
; }
3356 bool is_modify() const { return op
== MODIFY
; }
3357 bool is_promote() const { return op
== PROMOTE
; }
3358 bool is_clean() const { return op
== CLEAN
; }
3359 bool is_backlog() const { return op
== BACKLOG
; }
3360 bool is_lost_revert() const { return op
== LOST_REVERT
; }
3361 bool is_lost_delete() const { return op
== LOST_DELETE
; }
3362 bool is_lost_mark() const { return op
== LOST_MARK
; }
3363 bool is_error() const { return op
== ERROR
; }
3365 bool is_update() const {
3367 is_clone() || is_modify() || is_promote() || is_clean() ||
3368 is_backlog() || is_lost_revert() || is_lost_mark();
3370 bool is_delete() const {
3371 return op
== DELETE
|| op
== LOST_DELETE
;
3374 bool can_rollback() const {
3375 return mod_desc
.can_rollback();
3378 void mark_unrollbackable() {
3379 mod_desc
.mark_unrollbackable();
3382 bool requires_kraken() const {
3383 return mod_desc
.requires_kraken();
3386 // Errors are only used for dup detection, whereas
3387 // the index by objects is used by recovery, copy_get,
3388 // and other facilities that don't expect or need to
3389 // be aware of error entries.
3390 bool object_is_indexed() const {
3394 bool reqid_is_indexed() const {
3395 return reqid
!= osd_reqid_t() &&
3396 (op
== MODIFY
|| op
== DELETE
|| op
== ERROR
);
3399 string
get_key_name() const;
3400 void encode_with_checksum(bufferlist
& bl
) const;
3401 void decode_with_checksum(bufferlist::iterator
& p
);
3403 void encode(bufferlist
&bl
) const;
3404 void decode(bufferlist::iterator
&bl
);
3405 void dump(Formatter
*f
) const;
3406 static void generate_test_instances(list
<pg_log_entry_t
*>& o
);
3409 WRITE_CLASS_ENCODER(pg_log_entry_t
)
3411 ostream
& operator<<(ostream
& out
, const pg_log_entry_t
& e
);
3413 struct pg_log_dup_t
{
3414 osd_reqid_t reqid
; // caller+tid to uniquely identify request
3416 version_t user_version
; // the user version for this entry
3417 int32_t return_code
; // only stored for ERRORs for dup detection
3420 : user_version(0), return_code(0)
3422 explicit pg_log_dup_t(const pg_log_entry_t
& entry
)
3423 : reqid(entry
.reqid
), version(entry
.version
),
3424 user_version(entry
.user_version
), return_code(entry
.return_code
)
3426 pg_log_dup_t(const eversion_t
& v
, version_t uv
,
3427 const osd_reqid_t
& rid
, int return_code
)
3428 : reqid(rid
), version(v
), user_version(uv
),
3429 return_code(return_code
)
3432 string
get_key_name() const;
3433 void encode(bufferlist
&bl
) const;
3434 void decode(bufferlist::iterator
&bl
);
3435 void dump(Formatter
*f
) const;
3436 static void generate_test_instances(list
<pg_log_dup_t
*>& o
);
3438 friend std::ostream
& operator<<(std::ostream
& out
, const pg_log_dup_t
& e
);
3440 WRITE_CLASS_ENCODER(pg_log_dup_t
)
3442 std::ostream
& operator<<(std::ostream
& out
, const pg_log_dup_t
& e
);
3445 * pg_log_t - incremental log of recent pg changes.
3447 * serves as a recovery queue for recent changes.
3451 * head - newest entry (update|delete)
3452 * tail - entry previous to oldest (update|delete) for which we have
3453 * complete negative information.
3454 * i.e. we can infer pg contents for any store whose last_update >= tail.
3456 eversion_t head
; // newest entry
3457 eversion_t tail
; // version prior to oldest
3460 // We can rollback rollback-able entries > can_rollback_to
3461 eversion_t can_rollback_to
;
3463 // always <= can_rollback_to, indicates how far stashed rollback
3464 // data can be found
3465 eversion_t rollback_info_trimmed_to
;
3469 mempool::osd_pglog::list
<pg_log_entry_t
> log
;
3471 // entries just for dup op detection ordered oldest to newest
3472 mempool::osd_pglog::list
<pg_log_dup_t
> dups
;
3474 pg_log_t() = default;
3475 pg_log_t(const eversion_t
&last_update
,
3476 const eversion_t
&log_tail
,
3477 const eversion_t
&can_rollback_to
,
3478 const eversion_t
&rollback_info_trimmed_to
,
3479 mempool::osd_pglog::list
<pg_log_entry_t
> &&entries
,
3480 mempool::osd_pglog::list
<pg_log_dup_t
> &&dup_entries
)
3481 : head(last_update
), tail(log_tail
), can_rollback_to(can_rollback_to
),
3482 rollback_info_trimmed_to(rollback_info_trimmed_to
),
3483 log(std::move(entries
)), dups(std::move(dup_entries
)) {}
3484 pg_log_t(const eversion_t
&last_update
,
3485 const eversion_t
&log_tail
,
3486 const eversion_t
&can_rollback_to
,
3487 const eversion_t
&rollback_info_trimmed_to
,
3488 const std::list
<pg_log_entry_t
> &entries
,
3489 const std::list
<pg_log_dup_t
> &dup_entries
)
3490 : head(last_update
), tail(log_tail
), can_rollback_to(can_rollback_to
),
3491 rollback_info_trimmed_to(rollback_info_trimmed_to
) {
3492 for (auto &&entry
: entries
) {
3493 log
.push_back(entry
);
3495 for (auto &&entry
: dup_entries
) {
3496 dups
.push_back(entry
);
3502 rollback_info_trimmed_to
= can_rollback_to
= head
= tail
= z
;
3507 eversion_t
get_rollback_info_trimmed_to() const {
3508 return rollback_info_trimmed_to
;
3510 eversion_t
get_can_rollback_to() const {
3511 return can_rollback_to
;
3515 pg_log_t
split_out_child(pg_t child_pgid
, unsigned split_bits
) {
3516 mempool::osd_pglog::list
<pg_log_entry_t
> oldlog
, childlog
;
3519 eversion_t old_tail
;
3520 unsigned mask
= ~((~0)<<split_bits
);
3521 for (auto i
= oldlog
.begin();
3524 if ((i
->soid
.get_hash() & mask
) == child_pgid
.m_seed
) {
3525 childlog
.push_back(*i
);
3532 // osd_reqid is unique, so it doesn't matter if there are extra
3533 // dup entries in each pg. To avoid storing oid with the dup
3534 // entries, just copy the whole list.
3535 auto childdups(dups
);
3541 rollback_info_trimmed_to
,
3542 std::move(childlog
),
3543 std::move(childdups
));
3546 mempool::osd_pglog::list
<pg_log_entry_t
> rewind_from_head(eversion_t newhead
) {
3547 assert(newhead
>= tail
);
3549 mempool::osd_pglog::list
<pg_log_entry_t
>::iterator p
= log
.end();
3550 mempool::osd_pglog::list
<pg_log_entry_t
> divergent
;
3552 if (p
== log
.begin()) {
3553 // yikes, the whole thing is divergent!
3555 swap(divergent
, log
);
3559 if (p
->version
.version
<= newhead
.version
) {
3561 * look at eversion.version here. we want to avoid a situation like:
3562 * our log: 100'10 (0'0) m 10000004d3a.00000000/head by client4225.1:18529
3563 * new log: 122'10 (0'0) m 10000004d3a.00000000/head by client4225.1:18529
3564 * lower_bound = 100'9
3565 * i.e, same request, different version. If the eversion.version is > the
3566 * lower_bound, we it is divergent.
3569 divergent
.splice(divergent
.begin(), log
, p
, log
.end());
3572 assert(p
->version
> newhead
);
3576 if (can_rollback_to
> newhead
)
3577 can_rollback_to
= newhead
;
3579 if (rollback_info_trimmed_to
> newhead
)
3580 rollback_info_trimmed_to
= newhead
;
3585 bool empty() const {
3590 return head
.version
== 0 && head
.epoch
== 0;
3593 size_t approx_size() const {
3594 return head
.version
- tail
.version
;
3597 static void filter_log(spg_t import_pgid
, const OSDMap
&curmap
,
3598 const string
&hit_set_namespace
, const pg_log_t
&in
,
3599 pg_log_t
&out
, pg_log_t
&reject
);
3602 * copy entries from the tail of another pg_log_t
3604 * @param other pg_log_t to copy from
3605 * @param from copy entries after this version
3607 void copy_after(const pg_log_t
&other
, eversion_t from
);
3610 * copy a range of entries from another pg_log_t
3612 * @param other pg_log_t to copy from
3613 * @param from copy entries after this version
3614 * @param to up to and including this version
3616 void copy_range(const pg_log_t
&other
, eversion_t from
, eversion_t to
);
3619 * copy up to N entries
3621 * @param other source log
3622 * @param max max number of entries to copy
3624 void copy_up_to(const pg_log_t
&other
, int max
);
3626 ostream
& print(ostream
& out
) const;
3628 void encode(bufferlist
&bl
) const;
3629 void decode(bufferlist::iterator
&bl
, int64_t pool
= -1);
3630 void dump(Formatter
*f
) const;
3631 static void generate_test_instances(list
<pg_log_t
*>& o
);
3633 WRITE_CLASS_ENCODER(pg_log_t
)
3635 inline ostream
& operator<<(ostream
& out
, const pg_log_t
& log
)
3637 out
<< "log((" << log
.tail
<< "," << log
.head
<< "], crt="
3638 << log
.get_can_rollback_to() << ")";
3644 * pg_missing_t - summary of missing objects.
3646 * kept in memory, as a supplement to pg_log_t
3647 * also used to pass missing info in messages.
3649 struct pg_missing_item
{
3650 eversion_t need
, have
;
3651 enum missing_flags_t
{
3655 pg_missing_item() : flags(FLAG_NONE
) {}
3656 explicit pg_missing_item(eversion_t n
) : need(n
), flags(FLAG_NONE
) {} // have no old version
3657 pg_missing_item(eversion_t n
, eversion_t h
, bool is_delete
=false) : need(n
), have(h
) {
3658 set_delete(is_delete
);
3661 void encode(bufferlist
& bl
, uint64_t features
) const {
3662 if (HAVE_FEATURE(features
, OSD_RECOVERY_DELETES
)) {
3663 // encoding a zeroed eversion_t to differentiate between this and
3664 // legacy unversioned encoding - a need value of 0'0 is not
3665 // possible. This can be replaced with the legacy encoding
3666 // macros post-luminous.
3671 ::encode(static_cast<uint8_t>(flags
), bl
);
3673 // legacy unversioned encoding
3678 void decode(bufferlist::iterator
& bl
) {
3681 if (e
!= eversion_t()) {
3682 // legacy encoding, this is the need value
3690 flags
= static_cast<missing_flags_t
>(f
);
3694 void set_delete(bool is_delete
) {
3695 flags
= is_delete
? FLAG_DELETE
: FLAG_NONE
;
3698 bool is_delete() const {
3699 return (flags
& FLAG_DELETE
) == FLAG_DELETE
;
3702 string
flag_str() const {
3703 if (flags
== FLAG_NONE
) {
3710 void dump(Formatter
*f
) const {
3711 f
->dump_stream("need") << need
;
3712 f
->dump_stream("have") << have
;
3713 f
->dump_stream("flags") << flag_str();
3715 static void generate_test_instances(list
<pg_missing_item
*>& o
) {
3716 o
.push_back(new pg_missing_item
);
3717 o
.push_back(new pg_missing_item
);
3718 o
.back()->need
= eversion_t(1, 2);
3719 o
.back()->have
= eversion_t(1, 1);
3720 o
.push_back(new pg_missing_item
);
3721 o
.back()->need
= eversion_t(3, 5);
3722 o
.back()->have
= eversion_t(3, 4);
3723 o
.back()->flags
= FLAG_DELETE
;
3725 bool operator==(const pg_missing_item
&rhs
) const {
3726 return need
== rhs
.need
&& have
== rhs
.have
&& flags
== rhs
.flags
;
3728 bool operator!=(const pg_missing_item
&rhs
) const {
3729 return !(*this == rhs
);
3732 WRITE_CLASS_ENCODER_FEATURES(pg_missing_item
)
3733 ostream
& operator<<(ostream
& out
, const pg_missing_item
&item
);
3735 class pg_missing_const_i
{
3737 virtual const map
<hobject_t
, pg_missing_item
> &
3738 get_items() const = 0;
3739 virtual const map
<version_t
, hobject_t
> &get_rmissing() const = 0;
3740 virtual bool get_may_include_deletes() const = 0;
3741 virtual unsigned int num_missing() const = 0;
3742 virtual bool have_missing() const = 0;
3743 virtual bool is_missing(const hobject_t
& oid
, pg_missing_item
*out
= nullptr) const = 0;
3744 virtual bool is_missing(const hobject_t
& oid
, eversion_t v
) const = 0;
3745 virtual eversion_t
have_old(const hobject_t
& oid
) const = 0;
3746 virtual ~pg_missing_const_i() {}
3750 template <bool Track
>
3751 class ChangeTracker
{
3753 void changed(const hobject_t
&obj
) {}
3754 template <typename F
>
3755 void get_changed(F
&&f
) const {}
3757 bool is_clean() const {
3762 class ChangeTracker
<true> {
3763 set
<hobject_t
> _changed
;
3765 void changed(const hobject_t
&obj
) {
3766 _changed
.insert(obj
);
3768 template <typename F
>
3769 void get_changed(F
&&f
) const {
3770 for (auto const &i
: _changed
) {
3777 bool is_clean() const {
3778 return _changed
.empty();
3782 template <bool TrackChanges
>
3783 class pg_missing_set
: public pg_missing_const_i
{
3784 using item
= pg_missing_item
;
3785 map
<hobject_t
, item
> missing
; // oid -> (need v, have v)
3786 map
<version_t
, hobject_t
> rmissing
; // v -> oid
3787 ChangeTracker
<TrackChanges
> tracker
;
3790 pg_missing_set() = default;
3792 template <typename missing_type
>
3793 pg_missing_set(const missing_type
&m
) {
3794 missing
= m
.get_items();
3795 rmissing
= m
.get_rmissing();
3796 may_include_deletes
= m
.get_may_include_deletes();
3797 for (auto &&i
: missing
)
3798 tracker
.changed(i
.first
);
3801 bool may_include_deletes
= false;
3803 const map
<hobject_t
, item
> &get_items() const override
{
3806 const map
<version_t
, hobject_t
> &get_rmissing() const override
{
3809 bool get_may_include_deletes() const override
{
3810 return may_include_deletes
;
3812 unsigned int num_missing() const override
{
3813 return missing
.size();
3815 bool have_missing() const override
{
3816 return !missing
.empty();
3818 bool is_missing(const hobject_t
& oid
, pg_missing_item
*out
= nullptr) const override
{
3819 auto iter
= missing
.find(oid
);
3820 if (iter
== missing
.end())
3823 *out
= iter
->second
;
3826 bool is_missing(const hobject_t
& oid
, eversion_t v
) const override
{
3827 map
<hobject_t
, item
>::const_iterator m
=
3829 if (m
== missing
.end())
3831 const item
&item(m
->second
);
3836 eversion_t
have_old(const hobject_t
& oid
) const override
{
3837 map
<hobject_t
, item
>::const_iterator m
=
3839 if (m
== missing
.end())
3840 return eversion_t();
3841 const item
&item(m
->second
);
3845 void claim(pg_missing_set
& o
) {
3846 static_assert(!TrackChanges
, "Can't use claim with TrackChanges");
3847 missing
.swap(o
.missing
);
3848 rmissing
.swap(o
.rmissing
);
3852 * this needs to be called in log order as we extend the log. it
3853 * assumes missing is accurate up through the previous log entry.
3855 void add_next_event(const pg_log_entry_t
& e
) {
3856 map
<hobject_t
, item
>::iterator missing_it
;
3857 missing_it
= missing
.find(e
.soid
);
3858 bool is_missing_divergent_item
= missing_it
!= missing
.end();
3859 if (e
.prior_version
== eversion_t() || e
.is_clone()) {
3861 if (is_missing_divergent_item
) { // use iterator
3862 rmissing
.erase((missing_it
->second
).need
.version
);
3863 missing_it
->second
= item(e
.version
, eversion_t(), e
.is_delete()); // .have = nil
3864 } else // create new element in missing map
3865 missing
[e
.soid
] = item(e
.version
, eversion_t(), e
.is_delete()); // .have = nil
3866 } else if (is_missing_divergent_item
) {
3867 // already missing (prior).
3868 rmissing
.erase((missing_it
->second
).need
.version
);
3869 (missing_it
->second
).need
= e
.version
; // leave .have unchanged.
3870 missing_it
->second
.set_delete(e
.is_delete());
3871 } else if (e
.is_backlog()) {
3872 // May not have prior version
3873 assert(0 == "these don't exist anymore");
3875 // not missing, we must have prior_version (if any)
3876 assert(!is_missing_divergent_item
);
3877 missing
[e
.soid
] = item(e
.version
, e
.prior_version
, e
.is_delete());
3879 rmissing
[e
.version
.version
] = e
.soid
;
3880 tracker
.changed(e
.soid
);
3883 void revise_need(hobject_t oid
, eversion_t need
, bool is_delete
) {
3884 if (missing
.count(oid
)) {
3885 rmissing
.erase(missing
[oid
].need
.version
);
3886 missing
[oid
].need
= need
; // no not adjust .have
3887 missing
[oid
].set_delete(is_delete
);
3889 missing
[oid
] = item(need
, eversion_t(), is_delete
);
3891 rmissing
[need
.version
] = oid
;
3893 tracker
.changed(oid
);
3896 void revise_have(hobject_t oid
, eversion_t have
) {
3897 if (missing
.count(oid
)) {
3898 tracker
.changed(oid
);
3899 missing
[oid
].have
= have
;
3903 void add(const hobject_t
& oid
, eversion_t need
, eversion_t have
,
3905 missing
[oid
] = item(need
, have
, is_delete
);
3906 rmissing
[need
.version
] = oid
;
3907 tracker
.changed(oid
);
3910 void rm(const hobject_t
& oid
, eversion_t v
) {
3911 std::map
<hobject_t
, item
>::iterator p
= missing
.find(oid
);
3912 if (p
!= missing
.end() && p
->second
.need
<= v
)
3916 void rm(std::map
<hobject_t
, item
>::const_iterator m
) {
3917 tracker
.changed(m
->first
);
3918 rmissing
.erase(m
->second
.need
.version
);
3922 void got(const hobject_t
& oid
, eversion_t v
) {
3923 std::map
<hobject_t
, item
>::iterator p
= missing
.find(oid
);
3924 assert(p
!= missing
.end());
3925 assert(p
->second
.need
<= v
|| p
->second
.is_delete());
3929 void got(std::map
<hobject_t
, item
>::const_iterator m
) {
3930 tracker
.changed(m
->first
);
3931 rmissing
.erase(m
->second
.need
.version
);
3937 unsigned split_bits
,
3938 pg_missing_set
*omissing
) {
3939 omissing
->may_include_deletes
= may_include_deletes
;
3940 unsigned mask
= ~((~0)<<split_bits
);
3941 for (map
<hobject_t
, item
>::iterator i
= missing
.begin();
3944 if ((i
->first
.get_hash() & mask
) == child_pgid
.m_seed
) {
3945 omissing
->add(i
->first
, i
->second
.need
, i
->second
.have
,
3946 i
->second
.is_delete());
3955 for (auto const &i
: missing
)
3956 tracker
.changed(i
.first
);
3961 void encode(bufferlist
&bl
) const {
3962 ENCODE_START(4, 2, bl
);
3963 ::encode(missing
, bl
, may_include_deletes
? CEPH_FEATURE_OSD_RECOVERY_DELETES
: 0);
3964 ::encode(may_include_deletes
, bl
);
3967 void decode(bufferlist::iterator
&bl
, int64_t pool
= -1) {
3968 for (auto const &i
: missing
)
3969 tracker
.changed(i
.first
);
3970 DECODE_START_LEGACY_COMPAT_LEN(4, 2, 2, bl
);
3971 ::decode(missing
, bl
);
3972 if (struct_v
>= 4) {
3973 ::decode(may_include_deletes
, bl
);
3978 // Handle hobject_t upgrade
3979 map
<hobject_t
, item
> tmp
;
3980 for (map
<hobject_t
, item
>::iterator i
=
3984 if (!i
->first
.is_max() && i
->first
.pool
== -1) {
3985 hobject_t
to_insert(i
->first
);
3986 to_insert
.pool
= pool
;
3987 tmp
[to_insert
] = i
->second
;
3993 missing
.insert(tmp
.begin(), tmp
.end());
3996 for (map
<hobject_t
,item
>::iterator it
=
3998 it
!= missing
.end();
4000 rmissing
[it
->second
.need
.version
] = it
->first
;
4001 for (auto const &i
: missing
)
4002 tracker
.changed(i
.first
);
4004 void dump(Formatter
*f
) const {
4005 f
->open_array_section("missing");
4006 for (map
<hobject_t
,item
>::const_iterator p
=
4007 missing
.begin(); p
!= missing
.end(); ++p
) {
4008 f
->open_object_section("item");
4009 f
->dump_stream("object") << p
->first
;
4014 f
->dump_bool("may_include_deletes", may_include_deletes
);
4016 template <typename F
>
4017 void filter_objects(F
&&f
) {
4018 for (auto i
= missing
.begin(); i
!= missing
.end();) {
4026 static void generate_test_instances(list
<pg_missing_set
*>& o
) {
4027 o
.push_back(new pg_missing_set
);
4028 o
.push_back(new pg_missing_set
);
4030 hobject_t(object_t("foo"), "foo", 123, 456, 0, ""),
4031 eversion_t(5, 6), eversion_t(5, 1), false);
4032 o
.push_back(new pg_missing_set
);
4034 hobject_t(object_t("foo"), "foo", 123, 456, 0, ""),
4035 eversion_t(5, 6), eversion_t(5, 1), true);
4036 o
.back()->may_include_deletes
= true;
4038 template <typename F
>
4039 void get_changed(F
&&f
) const {
4040 tracker
.get_changed(f
);
4045 bool is_clean() const {
4046 return tracker
.is_clean();
4048 template <typename missing_t
>
4049 bool debug_verify_from_init(
4050 const missing_t
&init_missing
,
4051 ostream
*oss
) const {
4054 auto check_missing(init_missing
.get_items());
4055 tracker
.get_changed([&](const hobject_t
&hoid
) {
4056 check_missing
.erase(hoid
);
4057 if (missing
.count(hoid
)) {
4058 check_missing
.insert(*(missing
.find(hoid
)));
4062 if (check_missing
.size() != missing
.size()) {
4064 *oss
<< "Size mismatch, check: " << check_missing
.size()
4065 << ", actual: " << missing
.size() << "\n";
4069 for (auto &i
: missing
) {
4070 if (!check_missing
.count(i
.first
)) {
4072 *oss
<< "check_missing missing " << i
.first
<< "\n";
4074 } else if (check_missing
[i
.first
] != i
.second
) {
4076 *oss
<< "check_missing missing item mismatch on " << i
.first
4077 << ", check: " << check_missing
[i
.first
]
4078 << ", actual: " << i
.second
<< "\n";
4083 *oss
<< "check_missing: " << check_missing
<< "\n";
4084 set
<hobject_t
> changed
;
4085 tracker
.get_changed([&](const hobject_t
&hoid
) { changed
.insert(hoid
); });
4086 *oss
<< "changed: " << changed
<< "\n";
4091 template <bool TrackChanges
>
4093 const pg_missing_set
<TrackChanges
> &c
, bufferlist
&bl
, uint64_t features
=0) {
4096 ENCODE_DUMP_POST(cl
);
4098 template <bool TrackChanges
>
4099 void decode(pg_missing_set
<TrackChanges
> &c
, bufferlist::iterator
&p
) {
4102 template <bool TrackChanges
>
4103 ostream
& operator<<(ostream
& out
, const pg_missing_set
<TrackChanges
> &missing
)
4105 out
<< "missing(" << missing
.num_missing()
4106 << " may_include_deletes = " << missing
.may_include_deletes
;
4107 //if (missing.num_lost()) out << ", " << missing.num_lost() << " lost";
4112 using pg_missing_t
= pg_missing_set
<false>;
4113 using pg_missing_tracker_t
= pg_missing_set
<true>;
4117 * pg list objects response format
4120 struct pg_nls_response_t
{
4121 collection_list_handle_t handle
;
4122 list
<librados::ListObjectImpl
> entries
;
4124 void encode(bufferlist
& bl
) const {
4125 ENCODE_START(1, 1, bl
);
4126 ::encode(handle
, bl
);
4127 __u32 n
= (__u32
)entries
.size();
4129 for (list
<librados::ListObjectImpl
>::const_iterator i
= entries
.begin(); i
!= entries
.end(); ++i
) {
4130 ::encode(i
->nspace
, bl
);
4131 ::encode(i
->oid
, bl
);
4132 ::encode(i
->locator
, bl
);
4136 void decode(bufferlist::iterator
& bl
) {
4137 DECODE_START(1, bl
);
4138 ::decode(handle
, bl
);
4143 librados::ListObjectImpl i
;
4144 ::decode(i
.nspace
, bl
);
4145 ::decode(i
.oid
, bl
);
4146 ::decode(i
.locator
, bl
);
4147 entries
.push_back(i
);
4151 void dump(Formatter
*f
) const {
4152 f
->dump_stream("handle") << handle
;
4153 f
->open_array_section("entries");
4154 for (list
<librados::ListObjectImpl
>::const_iterator p
= entries
.begin(); p
!= entries
.end(); ++p
) {
4155 f
->open_object_section("object");
4156 f
->dump_string("namespace", p
->nspace
);
4157 f
->dump_string("object", p
->oid
);
4158 f
->dump_string("key", p
->locator
);
4163 static void generate_test_instances(list
<pg_nls_response_t
*>& o
) {
4164 o
.push_back(new pg_nls_response_t
);
4165 o
.push_back(new pg_nls_response_t
);
4166 o
.back()->handle
= hobject_t(object_t("hi"), "key", 1, 2, -1, "");
4167 o
.back()->entries
.push_back(librados::ListObjectImpl("", "one", ""));
4168 o
.back()->entries
.push_back(librados::ListObjectImpl("", "two", "twokey"));
4169 o
.back()->entries
.push_back(librados::ListObjectImpl("", "three", ""));
4170 o
.push_back(new pg_nls_response_t
);
4171 o
.back()->handle
= hobject_t(object_t("hi"), "key", 3, 4, -1, "");
4172 o
.back()->entries
.push_back(librados::ListObjectImpl("n1", "n1one", ""));
4173 o
.back()->entries
.push_back(librados::ListObjectImpl("n1", "n1two", "n1twokey"));
4174 o
.back()->entries
.push_back(librados::ListObjectImpl("n1", "n1three", ""));
4175 o
.push_back(new pg_nls_response_t
);
4176 o
.back()->handle
= hobject_t(object_t("hi"), "key", 5, 6, -1, "");
4177 o
.back()->entries
.push_back(librados::ListObjectImpl("", "one", ""));
4178 o
.back()->entries
.push_back(librados::ListObjectImpl("", "two", "twokey"));
4179 o
.back()->entries
.push_back(librados::ListObjectImpl("", "three", ""));
4180 o
.back()->entries
.push_back(librados::ListObjectImpl("n1", "n1one", ""));
4181 o
.back()->entries
.push_back(librados::ListObjectImpl("n1", "n1two", "n1twokey"));
4182 o
.back()->entries
.push_back(librados::ListObjectImpl("n1", "n1three", ""));
4186 WRITE_CLASS_ENCODER(pg_nls_response_t
)
4188 // For backwards compatibility with older OSD requests
4189 struct pg_ls_response_t
{
4190 collection_list_handle_t handle
;
4191 list
<pair
<object_t
, string
> > entries
;
4193 void encode(bufferlist
& bl
) const {
4196 ::encode(handle
, bl
);
4197 ::encode(entries
, bl
);
4199 void decode(bufferlist::iterator
& bl
) {
4203 ::decode(handle
, bl
);
4204 ::decode(entries
, bl
);
4206 void dump(Formatter
*f
) const {
4207 f
->dump_stream("handle") << handle
;
4208 f
->open_array_section("entries");
4209 for (list
<pair
<object_t
, string
> >::const_iterator p
= entries
.begin(); p
!= entries
.end(); ++p
) {
4210 f
->open_object_section("object");
4211 f
->dump_stream("object") << p
->first
;
4212 f
->dump_string("key", p
->second
);
4217 static void generate_test_instances(list
<pg_ls_response_t
*>& o
) {
4218 o
.push_back(new pg_ls_response_t
);
4219 o
.push_back(new pg_ls_response_t
);
4220 o
.back()->handle
= hobject_t(object_t("hi"), "key", 1, 2, -1, "");
4221 o
.back()->entries
.push_back(make_pair(object_t("one"), string()));
4222 o
.back()->entries
.push_back(make_pair(object_t("two"), string("twokey")));
4226 WRITE_CLASS_ENCODER(pg_ls_response_t
)
4229 * object_copy_cursor_t
4231 struct object_copy_cursor_t
{
4232 uint64_t data_offset
;
4238 object_copy_cursor_t()
4240 attr_complete(false),
4241 data_complete(false),
4242 omap_complete(false)
4245 bool is_initial() const {
4246 return !attr_complete
&& data_offset
== 0 && omap_offset
.empty();
4248 bool is_complete() const {
4249 return attr_complete
&& data_complete
&& omap_complete
;
4252 static void generate_test_instances(list
<object_copy_cursor_t
*>& o
);
4253 void encode(bufferlist
& bl
) const;
4254 void decode(bufferlist::iterator
&bl
);
4255 void dump(Formatter
*f
) const;
4257 WRITE_CLASS_ENCODER(object_copy_cursor_t
)
4260 * object_copy_data_t
4262 * Return data from a copy request. The semantics are a little strange
4263 * as a result of the encoding's heritage.
4265 * In particular, the sender unconditionally fills in the cursor (from what
4266 * it receives and sends), the size, and the mtime, but is responsible for
4267 * figuring out whether it should put any data in the attrs, data, or
4268 * omap members (corresponding to xattrs, object data, and the omap entries)
4269 * based on external data (the client includes a max amount to return with
4270 * the copy request). The client then looks into the attrs, data, and/or omap
4271 * based on the contents of the cursor.
4273 struct object_copy_data_t
{
4275 FLAG_DATA_DIGEST
= 1<<0,
4276 FLAG_OMAP_DIGEST
= 1<<1,
4278 object_copy_cursor_t cursor
;
4281 uint32_t data_digest
, omap_digest
;
4283 map
<string
, bufferlist
> attrs
;
4285 bufferlist omap_header
;
4286 bufferlist omap_data
;
4288 /// which snaps we are defined for (if a snap and not the head)
4289 vector
<snapid_t
> snaps
;
4290 ///< latest snap seq for the object (if head)
4293 ///< recent reqids on this object
4294 mempool::osd_pglog::vector
<pair
<osd_reqid_t
, version_t
> > reqids
;
4296 uint64_t truncate_seq
;
4297 uint64_t truncate_size
;
4300 object_copy_data_t() :
4301 size((uint64_t)-1), data_digest(-1),
4302 omap_digest(-1), flags(0),
4306 static void generate_test_instances(list
<object_copy_data_t
*>& o
);
4307 void encode(bufferlist
& bl
, uint64_t features
) const;
4308 void decode(bufferlist::iterator
& bl
);
4309 void dump(Formatter
*f
) const;
4311 WRITE_CLASS_ENCODER_FEATURES(object_copy_data_t
)
4316 struct pg_create_t
{
4317 epoch_t created
; // epoch pg created
4318 pg_t parent
; // split from parent (if != pg_t())
4322 : created(0), split_bits(0) {}
4323 pg_create_t(unsigned c
, pg_t p
, int s
)
4324 : created(c
), parent(p
), split_bits(s
) {}
4326 void encode(bufferlist
&bl
) const;
4327 void decode(bufferlist::iterator
&bl
);
4328 void dump(Formatter
*f
) const;
4329 static void generate_test_instances(list
<pg_create_t
*>& o
);
4331 WRITE_CLASS_ENCODER(pg_create_t
)
4333 // -----------------------------------------
4335 struct osd_peer_stat_t
{
4338 osd_peer_stat_t() { }
4340 void encode(bufferlist
&bl
) const;
4341 void decode(bufferlist::iterator
&bl
);
4342 void dump(Formatter
*f
) const;
4343 static void generate_test_instances(list
<osd_peer_stat_t
*>& o
);
4345 WRITE_CLASS_ENCODER(osd_peer_stat_t
)
4347 ostream
& operator<<(ostream
& out
, const osd_peer_stat_t
&stat
);
4350 // -----------------------------------------
4352 class ObjectExtent
{
4354 * ObjectExtents are used for specifying IO behavior against RADOS
4355 * objects when one is using the ObjectCacher.
4357 * To use this in a real system, *every member* must be filled
4358 * out correctly. In particular, make sure to initialize the
4359 * oloc correctly, as its default values are deliberate poison
4360 * and will cause internal ObjectCacher asserts.
4362 * Similarly, your buffer_extents vector *must* specify a total
4363 * size equal to your length. If the buffer_extents inadvertently
4364 * contain less space than the length member specifies, you
4365 * will get unintelligible asserts deep in the ObjectCacher.
4367 * If you are trying to do testing and don't care about actual
4368 * RADOS function, the simplest thing to do is to initialize
4369 * the ObjectExtent (truncate_size can be 0), create a single entry
4370 * in buffer_extents matching the length, and set oloc.pool to 0.
4373 object_t oid
; // object id
4375 uint64_t offset
; // in object
4376 uint64_t length
; // in object
4377 uint64_t truncate_size
; // in object
4379 object_locator_t oloc
; // object locator (pool etc)
4381 vector
<pair
<uint64_t,uint64_t> > buffer_extents
; // off -> len. extents in buffer being mapped (may be fragmented bc of striping!)
4383 ObjectExtent() : objectno(0), offset(0), length(0), truncate_size(0) {}
4384 ObjectExtent(object_t o
, uint64_t ono
, uint64_t off
, uint64_t l
, uint64_t ts
) :
4385 oid(o
), objectno(ono
), offset(off
), length(l
), truncate_size(ts
) { }
4388 inline ostream
& operator<<(ostream
& out
, const ObjectExtent
&ex
)
4390 return out
<< "extent("
4391 << ex
.oid
<< " (" << ex
.objectno
<< ") in " << ex
.oloc
4392 << " " << ex
.offset
<< "~" << ex
.length
4393 << " -> " << ex
.buffer_extents
4398 // ---------------------------------------
4400 class OSDSuperblock
{
4402 uuid_d cluster_fsid
, osd_fsid
;
4403 int32_t whoami
; // my role in this fs.
4404 epoch_t current_epoch
; // most recent epoch
4405 epoch_t oldest_map
, newest_map
; // oldest/newest maps we have.
4408 CompatSet compat_features
;
4410 // last interval over which i mounted and was then active
4411 epoch_t mounted
; // last epoch i mounted
4412 epoch_t clean_thru
; // epoch i was active and clean thru
4416 current_epoch(0), oldest_map(0), newest_map(0), weight(0),
4417 mounted(0), clean_thru(0) {
4420 void encode(bufferlist
&bl
) const;
4421 void decode(bufferlist::iterator
&bl
);
4422 void dump(Formatter
*f
) const;
4423 static void generate_test_instances(list
<OSDSuperblock
*>& o
);
4425 WRITE_CLASS_ENCODER(OSDSuperblock
)
4427 inline ostream
& operator<<(ostream
& out
, const OSDSuperblock
& sb
)
4429 return out
<< "sb(" << sb
.cluster_fsid
4430 << " osd." << sb
.whoami
4431 << " " << sb
.osd_fsid
4432 << " e" << sb
.current_epoch
4433 << " [" << sb
.oldest_map
<< "," << sb
.newest_map
<< "]"
4434 << " lci=[" << sb
.mounted
<< "," << sb
.clean_thru
<< "]"
4447 * attached to object head. describes most recent snap context, and
4448 * set of existing clones.
4453 vector
<snapid_t
> snaps
; // descending
4454 vector
<snapid_t
> clones
; // ascending
4455 map
<snapid_t
, interval_set
<uint64_t> > clone_overlap
; // overlap w/ next newest
4456 map
<snapid_t
, uint64_t> clone_size
;
4457 map
<snapid_t
, vector
<snapid_t
>> clone_snaps
; // descending
4459 SnapSet() : seq(0), head_exists(false) {}
4460 explicit SnapSet(bufferlist
& bl
) {
4461 bufferlist::iterator p
= bl
.begin();
4465 bool is_legacy() const {
4466 return clone_snaps
.size() < clones
.size() || !head_exists
;
4469 /// populate SnapSet from a librados::snap_set_t
4470 void from_snap_set(const librados::snap_set_t
& ss
, bool legacy
);
4472 /// get space accounted to clone
4473 uint64_t get_clone_bytes(snapid_t clone
) const;
4475 void encode(bufferlist
& bl
) const;
4476 void decode(bufferlist::iterator
& bl
);
4477 void dump(Formatter
*f
) const;
4478 static void generate_test_instances(list
<SnapSet
*>& o
);
4480 SnapContext
get_ssc_as_of(snapid_t as_of
) const {
4483 for (vector
<snapid_t
>::const_iterator i
= snaps
.begin();
4487 out
.snaps
.push_back(*i
);
4492 // return min element of snaps > after, return max if no such element
4493 snapid_t
get_first_snap_after(snapid_t after
, snapid_t max
) const {
4494 for (vector
<snapid_t
>::const_reverse_iterator i
= snaps
.rbegin();
4503 SnapSet
get_filtered(const pg_pool_t
&pinfo
) const;
4504 void filter(const pg_pool_t
&pinfo
);
4506 WRITE_CLASS_ENCODER(SnapSet
)
4508 ostream
& operator<<(ostream
& out
, const SnapSet
& cs
);
4513 #define SS_ATTR "snapset"
4515 struct watch_info_t
{
4517 uint32_t timeout_seconds
;
4520 watch_info_t() : cookie(0), timeout_seconds(0) { }
4521 watch_info_t(uint64_t c
, uint32_t t
, const entity_addr_t
& a
) : cookie(c
), timeout_seconds(t
), addr(a
) {}
4523 void encode(bufferlist
& bl
, uint64_t features
) const;
4524 void decode(bufferlist::iterator
& bl
);
4525 void dump(Formatter
*f
) const;
4526 static void generate_test_instances(list
<watch_info_t
*>& o
);
4528 WRITE_CLASS_ENCODER_FEATURES(watch_info_t
)
4530 static inline bool operator==(const watch_info_t
& l
, const watch_info_t
& r
) {
4531 return l
.cookie
== r
.cookie
&& l
.timeout_seconds
== r
.timeout_seconds
4532 && l
.addr
== r
.addr
;
4535 static inline ostream
& operator<<(ostream
& out
, const watch_info_t
& w
) {
4536 return out
<< "watch(cookie " << w
.cookie
<< " " << w
.timeout_seconds
<< "s"
4537 << " " << w
.addr
<< ")";
4540 struct notify_info_t
{
4547 static inline ostream
& operator<<(ostream
& out
, const notify_info_t
& n
) {
4548 return out
<< "notify(cookie " << n
.cookie
4549 << " notify" << n
.notify_id
4550 << " " << n
.timeout
<< "s)";
4553 struct object_info_t
;
4554 struct object_manifest_t
{
4557 TYPE_REDIRECT
= 1, // start with this
4558 TYPE_CHUNKED
= 2, // do this later
4560 uint8_t type
; // redirect, chunked, ...
4561 hobject_t redirect_target
;
4563 object_manifest_t() : type(0) { }
4564 object_manifest_t(uint8_t type
, const hobject_t
& redirect_target
)
4565 : type(type
), redirect_target(redirect_target
) { }
4567 bool is_empty() const {
4568 return type
== TYPE_NONE
;
4570 bool is_redirect() const {
4571 return type
== TYPE_REDIRECT
;
4573 bool is_chunked() const {
4574 return type
== TYPE_CHUNKED
;
4576 static const char *get_type_name(uint8_t m
) {
4578 case TYPE_NONE
: return "none";
4579 case TYPE_REDIRECT
: return "redirect";
4580 case TYPE_CHUNKED
: return "chunked";
4581 default: return "unknown";
4584 const char *get_type_name() const {
4585 return get_type_name(type
);
4587 static void generate_test_instances(list
<object_manifest_t
*>& o
);
4588 void encode(bufferlist
&bl
) const;
4589 void decode(bufferlist::iterator
&bl
);
4590 void dump(Formatter
*f
) const;
4591 friend ostream
& operator<<(ostream
& out
, const object_info_t
& oi
);
4593 WRITE_CLASS_ENCODER(object_manifest_t
)
4594 ostream
& operator<<(ostream
& out
, const object_manifest_t
& oi
);
4596 struct object_info_t
{
4598 eversion_t version
, prior_version
;
4599 version_t user_version
;
4600 osd_reqid_t last_reqid
;
4604 utime_t local_mtime
; // local mtime
4606 // note: these are currently encoded into a total 16 bits; see
4607 // encode()/decode() for the weirdness.
4610 FLAG_WHITEOUT
= 1<<1, // object logically does not exist
4611 FLAG_DIRTY
= 1<<2, // object has been modified since last flushed or undirtied
4612 FLAG_OMAP
= 1 << 3, // has (or may have) some/any omap data
4613 FLAG_DATA_DIGEST
= 1 << 4, // has data crc
4614 FLAG_OMAP_DIGEST
= 1 << 5, // has omap crc
4615 FLAG_CACHE_PIN
= 1 << 6, // pin the object in cache tier
4616 FLAG_MANIFEST
= 1 << 7, // has manifest
4618 FLAG_USES_TMAP
= 1<<8, // deprecated; no longer used.
4623 static string
get_flag_string(flag_t flags
) {
4625 if (flags
& FLAG_LOST
)
4627 if (flags
& FLAG_WHITEOUT
)
4629 if (flags
& FLAG_DIRTY
)
4631 if (flags
& FLAG_USES_TMAP
)
4633 if (flags
& FLAG_OMAP
)
4635 if (flags
& FLAG_DATA_DIGEST
)
4636 s
+= "|data_digest";
4637 if (flags
& FLAG_OMAP_DIGEST
)
4638 s
+= "|omap_digest";
4639 if (flags
& FLAG_CACHE_PIN
)
4641 if (flags
& FLAG_MANIFEST
)
4647 string
get_flag_string() const {
4648 return get_flag_string(flags
);
4651 /// [clone] descending. pre-luminous; moved to SnapSet
4652 vector
<snapid_t
> legacy_snaps
;
4654 uint64_t truncate_seq
, truncate_size
;
4656 map
<pair
<uint64_t, entity_name_t
>, watch_info_t
> watchers
;
4658 // opportunistic checksums; may or may not be present
4659 __u32 data_digest
; ///< data crc32c
4660 __u32 omap_digest
; ///< omap crc32c
4662 // alloc hint attribute
4663 uint64_t expected_object_size
, expected_write_size
;
4664 uint32_t alloc_hint_flags
;
4666 struct object_manifest_t manifest
;
4668 void copy_user_bits(const object_info_t
& other
);
4670 static ps_t
legacy_object_locator_to_ps(const object_t
&oid
,
4671 const object_locator_t
&loc
);
4673 bool test_flag(flag_t f
) const {
4674 return (flags
& f
) == f
;
4676 void set_flag(flag_t f
) {
4677 flags
= (flag_t
)(flags
| f
);
4679 void clear_flag(flag_t f
) {
4680 flags
= (flag_t
)(flags
& ~f
);
4682 bool is_lost() const {
4683 return test_flag(FLAG_LOST
);
4685 bool is_whiteout() const {
4686 return test_flag(FLAG_WHITEOUT
);
4688 bool is_dirty() const {
4689 return test_flag(FLAG_DIRTY
);
4691 bool is_omap() const {
4692 return test_flag(FLAG_OMAP
);
4694 bool is_data_digest() const {
4695 return test_flag(FLAG_DATA_DIGEST
);
4697 bool is_omap_digest() const {
4698 return test_flag(FLAG_OMAP_DIGEST
);
4700 bool is_cache_pinned() const {
4701 return test_flag(FLAG_CACHE_PIN
);
4703 bool has_manifest() const {
4704 return test_flag(FLAG_MANIFEST
);
4707 void set_data_digest(__u32 d
) {
4708 set_flag(FLAG_DATA_DIGEST
);
4711 void set_omap_digest(__u32 d
) {
4712 set_flag(FLAG_OMAP_DIGEST
);
4715 void clear_data_digest() {
4716 clear_flag(FLAG_DATA_DIGEST
);
4719 void clear_omap_digest() {
4720 clear_flag(FLAG_OMAP_DIGEST
);
4724 set_data_digest(-1);
4725 set_omap_digest(-1);
4728 void encode(bufferlist
& bl
, uint64_t features
) const;
4729 void decode(bufferlist::iterator
& bl
);
4730 void decode(bufferlist
& bl
) {
4731 bufferlist::iterator p
= bl
.begin();
4734 void dump(Formatter
*f
) const;
4735 static void generate_test_instances(list
<object_info_t
*>& o
);
4737 explicit object_info_t()
4738 : user_version(0), size(0), flags((flag_t
)0),
4739 truncate_seq(0), truncate_size(0),
4740 data_digest(-1), omap_digest(-1),
4741 expected_object_size(0), expected_write_size(0),
4745 explicit object_info_t(const hobject_t
& s
)
4747 user_version(0), size(0), flags((flag_t
)0),
4748 truncate_seq(0), truncate_size(0),
4749 data_digest(-1), omap_digest(-1),
4750 expected_object_size(0), expected_write_size(0),
4754 explicit object_info_t(bufferlist
& bl
) {
4758 WRITE_CLASS_ENCODER_FEATURES(object_info_t
)
4760 ostream
& operator<<(ostream
& out
, const object_info_t
& oi
);
4765 struct ObjectRecoveryInfo
{
4770 SnapSet ss
; // only populated if soid is_snap()
4771 interval_set
<uint64_t> copy_subset
;
4772 map
<hobject_t
, interval_set
<uint64_t>> clone_subset
;
4774 ObjectRecoveryInfo() : size(0) { }
4776 static void generate_test_instances(list
<ObjectRecoveryInfo
*>& o
);
4777 void encode(bufferlist
&bl
, uint64_t features
) const;
4778 void decode(bufferlist::iterator
&bl
, int64_t pool
= -1);
4779 ostream
&print(ostream
&out
) const;
4780 void dump(Formatter
*f
) const;
4782 WRITE_CLASS_ENCODER_FEATURES(ObjectRecoveryInfo
)
4783 ostream
& operator<<(ostream
& out
, const ObjectRecoveryInfo
&inf
);
4785 struct ObjectRecoveryProgress
{
4786 uint64_t data_recovered_to
;
4787 string omap_recovered_to
;
4793 ObjectRecoveryProgress()
4794 : data_recovered_to(0),
4796 data_complete(false), omap_complete(false) { }
4798 bool is_complete(const ObjectRecoveryInfo
& info
) const {
4799 return (data_recovered_to
>= (
4800 info
.copy_subset
.empty() ?
4801 0 : info
.copy_subset
.range_end())) &&
4805 static void generate_test_instances(list
<ObjectRecoveryProgress
*>& o
);
4806 void encode(bufferlist
&bl
) const;
4807 void decode(bufferlist::iterator
&bl
);
4808 ostream
&print(ostream
&out
) const;
4809 void dump(Formatter
*f
) const;
4811 WRITE_CLASS_ENCODER(ObjectRecoveryProgress
)
4812 ostream
& operator<<(ostream
& out
, const ObjectRecoveryProgress
&prog
);
4814 struct PushReplyOp
{
4817 static void generate_test_instances(list
<PushReplyOp
*>& o
);
4818 void encode(bufferlist
&bl
) const;
4819 void decode(bufferlist::iterator
&bl
);
4820 ostream
&print(ostream
&out
) const;
4821 void dump(Formatter
*f
) const;
4823 uint64_t cost(CephContext
*cct
) const;
4825 WRITE_CLASS_ENCODER(PushReplyOp
)
4826 ostream
& operator<<(ostream
& out
, const PushReplyOp
&op
);
4831 ObjectRecoveryInfo recovery_info
;
4832 ObjectRecoveryProgress recovery_progress
;
4834 static void generate_test_instances(list
<PullOp
*>& o
);
4835 void encode(bufferlist
&bl
, uint64_t features
) const;
4836 void decode(bufferlist::iterator
&bl
);
4837 ostream
&print(ostream
&out
) const;
4838 void dump(Formatter
*f
) const;
4840 uint64_t cost(CephContext
*cct
) const;
4842 WRITE_CLASS_ENCODER_FEATURES(PullOp
)
4843 ostream
& operator<<(ostream
& out
, const PullOp
&op
);
4849 interval_set
<uint64_t> data_included
;
4850 bufferlist omap_header
;
4851 map
<string
, bufferlist
> omap_entries
;
4852 map
<string
, bufferlist
> attrset
;
4854 ObjectRecoveryInfo recovery_info
;
4855 ObjectRecoveryProgress before_progress
;
4856 ObjectRecoveryProgress after_progress
;
4858 static void generate_test_instances(list
<PushOp
*>& o
);
4859 void encode(bufferlist
&bl
, uint64_t features
) const;
4860 void decode(bufferlist::iterator
&bl
);
4861 ostream
&print(ostream
&out
) const;
4862 void dump(Formatter
*f
) const;
4864 uint64_t cost(CephContext
*cct
) const;
4866 WRITE_CLASS_ENCODER_FEATURES(PushOp
)
4867 ostream
& operator<<(ostream
& out
, const PushOp
&op
);
4871 * summarize pg contents for purposes of a scrub
4875 map
<string
,bufferptr
> attrs
;
4877 __u32 omap_digest
; ///< omap crc32c
4878 __u32 digest
; ///< data crc32c
4880 bool digest_present
:1;
4881 bool omap_digest_present
:1;
4884 bool ec_hash_mismatch
:1;
4885 bool ec_size_mismatch
:1;
4888 // Init invalid size so it won't match if we get a stat EIO error
4889 size(-1), omap_digest(0), digest(0),
4890 negative(false), digest_present(false), omap_digest_present(false),
4891 read_error(false), stat_error(false), ec_hash_mismatch(false), ec_size_mismatch(false) {}
4893 void encode(bufferlist
& bl
) const;
4894 void decode(bufferlist::iterator
& bl
);
4895 void dump(Formatter
*f
) const;
4896 static void generate_test_instances(list
<object
*>& o
);
4898 WRITE_CLASS_ENCODER(object
)
4900 map
<hobject_t
,object
> objects
;
4901 eversion_t valid_through
;
4902 eversion_t incr_since
;
4904 void merge_incr(const ScrubMap
&l
);
4905 void insert(const ScrubMap
&r
) {
4906 objects
.insert(r
.objects
.begin(), r
.objects
.end());
4908 void swap(ScrubMap
&r
) {
4910 swap(objects
, r
.objects
);
4911 swap(valid_through
, r
.valid_through
);
4912 swap(incr_since
, r
.incr_since
);
4915 void encode(bufferlist
& bl
) const;
4916 void decode(bufferlist::iterator
& bl
, int64_t pool
=-1);
4917 void dump(Formatter
*f
) const;
4918 static void generate_test_instances(list
<ScrubMap
*>& o
);
4920 WRITE_CLASS_ENCODER(ScrubMap::object
)
4921 WRITE_CLASS_ENCODER(ScrubMap
)
4927 bufferlist indata
, outdata
;
4931 memset(&op
, 0, sizeof(ceph_osd_op
));
4935 * split a bufferlist into constituent indata members of a vector of OSDOps
4937 * @param ops [out] vector of OSDOps
4938 * @param in [in] combined data buffer
4940 static void split_osd_op_vector_in_data(vector
<OSDOp
>& ops
, bufferlist
& in
);
4943 * merge indata members of a vector of OSDOp into a single bufferlist
4945 * Notably this also encodes certain other OSDOp data into the data
4946 * buffer, including the sobject_t soid.
4948 * @param ops [in] vector of OSDOps
4949 * @param out [out] combined data buffer
4951 static void merge_osd_op_vector_in_data(vector
<OSDOp
>& ops
, bufferlist
& out
);
4954 * split a bufferlist into constituent outdata members of a vector of OSDOps
4956 * @param ops [out] vector of OSDOps
4957 * @param in [in] combined data buffer
4959 static void split_osd_op_vector_out_data(vector
<OSDOp
>& ops
, bufferlist
& in
);
4962 * merge outdata members of a vector of OSDOps into a single bufferlist
4964 * @param ops [in] vector of OSDOps
4965 * @param out [out] combined data buffer
4967 static void merge_osd_op_vector_out_data(vector
<OSDOp
>& ops
, bufferlist
& out
);
4970 * Clear data as much as possible, leave minimal data for historical op dump
4972 * @param ops [in] vector of OSDOps
4974 static void clear_data(vector
<OSDOp
>& ops
);
4977 ostream
& operator<<(ostream
& out
, const OSDOp
& op
);
4979 struct watch_item_t
{
4982 uint32_t timeout_seconds
;
4985 watch_item_t() : cookie(0), timeout_seconds(0) { }
4986 watch_item_t(entity_name_t name
, uint64_t cookie
, uint32_t timeout
,
4987 const entity_addr_t
& addr
)
4988 : name(name
), cookie(cookie
), timeout_seconds(timeout
),
4991 void encode(bufferlist
&bl
, uint64_t features
) const {
4992 ENCODE_START(2, 1, bl
);
4994 ::encode(cookie
, bl
);
4995 ::encode(timeout_seconds
, bl
);
4996 ::encode(addr
, bl
, features
);
4999 void decode(bufferlist::iterator
&bl
) {
5000 DECODE_START(2, bl
);
5002 ::decode(cookie
, bl
);
5003 ::decode(timeout_seconds
, bl
);
5004 if (struct_v
>= 2) {
5010 WRITE_CLASS_ENCODER_FEATURES(watch_item_t
)
5012 struct obj_watch_item_t
{
5018 * obj list watch response format
5021 struct obj_list_watch_response_t
{
5022 list
<watch_item_t
> entries
;
5024 void encode(bufferlist
& bl
, uint64_t features
) const {
5025 ENCODE_START(1, 1, bl
);
5026 ::encode(entries
, bl
, features
);
5029 void decode(bufferlist::iterator
& bl
) {
5030 DECODE_START(1, bl
);
5031 ::decode(entries
, bl
);
5034 void dump(Formatter
*f
) const {
5035 f
->open_array_section("entries");
5036 for (list
<watch_item_t
>::const_iterator p
= entries
.begin(); p
!= entries
.end(); ++p
) {
5037 f
->open_object_section("watch");
5038 f
->dump_stream("watcher") << p
->name
;
5039 f
->dump_int("cookie", p
->cookie
);
5040 f
->dump_int("timeout", p
->timeout_seconds
);
5041 f
->open_object_section("addr");
5048 static void generate_test_instances(list
<obj_list_watch_response_t
*>& o
) {
5050 o
.push_back(new obj_list_watch_response_t
);
5051 o
.push_back(new obj_list_watch_response_t
);
5052 ea
.set_type(entity_addr_t::TYPE_LEGACY
);
5054 ea
.set_family(AF_INET
);
5055 ea
.set_in4_quad(0, 127);
5056 ea
.set_in4_quad(1, 0);
5057 ea
.set_in4_quad(2, 0);
5058 ea
.set_in4_quad(3, 1);
5060 o
.back()->entries
.push_back(watch_item_t(entity_name_t(entity_name_t::TYPE_CLIENT
, 1), 10, 30, ea
));
5062 ea
.set_in4_quad(3, 2);
5064 o
.back()->entries
.push_back(watch_item_t(entity_name_t(entity_name_t::TYPE_CLIENT
, 2), 20, 60, ea
));
5067 WRITE_CLASS_ENCODER_FEATURES(obj_list_watch_response_t
)
5071 vector
<snapid_t
> snaps
; // ascending
5072 vector
< pair
<uint64_t,uint64_t> > overlap
;
5075 clone_info() : cloneid(CEPH_NOSNAP
), size(0) {}
5077 void encode(bufferlist
& bl
) const {
5078 ENCODE_START(1, 1, bl
);
5079 ::encode(cloneid
, bl
);
5080 ::encode(snaps
, bl
);
5081 ::encode(overlap
, bl
);
5085 void decode(bufferlist::iterator
& bl
) {
5086 DECODE_START(1, bl
);
5087 ::decode(cloneid
, bl
);
5088 ::decode(snaps
, bl
);
5089 ::decode(overlap
, bl
);
5093 void dump(Formatter
*f
) const {
5094 if (cloneid
== CEPH_NOSNAP
)
5095 f
->dump_string("cloneid", "HEAD");
5097 f
->dump_unsigned("cloneid", cloneid
.val
);
5098 f
->open_array_section("snapshots");
5099 for (vector
<snapid_t
>::const_iterator p
= snaps
.begin(); p
!= snaps
.end(); ++p
) {
5100 f
->open_object_section("snap");
5101 f
->dump_unsigned("id", p
->val
);
5105 f
->open_array_section("overlaps");
5106 for (vector
< pair
<uint64_t,uint64_t> >::const_iterator q
= overlap
.begin();
5107 q
!= overlap
.end(); ++q
) {
5108 f
->open_object_section("overlap");
5109 f
->dump_unsigned("offset", q
->first
);
5110 f
->dump_unsigned("length", q
->second
);
5114 f
->dump_unsigned("size", size
);
5116 static void generate_test_instances(list
<clone_info
*>& o
) {
5117 o
.push_back(new clone_info
);
5118 o
.push_back(new clone_info
);
5119 o
.back()->cloneid
= 1;
5120 o
.back()->snaps
.push_back(1);
5121 o
.back()->overlap
.push_back(pair
<uint64_t,uint64_t>(0,4096));
5122 o
.back()->overlap
.push_back(pair
<uint64_t,uint64_t>(8192,4096));
5123 o
.back()->size
= 16384;
5124 o
.push_back(new clone_info
);
5125 o
.back()->cloneid
= CEPH_NOSNAP
;
5126 o
.back()->size
= 32768;
5129 WRITE_CLASS_ENCODER(clone_info
)
5132 * obj list snaps response format
5135 struct obj_list_snap_response_t
{
5136 vector
<clone_info
> clones
; // ascending
5139 void encode(bufferlist
& bl
) const {
5140 ENCODE_START(2, 1, bl
);
5141 ::encode(clones
, bl
);
5145 void decode(bufferlist::iterator
& bl
) {
5146 DECODE_START(2, bl
);
5147 ::decode(clones
, bl
);
5154 void dump(Formatter
*f
) const {
5155 f
->open_array_section("clones");
5156 for (vector
<clone_info
>::const_iterator p
= clones
.begin(); p
!= clones
.end(); ++p
) {
5157 f
->open_object_section("clone");
5161 f
->dump_unsigned("seq", seq
);
5164 static void generate_test_instances(list
<obj_list_snap_response_t
*>& o
) {
5165 o
.push_back(new obj_list_snap_response_t
);
5166 o
.push_back(new obj_list_snap_response_t
);
5169 cl
.snaps
.push_back(1);
5170 cl
.overlap
.push_back(pair
<uint64_t,uint64_t>(0,4096));
5171 cl
.overlap
.push_back(pair
<uint64_t,uint64_t>(8192,4096));
5173 o
.back()->clones
.push_back(cl
);
5174 cl
.cloneid
= CEPH_NOSNAP
;
5178 o
.back()->clones
.push_back(cl
);
5179 o
.back()->seq
= 123;
5183 WRITE_CLASS_ENCODER(obj_list_snap_response_t
)
5187 struct PromoteCounter
{
5188 std::atomic_ullong attempts
{0};
5189 std::atomic_ullong objects
{0};
5190 std::atomic_ullong bytes
{0};
5196 void finish(uint64_t size
) {
5201 void sample_and_attenuate(uint64_t *a
, uint64_t *o
, uint64_t *b
) {
5212 * ObjectStore full statfs information
5214 struct store_statfs_t
5216 uint64_t total
= 0; // Total bytes
5217 uint64_t available
= 0; // Free bytes available
5219 int64_t allocated
= 0; // Bytes allocated by the store
5220 int64_t stored
= 0; // Bytes actually stored by the user
5221 int64_t compressed
= 0; // Bytes stored after compression
5222 int64_t compressed_allocated
= 0; // Bytes allocated for compressed data
5223 int64_t compressed_original
= 0; // Bytes that were successfully compressed
5226 *this = store_statfs_t();
5228 bool operator ==(const store_statfs_t
& other
) const;
5229 void dump(Formatter
*f
) const;
5231 ostream
&operator<<(ostream
&lhs
, const store_statfs_t
&rhs
);