-
Notifications
You must be signed in to change notification settings - Fork 1.2k
/
repair.cc
1976 lines (1851 loc) · 97.3 KB
/
repair.cc
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
* Copyright (C) 2015-present ScyllaDB
*/
/*
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
#include "repair.hh"
#include "repair/row_level.hh"
#include "mutation/atomic_cell_hash.hh"
#include "dht/sharder.hh"
#include "streaming/stream_reason.hh"
#include "gms/inet_address.hh"
#include "utils/fb_utilities.hh"
#include "gms/gossiper.hh"
#include "message/messaging_service.hh"
#include "sstables/sstables.hh"
#include "replica/database.hh"
#include "db/config.hh"
#include "utils/hashers.hh"
#include "locator/network_topology_strategy.hh"
#include "service/migration_manager.hh"
#include "partition_range_compat.hh"
#include "gms/feature_service.hh"
#include <boost/algorithm/string/predicate.hpp>
#include <boost/algorithm/string/split.hpp>
#include <boost/algorithm/string/classification.hpp>
#include <boost/algorithm/cxx11/any_of.hpp>
#include <boost/range/algorithm.hpp>
#include <boost/range/algorithm_ext.hpp>
#include <boost/range/adaptor/map.hpp>
#include <seastar/core/gate.hh>
#include <seastar/util/defer.hh>
#include <seastar/core/metrics_registration.hh>
#include <seastar/core/coroutine.hh>
#include <seastar/coroutine/parallel_for_each.hh>
#include <seastar/core/sleep.hh>
#include <cfloat>
#include <algorithm>
#include "idl/position_in_partition.dist.hh"
#include "idl/partition_checksum.dist.hh"
logging::logger rlogger("repair");
node_ops_info::node_ops_info(node_ops_id ops_uuid_, shared_ptr<abort_source> as_, std::list<gms::inet_address>&& ignore_nodes_) noexcept
: ops_uuid(ops_uuid_)
, as(std::move(as_))
, ignore_nodes(std::move(ignore_nodes_))
{}
void node_ops_info::check_abort() {
if (as && as->abort_requested()) {
auto msg = format("Node operation with ops_uuid={} is aborted", ops_uuid);
rlogger.warn("{}", msg);
throw std::runtime_error(msg);
}
}
node_ops_metrics::node_ops_metrics(shared_ptr<repair::task_manager_module> module)
: _module(module)
{
namespace sm = seastar::metrics;
auto ops_label_type = sm::label("ops");
_metrics.add_group("node_ops", {
sm::make_gauge("finished_percentage", [this] { return bootstrap_finished_percentage(); },
sm::description("Finished percentage of node operation on this shard"), {ops_label_type("bootstrap")}),
sm::make_gauge("finished_percentage", [this] { return replace_finished_percentage(); },
sm::description("Finished percentage of node operation on this shard"), {ops_label_type("replace")}),
sm::make_gauge("finished_percentage", [this] { return rebuild_finished_percentage(); },
sm::description("Finished percentage of node operation on this shard"), {ops_label_type("rebuild")}),
sm::make_gauge("finished_percentage", [this] { return decommission_finished_percentage(); },
sm::description("Finished percentage of node operation on this shard"), {ops_label_type("decommission")}),
sm::make_gauge("finished_percentage", [this] { return removenode_finished_percentage(); },
sm::description("Finished percentage of node operation on this shard"), {ops_label_type("removenode")}),
sm::make_gauge("finished_percentage", [this] { return repair_finished_percentage(); },
sm::description("Finished percentage of node operation on this shard"), {ops_label_type("repair")}),
});
}
float node_ops_metrics::bootstrap_finished_percentage() {
return bootstrap_total_ranges == 0 ? 1 : float(bootstrap_finished_ranges) / float(bootstrap_total_ranges);
}
float node_ops_metrics::replace_finished_percentage() {
return replace_total_ranges == 0 ? 1 : float(replace_finished_ranges) / float(replace_total_ranges);
}
float node_ops_metrics::rebuild_finished_percentage() {
return rebuild_total_ranges == 0 ? 1 : float(rebuild_finished_ranges) / float(rebuild_total_ranges);
}
float node_ops_metrics::decommission_finished_percentage() {
return decommission_total_ranges == 0 ? 1 : float(decommission_finished_ranges) / float(decommission_total_ranges);
}
float node_ops_metrics::removenode_finished_percentage() {
return removenode_total_ranges == 0 ? 1 : float(removenode_finished_ranges) / float(removenode_total_ranges);
}
template <typename T1, typename T2>
inline
static std::ostream& operator<<(std::ostream& os, const std::unordered_map<T1, T2>& v) {
bool first = true;
os << "{";
for (auto&& elem : v) {
if (!first) {
os << ", ";
} else {
first = false;
}
os << elem.first << "=" << elem.second;
}
os << "}";
return os;
}
std::ostream& operator<<(std::ostream& out, row_level_diff_detect_algorithm algo) {
switch (algo) {
case row_level_diff_detect_algorithm::send_full_set:
return out << "send_full_set";
case row_level_diff_detect_algorithm::send_full_set_rpc_stream:
return out << "send_full_set_rpc_stream";
};
return out << "unknown";
}
static size_t get_nr_tables(const replica::database& db, const sstring& keyspace) {
auto& m = db.get_column_families_mapping();
return std::count_if(m.begin(), m.end(), [&keyspace] (auto& e) {
return e.first.first == keyspace;
});
}
static std::vector<sstring> list_column_families(const replica::database& db, const sstring& keyspace) {
std::vector<sstring> ret;
for (auto &&e : db.get_column_families_mapping()) {
if (e.first.first == keyspace) {
ret.push_back(e.first.second);
}
}
return ret;
}
static const replica::column_family* find_column_family_if_exists(const replica::database& db, std::string_view ks_name, std::string_view cf_name, bool warn = true) {
try {
auto uuid = db.find_uuid(std::move(ks_name), std::move(cf_name));
return &db.find_column_family(uuid);
} catch (replica::no_such_column_family&) {
if (warn) {
rlogger.warn("{}", std::current_exception());
}
return nullptr;
}
}
static const replica::column_family* find_column_family_if_exists(const replica::database& db, const table_id& uuid, bool warn = true) {
try {
return &db.find_column_family(uuid);
} catch (...) {
if (warn) {
rlogger.warn("{}", std::current_exception());
}
return nullptr;
}
}
std::ostream& operator<<(std::ostream& os, const repair_uniq_id& x) {
return os << format("[id={}, uuid={}]", x.id, x.uuid());
}
// Must run inside a seastar thread
static std::vector<table_id> get_table_ids(const replica::database& db, const sstring& keyspace, const std::vector<sstring>& tables) {
std::vector<table_id> table_ids;
table_ids.reserve(tables.size());
for (auto& table : tables) {
thread::maybe_yield();
try {
table_ids.push_back(db.find_uuid(keyspace, table));
} catch (replica::no_such_column_family&) {
rlogger.warn("Column family {} does not exist in keyspace {}", table, keyspace);
}
}
return table_ids;
}
static std::vector<sstring> get_table_names(const replica::database& db, const std::vector<table_id>& table_ids) {
std::vector<sstring> table_names;
table_names.reserve(table_ids.size());
for (auto& table_id : table_ids) {
auto* cf = find_column_family_if_exists(db, table_id);
table_names.push_back(cf ? cf->schema()->cf_name() : "");
}
return table_names;
}
template<typename Collection, typename T>
void remove_item(Collection& c, T& item) {
auto it = std::find(c.begin(), c.end(), item);
if (it != c.end()) {
c.erase(it);
}
}
// Return all of the neighbors with whom we share the provided range.
static std::vector<gms::inet_address> get_neighbors(
const locator::effective_replication_map& erm,
const sstring& ksname, query::range<dht::token> range,
const std::vector<sstring>& data_centers,
const std::vector<sstring>& hosts,
const std::unordered_set<gms::inet_address>& ignore_nodes) {
dht::token tok = range.end() ? range.end()->value() : dht::maximum_token();
auto ret = erm.get_natural_endpoints(tok);
remove_item(ret, utils::fb_utilities::get_broadcast_address());
if (!data_centers.empty()) {
auto dc_endpoints_map = erm.get_token_metadata().get_topology().get_datacenter_endpoints();
std::unordered_set<gms::inet_address> dc_endpoints;
for (const sstring& dc : data_centers) {
auto it = dc_endpoints_map.find(dc);
if (it == dc_endpoints_map.end()) {
std::vector<sstring> dcs;
for (const auto& e : dc_endpoints_map) {
dcs.push_back(e.first);
}
throw std::runtime_error(fmt::format("Unknown data center '{}'. "
"Known data centers: {}", dc, dcs));
}
for (const auto& endpoint : it->second) {
dc_endpoints.insert(endpoint);
}
}
// We require, like Cassandra does, that the current host must also
// be part of the repair
if (!dc_endpoints.contains(utils::fb_utilities::get_broadcast_address())) {
throw std::runtime_error("The current host must be part of the repair");
}
// The resulting list of nodes is the intersection of the nodes in the
// listed data centers, and the (range-dependent) list of neighbors.
std::unordered_set<gms::inet_address> neighbor_set(ret.begin(), ret.end());
ret.clear();
for (const auto& endpoint : dc_endpoints) {
if (neighbor_set.contains(endpoint)) {
ret.push_back(endpoint);
}
}
} else if (!hosts.empty()) {
bool found_me = false;
std::unordered_set<gms::inet_address> neighbor_set(ret.begin(), ret.end());
ret.clear();
for (const sstring& host : hosts) {
gms::inet_address endpoint;
try {
endpoint = gms::inet_address(host);
} catch(...) {
throw std::runtime_error(format("Unknown host specified: {}", host));
}
if (endpoint == utils::fb_utilities::get_broadcast_address()) {
found_me = true;
} else if (neighbor_set.contains(endpoint)) {
ret.push_back(endpoint);
// If same host is listed twice, don't add it again later
neighbor_set.erase(endpoint);
}
// Nodes which aren't neighbors for this range are ignored.
// This allows the user to give a list of "good" nodes, where
// for each different range, only the subset of nodes actually
// holding a replica of the given range is used. This,
// however, means the user is never warned if one of the nodes
// on the list isn't even part of the cluster.
}
// We require, like Cassandra does, that the current host must also
// be listed on the "-hosts" option - even those we don't want it in
// the returned list:
if (!found_me) {
throw std::runtime_error("The current host must be part of the repair");
}
if (ret.size() < 1) {
auto me = utils::fb_utilities::get_broadcast_address();
auto others = erm.get_natural_endpoints(tok);
remove_item(others, me);
throw std::runtime_error(fmt::format("Repair requires at least two "
"endpoints that are neighbors before it can continue, "
"the endpoint used for this repair is {}, other "
"available neighbors are {} but these neighbors were not "
"part of the supplied list of hosts to use during the "
"repair ({}).", me, others, hosts));
}
} else if (!ignore_nodes.empty()) {
auto it = std::remove_if(ret.begin(), ret.end(), [&ignore_nodes] (const gms::inet_address& node) {
return ignore_nodes.contains(node);
});
ret.erase(it, ret.end());
}
return boost::copy_range<std::vector<gms::inet_address>>(std::move(ret));
#if 0
// Origin's ActiveRepairService.getNeighbors() also verifies that the
// requested range fits into a local range
StorageService ss = StorageService.instance;
Map<Range<Token>, List<InetAddress>> replicaSets = ss.getRangeToAddressMap(keyspaceName);
Range<Token> rangeSuperSet = null;
for (Range<Token> range : ss.getLocalRanges(keyspaceName))
{
if (range.contains(toRepair))
{
rangeSuperSet = range;
break;
}
else if (range.intersects(toRepair))
{
throw new IllegalArgumentException("Requested range intersects a local range but is not fully contained in one; this would lead to imprecise repair");
}
}
if (rangeSuperSet == null || !replicaSets.containsKey(rangeSuperSet))
return Collections.emptySet();
#endif
}
static future<std::list<gms::inet_address>> get_hosts_participating_in_repair(
const locator::effective_replication_map& erm,
const sstring& ksname,
const dht::token_range_vector& ranges,
const std::vector<sstring>& data_centers,
const std::vector<sstring>& hosts,
const std::unordered_set<gms::inet_address>& ignore_nodes) {
std::unordered_set<gms::inet_address> participating_hosts;
// Repair coordinator must participate in repair, but it is never
// returned by get_neighbors - add it here
participating_hosts.insert(utils::fb_utilities::get_broadcast_address());
co_await do_for_each(ranges, [&] (const dht::token_range& range) {
const auto nbs = get_neighbors(erm, ksname, range, data_centers, hosts, ignore_nodes);
for (const auto& nb : nbs) {
participating_hosts.insert(nb);
}
});
co_return std::list<gms::inet_address>(participating_hosts.begin(), participating_hosts.end());
}
float node_ops_metrics::repair_finished_percentage() {
return _module->report_progress(streaming::stream_reason::repair);
}
repair::task_manager_module::task_manager_module(tasks::task_manager& tm, repair_service& rs, size_t max_repair_memory) noexcept
: tasks::task_manager::module(tm, "repair")
, _rs(rs)
, _range_parallelism_semaphore(std::max(size_t(1), size_t(max_repair_memory / max_repair_memory_per_range / 4)),
named_semaphore_exception_factory{"repair range parallelism"})
{
auto nr = _range_parallelism_semaphore.available_units();
rlogger.info("Setting max_repair_memory={}, max_repair_memory_per_range={}, max_repair_ranges_in_parallel={}",
max_repair_memory, max_repair_memory_per_range, nr);
}
void repair::task_manager_module::start(repair_uniq_id id) {
_pending_repairs.insert(id.uuid());
_status[id.id] = repair_status::RUNNING;
}
void repair::task_manager_module::done(repair_uniq_id id, bool succeeded) {
_pending_repairs.erase(id.uuid());
_aborted_pending_repairs.erase(id.uuid());
if (succeeded) {
_status.erase(id.id);
} else {
_status[id.id] = repair_status::FAILED;
}
_done_cond.broadcast();
}
repair_status repair::task_manager_module::get(int id) const {
if (std::cmp_greater(id, _sequence_number)) {
throw std::runtime_error(format("unknown repair id {}", id));
}
auto it = _status.find(id);
if (it == _status.end()) {
return repair_status::SUCCESSFUL;
} else {
return it->second;
}
}
future<repair_status> repair::task_manager_module::repair_await_completion(int id, std::chrono::steady_clock::time_point timeout) {
return seastar::with_gate(async_gate(), [this, id, timeout] {
if (std::cmp_greater(id, _sequence_number)) {
return make_exception_future<repair_status>(std::runtime_error(format("unknown repair id {}", id)));
}
return repeat_until_value([this, id, timeout] {
auto it = _status.find(id);
if (it == _status.end()) {
return make_ready_future<std::optional<repair_status>>(repair_status::SUCCESSFUL);
} else {
if (it->second == repair_status::FAILED) {
return make_ready_future<std::optional<repair_status>>(repair_status::FAILED);
} else {
return _done_cond.wait(timeout).then([] {
return make_ready_future<std::optional<repair_status>>(std::nullopt);
}).handle_exception_type([] (condition_variable_timed_out&) {
return make_ready_future<std::optional<repair_status>>(repair_status::RUNNING);
});
}
}
});
});
}
void repair::task_manager_module::check_in_shutdown() {
abort_source().check();
}
void repair::task_manager_module::add_shard_task_id(int id, tasks::task_id uuid) {
_repairs.emplace(id, uuid);
}
void repair::task_manager_module::remove_shard_task_id(int id) {
_repairs.erase(id);
}
tasks::task_manager::task_ptr repair::task_manager_module::get_shard_task_ptr(int id) {
auto it = _repairs.find(id);
if (it != _repairs.end()) {
auto task_it = _tasks.find(it->second);
if (task_it != _tasks.end()) {
return task_it->second;
}
}
return {};
}
std::vector<int> repair::task_manager_module::get_active() const {
std::vector<int> res;
boost::push_back(res, _status | boost::adaptors::filtered([] (auto& x) {
return x.second == repair_status::RUNNING;
}) | boost::adaptors::map_keys);
return res;
}
size_t repair::task_manager_module::nr_running_repair_jobs() {
size_t count = 0;
if (this_shard_id() != 0) {
return count;
}
for (auto& x : _status) {
auto& status = x.second;
if (status == repair_status::RUNNING) {
count++;
}
}
return count;
}
bool repair::task_manager_module::is_aborted(const tasks::task_id& uuid) {
return _aborted_pending_repairs.contains(uuid);
}
void repair::task_manager_module::abort_all_repairs() {
_aborted_pending_repairs = _pending_repairs;
for (auto& x : _repairs) {
auto it = _tasks.find(x.second);
if (it != _tasks.end()) {
auto& impl = dynamic_cast<repair::shard_repair_task_impl&>(*it->second->_impl);
// If the task is aborted, its state will change to failed. One can wait for this with task_manager::task::done().
(void)impl.abort();
}
}
rlogger.info0("Started to abort repair jobs={}, nr_jobs={}", _aborted_pending_repairs, _aborted_pending_repairs.size());
}
float repair::task_manager_module::report_progress(streaming::stream_reason reason) {
uint64_t nr_ranges_finished = 0;
uint64_t nr_ranges_total = 0;
for (auto& x : _repairs) {
auto it = _tasks.find(x.second);
if (it != _tasks.end()) {
auto& impl = dynamic_cast<repair::shard_repair_task_impl&>(*it->second->_impl);
if (impl.reason() == reason) {
nr_ranges_total += impl.ranges_size();
nr_ranges_finished += impl.nr_ranges_finished;
}
}
}
return nr_ranges_total == 0 ? 1 : float(nr_ranges_finished) / float(nr_ranges_total);
}
named_semaphore& repair::task_manager_module::range_parallelism_semaphore() {
return _range_parallelism_semaphore;
}
future<> repair::task_manager_module::run(repair_uniq_id id, std::function<void ()> func) {
return seastar::with_gate(async_gate(), [this, id, func = std::move(func)] () mutable {
start(id);
return seastar::async([func = std::move(func)] { func(); }).then([this, id] {
rlogger.info("repair[{}]: completed successfully", id.uuid());
done(id, true);
}).handle_exception([this, id] (std::exception_ptr ep) {
done(id, false);
return make_exception_future(std::move(ep));
});
});
}
future<uint64_t> estimate_partitions(seastar::sharded<replica::database>& db, const sstring& keyspace,
const sstring& cf, const dht::token_range& range) {
return db.map_reduce0(
[keyspace, cf, range] (auto& db) {
// FIXME: column_family should have a method to estimate the number of
// partitions (and of course it should use cardinality estimation bitmaps,
// not trivial sum). We shouldn't have this ugly code here...
// FIXME: If sstables are shared, they will be accounted more than
// once. However, shared sstables should exist for a short-time only.
auto sstables = db.find_column_family(keyspace, cf).get_sstables();
return boost::accumulate(*sstables, uint64_t(0),
[&range] (uint64_t x, auto&& sst) { return x + sst->estimated_keys_for_range(range); });
},
uint64_t(0),
std::plus<uint64_t>()
);
}
static
const dht::sharder&
get_sharder_for_tables(seastar::sharded<replica::database>& db, const sstring& keyspace, const std::vector<table_id>& table_ids) {
schema_ptr last_s;
for (size_t idx = 0 ; idx < table_ids.size(); idx++) {
schema_ptr s;
if (const auto* cf = find_column_family_if_exists(db.local(), table_ids[idx])) {
s = cf->schema();
} else {
continue;
}
if (last_s && last_s->get_sharder() != s->get_sharder()) {
throw std::runtime_error(
format("All tables repaired together have to have the same sharding logic. "
"Different sharding logic found: {} (for table {}) and {} (for table {})",
last_s->get_sharder(), last_s->cf_name(),
s->get_sharder(), s->cf_name()));
}
last_s = std::move(s);
}
if (!last_s) {
throw std::runtime_error(format("Failed to find sharder for keyspace={}, tables={}, no table in this keyspace",
keyspace, table_ids));
}
return last_s->get_sharder();
}
repair::shard_repair_task_impl::shard_repair_task_impl(tasks::task_manager::module_ptr module,
tasks::task_id id,
const sstring& keyspace,
repair_service& repair,
locator::effective_replication_map_ptr erm_,
const dht::token_range_vector& ranges_,
std::vector<table_id> table_ids_,
repair_uniq_id parent_id_,
const std::vector<sstring>& data_centers_,
const std::vector<sstring>& hosts_,
const std::unordered_set<gms::inet_address>& ignore_nodes_,
streaming::stream_reason reason_,
bool hints_batchlog_flushed)
: repair_task_impl(module, id, 0, keyspace, "", "", parent_id_.uuid(), reason_)
, rs(repair)
, db(repair.get_db())
, messaging(repair.get_messaging().container())
, sys_dist_ks(repair.get_sys_dist_ks())
, view_update_generator(repair.get_view_update_generator())
, mm(repair.get_migration_manager())
, gossiper(repair.get_gossiper())
, sharder(get_sharder_for_tables(db, keyspace, table_ids_))
, erm(std::move(erm_))
, ranges(ranges_)
, cfs(get_table_names(db.local(), table_ids_))
, table_ids(std::move(table_ids_))
, global_repair_id(parent_id_)
, data_centers(data_centers_)
, hosts(hosts_)
, ignore_nodes(ignore_nodes_)
, total_rf(erm->get_replication_factor())
, nr_ranges_total(ranges.size())
, _hints_batchlog_flushed(std::move(hints_batchlog_flushed))
{ }
void repair::shard_repair_task_impl::check_failed_ranges() {
rlogger.info("repair[{}]: stats: repair_reason={}, keyspace={}, tables={}, ranges_nr={}, {}",
global_repair_id.uuid(), _reason, _status.keyspace, table_names(), ranges.size(), _stats.get_stats());
if (nr_failed_ranges || _aborted || _failed_because) {
sstring failed_because = "N/A";
if (!_aborted) {
failed_because = _failed_because ? *_failed_because : "unknown";
}
auto msg = format("repair[{}]: {} out of {} ranges failed, keyspace={}, tables={}, repair_reason={}, nodes_down_during_repair={}, aborted_by_user={}, failed_because={}",
global_repair_id.uuid(), nr_failed_ranges, ranges_size(), _status.keyspace, table_names(), _reason, nodes_down, _aborted, failed_because);
rlogger.warn("{}", msg);
throw std::runtime_error(msg);
} else {
if (dropped_tables.size()) {
rlogger.warn("repair[{}]: completed successfully, keyspace={}, ignoring dropped tables={}", global_repair_id.uuid(), _status.keyspace, dropped_tables);
} else {
rlogger.info("repair[{}]: completed successfully, keyspace={}", global_repair_id.uuid(), _status.keyspace);
}
}
}
void repair::shard_repair_task_impl::check_in_abort_or_shutdown() {
try {
_as.check();
} catch (...) {
if (!_aborted) {
_aborted = true;
rlogger.warn("repair[{}]: Repair job aborted by user, job={}, keyspace={}, tables={}",
global_repair_id.uuid(), global_repair_id.uuid(), _status.keyspace, table_names());;
}
throw;
}
}
repair_neighbors repair::shard_repair_task_impl::get_repair_neighbors(const dht::token_range& range) {
return neighbors.empty() ?
repair_neighbors(get_neighbors(*erm, _status.keyspace, range, data_centers, hosts, ignore_nodes)) :
neighbors[range];
}
size_t repair::shard_repair_task_impl::ranges_size() {
return ranges.size() * table_ids.size();
}
// Repair a single local range, multiple column families.
// Comparable to RepairSession in Origin
future<> repair::shard_repair_task_impl::repair_range(const dht::token_range& range, ::table_id table_id) {
check_in_abort_or_shutdown();
ranges_index++;
repair_neighbors r_neighbors = get_repair_neighbors(range);
auto neighbors = std::move(r_neighbors.all);
auto mandatory_neighbors = std::move(r_neighbors.mandatory);
auto live_neighbors = boost::copy_range<std::vector<gms::inet_address>>(neighbors |
boost::adaptors::filtered([this] (const gms::inet_address& node) { return gossiper.is_alive(node); }));
for (auto& node : mandatory_neighbors) {
auto it = std::find(live_neighbors.begin(), live_neighbors.end(), node);
if (it == live_neighbors.end()) {
nr_failed_ranges++;
nodes_down.insert(node);
auto status = format("failed: mandatory neighbor={} is not alive", node);
rlogger.error("repair[{}]: Repair {} out of {} ranges, keyspace={}, table={}, range={}, peers={}, live_peers={}, status={}",
global_repair_id.uuid(), ranges_index, ranges_size(), _status.keyspace, table_names(), range, neighbors, live_neighbors, status);
// If the task is aborted, its state will change to failed. One can wait for this with task_manager::task::done().
(void)abort();
co_await coroutine::return_exception(std::runtime_error(format("Repair mandatory neighbor={} is not alive, keyspace={}, mandatory_neighbors={}",
node, _status.keyspace, mandatory_neighbors)));
}
}
if (live_neighbors.size() != neighbors.size()) {
nr_failed_ranges++;
std::unordered_set<gms::inet_address> live_neighbors_set(live_neighbors.begin(), live_neighbors.end());
for (auto& node : neighbors) {
if (!live_neighbors_set.contains(node)) {
nodes_down.insert(node);
}
}
auto status = live_neighbors.empty() ? "skipped_no_live_peers" : "partial";
rlogger.warn("repair[{}]: Repair {} out of {} ranges, keyspace={}, table={}, range={}, peers={}, live_peers={}, status={}",
global_repair_id.uuid(), ranges_index, ranges_size(), _status.keyspace, table_names(), range, neighbors, live_neighbors, status);
if (live_neighbors.empty()) {
co_return;
}
neighbors.swap(live_neighbors);
}
if (neighbors.empty()) {
auto status = "skipped_no_followers";
rlogger.warn("repair[{}]: Repair {} out of {} ranges, keyspace={}, table={}, range={}, peers={}, live_peers={}, status={}",
global_repair_id.uuid(), ranges_index, ranges_size(), _status.keyspace, table_names(), range, neighbors, live_neighbors, status);
co_return;
}
rlogger.debug("repair[{}]: Repair {} out of {} ranges, keyspace={}, table={}, range={}, peers={}, live_peers={}",
global_repair_id.uuid(), ranges_index, ranges_size(), _status.keyspace, table_names(), range, neighbors, live_neighbors);
co_await mm.sync_schema(db.local(), neighbors);
sstring cf;
try {
cf = db.local().find_column_family(table_id).schema()->cf_name();
} catch (replica::no_such_column_family&) {
co_return;
}
// Row level repair
if (dropped_tables.contains(cf)) {
co_return;
}
try {
co_await repair_cf_range_row_level(*this, cf, table_id, range, neighbors);
} catch (replica::no_such_column_family&) {
dropped_tables.insert(cf);
} catch (...) {
nr_failed_ranges++;
throw;
}
}
void repair_stats::add(const repair_stats& o) {
round_nr += o.round_nr;
round_nr_fast_path_already_synced += o.round_nr_fast_path_already_synced;
round_nr_fast_path_same_combined_hashes += o.round_nr_fast_path_same_combined_hashes;
round_nr_slow_path += o.round_nr_slow_path;
rpc_call_nr += o.rpc_call_nr;
tx_hashes_nr += o.tx_hashes_nr;
rx_hashes_nr += o.rx_hashes_nr;
tx_row_nr += o.tx_row_nr;
rx_row_nr += o.rx_row_nr;
tx_row_bytes += o.tx_row_bytes;
rx_row_bytes += o.rx_row_bytes;
auto add_map = [] (auto& target, auto& src) {
for (const auto& [k, v] : src) {
target[k] += v;
}
};
add_map(row_from_disk_bytes, o.row_from_disk_bytes);
add_map(row_from_disk_nr, o.row_from_disk_nr);
add_map(tx_row_nr_peer, o.tx_row_nr_peer);
add_map(rx_row_nr_peer, o.rx_row_nr_peer);
}
sstring repair_stats::get_stats() {
std::map<gms::inet_address, float> row_from_disk_bytes_per_sec;
std::map<gms::inet_address, float> row_from_disk_rows_per_sec;
auto duration = std::chrono::duration_cast<std::chrono::duration<float>>(lowres_clock::now() - start_time).count();
for (auto& x : row_from_disk_bytes) {
if (std::fabs(duration) > FLT_EPSILON) {
row_from_disk_bytes_per_sec[x.first] = x.second / duration / 1024 / 1024;
} else {
row_from_disk_bytes_per_sec[x.first] = 0;
}
}
for (auto& x : row_from_disk_nr) {
if (std::fabs(duration) > FLT_EPSILON) {
row_from_disk_rows_per_sec[x.first] = x.second / duration;
} else {
row_from_disk_rows_per_sec[x.first] = 0;
}
}
return format("round_nr={}, round_nr_fast_path_already_synced={}, round_nr_fast_path_same_combined_hashes={}, round_nr_slow_path={}, rpc_call_nr={}, tx_hashes_nr={}, rx_hashes_nr={}, duration={} seconds, tx_row_nr={}, rx_row_nr={}, tx_row_bytes={}, rx_row_bytes={}, row_from_disk_bytes={}, row_from_disk_nr={}, row_from_disk_bytes_per_sec={} MiB/s, row_from_disk_rows_per_sec={} Rows/s, tx_row_nr_peer={}, rx_row_nr_peer={}",
round_nr,
round_nr_fast_path_already_synced,
round_nr_fast_path_same_combined_hashes,
round_nr_slow_path,
rpc_call_nr,
tx_hashes_nr,
rx_hashes_nr,
duration,
tx_row_nr,
rx_row_nr,
tx_row_bytes,
rx_row_bytes,
row_from_disk_bytes,
row_from_disk_nr,
row_from_disk_bytes_per_sec,
row_from_disk_rows_per_sec,
tx_row_nr_peer,
rx_row_nr_peer);
}
struct repair_options {
// If primary_range is true, we should perform repair only on this node's
// primary ranges. The default of false means perform repair on all ranges
// held by the node. primary_range=true is useful if the user plans to
// repair all nodes.
bool primary_range = false;
// If ranges is not empty, it overrides the repair's default heuristics
// for determining the list of ranges to repair. In particular, "ranges"
// overrides the setting of "primary_range".
dht::token_range_vector ranges;
// If start_token and end_token are set, they define a range which is
// intersected with the ranges actually held by this node to decide what
// to repair.
sstring start_token;
sstring end_token;
// column_families is the list of column families to repair in the given
// keyspace. If this list is empty (the default), all the column families
// in this keyspace are repaired
std::vector<sstring> column_families;
// hosts specifies the list of known good hosts to repair with this host
// (note that this host is required to also be on this list). For each
// range repaired, only the relevant subset of the hosts (holding a
// replica of this range) is used.
std::vector<sstring> hosts;
// The ignore_nodes specifies the list of nodes to ignore in this repair,
// e.g., the user knows a node is down and wants to run repair without this
// specific node.
std::vector<sstring> ignore_nodes;
// data_centers is used to restrict the repair to the local data center.
// The node starting the repair must be in the data center; Issuing a
// repair to a data center other than the named one returns an error.
std::vector<sstring> data_centers;
repair_options(std::unordered_map<sstring, sstring> options) {
bool_opt(primary_range, options, PRIMARY_RANGE_KEY);
ranges_opt(ranges, options, RANGES_KEY);
list_opt(column_families, options, COLUMNFAMILIES_KEY);
list_opt(hosts, options, HOSTS_KEY);
list_opt(ignore_nodes, options, IGNORE_NODES_KEY);
list_opt(data_centers, options, DATACENTERS_KEY);
// We currently do not support incremental repair. We could probably
// ignore this option as it is just an optimization, but for now,
// let's make it an error.
bool incremental = false;
bool_opt(incremental, options, INCREMENTAL_KEY);
if (incremental) {
throw std::runtime_error("unsupported incremental repair");
}
// We do not currently support the distinction between "parallel" and
// "sequential" repair, and operate the same for both.
// We don't currently support "dc parallel" parallelism.
int parallelism = PARALLEL;
int_opt(parallelism, options, PARALLELISM_KEY);
if (parallelism != PARALLEL && parallelism != SEQUENTIAL) {
throw std::runtime_error(format("unsupported repair parallelism: {}", parallelism));
}
string_opt(start_token, options, START_TOKEN);
string_opt(end_token, options, END_TOKEN);
bool trace = false;
bool_opt(trace, options, TRACE_KEY);
if (trace) {
throw std::runtime_error("unsupported trace");
}
// Consume, ignore.
int job_threads;
int_opt(job_threads, options, JOB_THREADS_KEY);
// The parsing code above removed from the map options we have parsed.
// If anything is left there in the end, it's an unsupported option.
if (!options.empty()) {
throw std::runtime_error(format("unsupported repair options: {}",
options));
}
}
static constexpr const char* PRIMARY_RANGE_KEY = "primaryRange";
static constexpr const char* PARALLELISM_KEY = "parallelism";
static constexpr const char* INCREMENTAL_KEY = "incremental";
static constexpr const char* JOB_THREADS_KEY = "jobThreads";
static constexpr const char* RANGES_KEY = "ranges";
static constexpr const char* COLUMNFAMILIES_KEY = "columnFamilies";
static constexpr const char* DATACENTERS_KEY = "dataCenters";
static constexpr const char* HOSTS_KEY = "hosts";
static constexpr const char* IGNORE_NODES_KEY = "ignore_nodes";
static constexpr const char* TRACE_KEY = "trace";
static constexpr const char* START_TOKEN = "startToken";
static constexpr const char* END_TOKEN = "endToken";
// Settings of "parallelism" option. Numbers must match Cassandra's
// RepairParallelism enum, which is used by the caller.
enum repair_parallelism {
SEQUENTIAL=0, PARALLEL=1, DATACENTER_AWARE=2
};
private:
static void bool_opt(bool& var,
std::unordered_map<sstring, sstring>& options,
const sstring& key) {
auto it = options.find(key);
if (it != options.end()) {
// Same parsing as Boolean.parseBoolean does:
if (boost::algorithm::iequals(it->second, "true")) {
var = true;
} else {
var = false;
}
options.erase(it);
}
}
static void int_opt(int& var,
std::unordered_map<sstring, sstring>& options,
const sstring& key) {
auto it = options.find(key);
if (it != options.end()) {
errno = 0;
var = strtol(it->second.c_str(), nullptr, 10);
if (errno) {
throw(std::runtime_error(format("cannot parse integer: '{}'", it->second)));
}
options.erase(it);
}
}
static void string_opt(sstring& var,
std::unordered_map<sstring, sstring>& options,
const sstring& key) {
auto it = options.find(key);
if (it != options.end()) {
var = it->second;
options.erase(it);
}
}
// A range is expressed as start_token:end token and multiple ranges can
// be given as comma separated ranges(e.g. aaa:bbb,ccc:ddd).
static void ranges_opt(dht::token_range_vector& var,
std::unordered_map<sstring, sstring>& options,
const sstring& key) {
auto it = options.find(key);
if (it == options.end()) {
return;
}
std::vector<sstring> range_strings;
boost::split(range_strings, it->second, boost::algorithm::is_any_of(","));
for (auto range : range_strings) {
std::vector<sstring> token_strings;
boost::split(token_strings, range, boost::algorithm::is_any_of(":"));
if (token_strings.size() != 2) {
throw(std::runtime_error("range must have two components "
"separated by ':', got '" + range + "'"));
}
auto tok_start = dht::token::from_sstring(token_strings[0]);
auto tok_end = dht::token::from_sstring(token_strings[1]);
auto rng = wrapping_range<dht::token>(
::range<dht::token>::bound(tok_start, false),
::range<dht::token>::bound(tok_end, true));
::compat::unwrap_into(std::move(rng), dht::token_comparator(), [&] (dht::token_range&& x) {
var.push_back(std::move(x));
});
}
options.erase(it);
}
// A comma-separate list of strings
static void list_opt(std::vector<sstring>& var,
std::unordered_map<sstring, sstring>& options,
const sstring& key) {
auto it = options.find(key);
if (it == options.end()) {
return;
}
std::vector<sstring> range_strings;
boost::split(var, it->second, boost::algorithm::is_any_of(","));
options.erase(it);
}
};
future<> repair::shard_repair_task_impl::do_repair_ranges() {
// Repair tables in the keyspace one after another
assert(table_names().size() == table_ids.size());
for (size_t idx = 0; idx < table_ids.size(); idx++) {
auto table_id = table_ids[idx];
auto table_name = table_names()[idx];
// repair all the ranges in limited parallelism
rlogger.info("repair[{}]: Started to repair {} out of {} tables in keyspace={}, table={}, table_id={}, repair_reason={}",
global_repair_id.uuid(), idx + 1, table_ids.size(), _status.keyspace, table_name, table_id, _reason);
co_await coroutine::parallel_for_each(ranges, [this, table_id] (auto&& range) {
return with_semaphore(rs.get_repair_module().range_parallelism_semaphore(), 1, [this, &range, table_id] {
return repair_range(range, table_id).then([this] {
if (_reason == streaming::stream_reason::bootstrap) {
rs.get_metrics().bootstrap_finished_ranges++;
} else if (_reason == streaming::stream_reason::replace) {
rs.get_metrics().replace_finished_ranges++;
} else if (_reason == streaming::stream_reason::rebuild) {
rs.get_metrics().rebuild_finished_ranges++;
} else if (_reason == streaming::stream_reason::decommission) {
rs.get_metrics().decommission_finished_ranges++;
} else if (_reason == streaming::stream_reason::removenode) {
rs.get_metrics().removenode_finished_ranges++;
} else if (_reason == streaming::stream_reason::repair) {
rs.get_metrics().repair_finished_ranges_sum++;
nr_ranges_finished++;
}
rlogger.debug("repair[{}]: node ops progress bootstrap={}, replace={}, rebuild={}, decommission={}, removenode={}, repair={}",
global_repair_id.uuid(),
rs.get_metrics().bootstrap_finished_percentage(),
rs.get_metrics().replace_finished_percentage(),
rs.get_metrics().rebuild_finished_percentage(),
rs.get_metrics().decommission_finished_percentage(),
rs.get_metrics().removenode_finished_percentage(),
rs.get_metrics().repair_finished_percentage());
});
});
});
if (_reason != streaming::stream_reason::repair) {
try {
auto& table = db.local().find_column_family(table_id);
rlogger.debug("repair[{}]: Trigger off-strategy compaction for keyspace={}, table={}",
global_repair_id.uuid(), table.schema()->ks_name(), table.schema()->cf_name());
table.trigger_offstrategy_compaction();
} catch (replica::no_such_column_family&) {
// Ignore dropped table
}
}
}
co_return;
}
// Repairs a list of token ranges, each assumed to be a token
// range for which this node holds a replica, and, importantly, each range
// is assumed to be a indivisible in the sense that all the tokens in has the
// same nodes as replicas.
future<> repair::shard_repair_task_impl::run() {
rs.get_repair_module().add_shard_task_id(global_repair_id.id, _status.id);