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
|
/**
* Copyright (C) 2020-present MongoDB, Inc.
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the Server Side Public License, version 1,
* as published by MongoDB, Inc.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* Server Side Public License for more details.
*
* You should have received a copy of the Server Side Public License
* along with this program. If not, see
* <http://www.mongodb.com/licensing/server-side-public-license>.
*
* As a special exception, the copyright holders give permission to link the
* code of portions of this program with the OpenSSL library under certain
* conditions as described in each individual source file and distribute
* linked combinations including the program with the OpenSSL library. You
* must comply with the Server Side Public License in all respects for
* all of the code used other than as permitted herein. If you modify file(s)
* with this exception, you may extend this exception to your version of the
* file(s), but you are not obligated to do so. If you do not wish to do so,
* delete this exception statement from your version. If you delete this
* exception statement from all source files in the program, then also delete
* it in the license file.
*/
#pragma once
#include <boost/optional.hpp>
#include "mongo/bson/timestamp.h"
#include "mongo/db/operation_context.h"
#include "mongo/db/repl/optime.h"
#include "mongo/executor/task_executor.h"
namespace mongo {
/**
* The TenantMigrationAccessBlocker is used to block and eventually reject reads and writes to a
* database while the Atlas Serverless tenant that owns the database is being migrated from this
* replica set to another replica set.
*
* In order to preserve causal consistency across the migration, this replica set, the "donor",
* blocks writes and reads as of a particular "blockTimestamp". The donor then advances the
* recipient's clusterTime to "blockTimestamp" before committing the migration.
*
* Client writes are run inside a new loop, similar to writeConflictRetry:
*
* template <typename F>
* auto migrationConflictRetry(OperationContext* opCtx, const Database* db, F&& f) {
* while (true) {
* try {
* return f();
* } catch (const MigrationConflictException&) {
* TenantMigrationAccessBlocker::get(db).checkIfCanWriteOrBlock(opCtx);
* }
* }
* }
*
* Writes call checkIfCanWriteOrThrow after being assigned an OpTime but before committing. The
* method throws MigratingTenantConflict if writes are being blocked, which is caught in the loop.
* The write then blocks until the migration either commits (in which case checkIfCanWriteOrBlock
* throws an error that causes the write to be rejected) or aborts (in which case
* checkIfCanWriteOrBlock returns successfully and the write is retried in the loop). This loop is
* used because writes must not block after being assigned an OpTime but before committing.
*
* Reads with afterClusterTime or atClusterTime call checkIfCanReadOrBlock at some point after
* waiting for readConcern, that is, after waiting to reach their clusterTime, which includes
* waiting for all earlier oplog holes to be filled.
*
* Given this, the donor uses this class's API in the following way:
*
* 1. The donor primary creates a WriteUnitOfWork to do a write, call it the "start blocking" write.
* The donor primary calls startBlockingWrites before the write is assigned an OpTime. This write's
* Timestamp will be the "blockTimestamp".
*
* At this point:
* - Writes that have already passed checkIfCanWriteOrThrow must have been assigned an OpTime before
* the blockTimestamp, since the blockTimestamp hasn't been assigned yet, and OpTimes are handed
* out in monotonically increasing order.
* - Writes that have not yet passed checkIfCanWriteOrThrow will end up blocking. Some of these
* writes may have already been assigned an OpTime, or may end up being assigned an OpTime that is
* before the blockTimestamp, and so will end up blocking unnecessarily, but not incorrectly.
*
* 2. In the op observer after the "start blocking" write's OpTime is set, primaries and secondaries
* of the donor replica set call startBlockingReadsAfter with the write's Timestamp as
* "blockTimestamp".
*
* At this point:
* - Reads on the node that have already passed checkIfCanReadOrBlock must have a clusterTime before
* the blockTimestamp, since the write at blockTimestamp hasn't committed yet (i.e., there's still
* an oplog hole at blockTimestamp).
* - Reads on the node that have not yet passed checkIfCanReadOrBlock will end up blocking.
*
* If the "start blocking" write aborts or the write rolls back via replication rollback, the node
* calls rollBackStartBlocking.
*
* 4a. The donor primary commits the migration by doing another write, call it the "commit" write.
* The op observer for the "commit" write on primaries and secondaries calls commit, which
* asynchronously waits for the "commit" write's OpTime to become majority committed, then
* transitions the class to reject writes and reads.
*
* 4b. The donor primary can instead abort the migration by doing a write, call it the "abort"
* write. The op observer for the "abort" write on primaries and secondaries calls abort, which
* asynchronously waits for the "abort" write's OpTime to become majority committed, then
* transitions the class back to allowing reads and writes.
*
* If the "commit" or "abort" write aborts or rolls back via replication rollback, the node calls
* rollBackCommitOrAbort, which cancels the asynchronous task.
*/
class TenantMigrationAccessBlocker {
public:
TenantMigrationAccessBlocker(ServiceContext* serviceContext, executor::TaskExecutor* executor);
//
// Called by all writes and reads against the database.
//
void checkIfCanWriteOrThrow();
void checkIfCanWriteOrBlock(OperationContext* opCtx);
void checkIfLinearizableReadWasAllowedOrThrow(OperationContext* opCtx);
void checkIfCanDoClusterTimeReadOrBlock(OperationContext* opCtx,
const Timestamp& readTimestamp);
//
// Called while donating this database.
//
void startBlockingWrites();
void startBlockingReadsAfter(const Timestamp& timestamp);
void rollBackStartBlocking();
void commit(repl::OpTime opTime);
void abort(repl::OpTime opTime);
void rollBackCommitOrAbort();
void appendInfoForServerStatus(BSONObjBuilder* builder) const;
private:
void _waitForOpTimeToMajorityCommit(repl::OpTime opTime, std::function<void()> callbackFn);
enum class Access { kAllow, kBlockWrites, kBlockWritesAndReads, kReject };
ServiceContext* _serviceContext;
executor::TaskExecutor* _executor;
// Protects the state below.
mutable Mutex _mutex = MONGO_MAKE_LATCH("TenantMigrationAccessBlocker::_mutex");
Access _access{Access::kAllow};
boost::optional<Timestamp> _blockTimestamp;
boost::optional<repl::OpTime> _commitOrAbortOpTime;
OperationContext* _waitForCommitOrAbortToMajorityCommitOpCtx{nullptr};
stdx::condition_variable _transitionOutOfBlockingCV;
};
} // namespace mongo
|
