summaryrefslogtreecommitdiff
path: root/src/mongo/s/transaction_router.h
blob: 138600310f67183ddbe8ed0ba3ef3fad16b24915 (plain)
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
/**
 *    Copyright (C) 2018-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 <map>

#include "mongo/base/disallow_copying.h"
#include "mongo/db/commands/txn_cmds_gen.h"
#include "mongo/db/logical_session_id.h"
#include "mongo/db/operation_context.h"
#include "mongo/db/repl/read_concern_args.h"
#include "mongo/s/async_requests_sender.h"
#include "mongo/s/client/shard.h"
#include "mongo/s/shard_id.h"
#include "mongo/util/string_map.h"

namespace mongo {

/**
 * Keeps track of the transaction state. A session is in use when it is being used by a request.
 */
class TransactionRouter {
public:
    // The default value to use as the statement id of the first command in the transaction if none
    // was sent.
    static const StmtId kDefaultFirstStmtId = 0;

    /**
     * Represents the options for a transaction that are shared across all participants. These
     * cannot be changed without restarting the transactions that may have already been begun on
     * every participant, i.e. clearing the current participant list.
     */
    struct SharedTransactionOptions {
        // Set for all distributed transactions.
        TxnNumber txnNumber;
        repl::ReadConcernArgs readConcernArgs;

        // Only set for transactions with snapshot level read concern.
        boost::optional<LogicalTime> atClusterTime;
    };

    /**
     * Represents a shard participant in a distributed transaction. Lives only for the duration of
     * the transaction that created it.
     */
    struct Participant {
        Participant(bool isCoordinator,
                    StmtId stmtIdCreatedAt,
                    SharedTransactionOptions sharedOptions);

        /**
         * Attaches necessary fields if this is participating in a multi statement transaction.
         */
        BSONObj attachTxnFieldsIfNeeded(BSONObj cmd, bool isFirstStatementInThisParticipant) const;

        // True if the participant has been chosen as the coordinator for its transaction
        const bool isCoordinator{false};

        // The highest statement id of the request during which this participant was created.
        const StmtId stmtIdCreatedAt{kUninitializedStmtId};

        // Returns the shared transaction options this participant was created with
        const SharedTransactionOptions sharedOptions;
    };

    enum class TransactionActions { kStart, kContinue, kCommit };

    /**
     * Encapsulates the logic around selecting a global read timestamp for a sharded transaction at
     * snapshot level read concern.
     *
     * The first command in a transaction to target at least one shard must select a cluster time
     * timestamp before targeting, but may change the timestamp before contacting any shards to
     * allow optimizing the timestamp based on the targeted shards. If the first command encounters
     * a retryable error, e.g. StaleShardVersion or SnapshotTooOld, the retry may also select a new
     * timestamp. Once the first command has successfully completed, the timestamp cannot be
     * changed.
     */
    class AtClusterTime {
    public:
        /**
         * Cannot be called until a timestamp has been set.
         */
        LogicalTime getTime() const;

        /**
         * Sets the timestamp and remembers the statement id of the command that set it.
         */
        void setTime(LogicalTime atClusterTime, StmtId currentStmtId);

        /**
         * True if the timestamp can be changed by a command running at the given statement id.
         */
        bool canChange(StmtId currentStmtId) const;

    private:
        StmtId _stmtIdSelectedAt = kUninitializedStmtId;
        LogicalTime _atClusterTime;
    };

    TransactionRouter();
    ~TransactionRouter();

    /**
     * Extract the runtime state attached to the operation context. Returns nullptr if none is
     * attached.
     */
    static TransactionRouter* get(OperationContext* opCtx);

    /**
     * Starts a fresh transaction in this session or continue an existing one. Also cleans up the
     * previous transaction state.
     */
    void beginOrContinueTxn(OperationContext* opCtx,
                            TxnNumber txnNumber,
                            TransactionActions action);

    /**
     * Attaches the required transaction related fields for a request to be sent to the given
     * shard.
     *
     * Calling this method has the following side effects:
     * 1. Potentially selecting a coordinator.
     * 2. Adding the shard to the list of participants.
     * 3. Also append fields for first statements (ex. startTransaction, readConcern)
     *    if the shard was newly added to the list of participants.
     */
    BSONObj attachTxnFieldsIfNeeded(const ShardId& shardId, const BSONObj& cmdObj);

    /**
     * Updates the transaction state to allow for a retry of the current command on a stale version
     * error. This includes sending abortTransaction to all cleared participants. Will throw if the
     * transaction cannot be continued.
     */
    void onStaleShardOrDbError(OperationContext* opCtx,
                               StringData cmdName,
                               const Status& errorStatus);

    /**
     * Resets the transaction state to allow for a retry attempt. This includes clearing all
     * participants, clearing the coordinator, resetting the global read timestamp, and sending
     * abortTransaction to all cleared participants. Will throw if the transaction cannot be
     * continued.
     */
    void onSnapshotError(OperationContext* opCtx, const Status& errorStatus);

    /**
     * Updates the transaction tracking state to allow for a retry attempt on a view resolution
     * error. This includes sending abortTransaction to all cleared participants.
     */
    void onViewResolutionError(OperationContext* opCtx, const NamespaceString& nss);

    /**
     * Sets the atClusterTime for the current transaction to the latest time in the router's logical
     * clock. Does nothing if the transaction does not have snapshot read concern or an
     * atClusterTime has already been selected and cannot be changed.
     */
    void setDefaultAtClusterTime(OperationContext* opCtx);

    /**
     * Returns the global read timestamp for this transaction. Returns boost::none for transactions
     * that don't run at snapshot level read concern or if a timestamp has not yet been selected.
     */
    const boost::optional<AtClusterTime>& getAtClusterTime() const;

    /**
     * If a coordinator has been selected for the current transaction, returns its identifier
     */
    const boost::optional<ShardId>& getCoordinatorId() const;

    /**
     * Commits the transaction. For transactions with multiple participants, this will initiate
     * the two phase commit procedure.
     */
    Shard::CommandResponse commitTransaction(
        OperationContext* opCtx, const boost::optional<TxnRecoveryToken>& recoveryToken);

    /**
     * Sends abort to all participants and returns the responses from all shards.
     */
    std::vector<AsyncRequestsSender::Response> abortTransaction(OperationContext* opCtx,
                                                                bool isImplicit = false);

    /**
     * Sends abort to all shards in the current participant list. Will retry on retryable errors,
     * but ignores the responses from each shard.
     */
    void implicitlyAbortTransaction(OperationContext* opCtx, const Status& errorStatus);

    /**
     * Returns the participant for this transaction or nullptr if the specified shard is not
     * participant of this transaction.
     */
    Participant* getParticipant(const ShardId& shard);

    /**
     * If a coordinator has been selected for this transaction already, constructs a recovery token,
     * which can be used to resume commit or abort of the transaction from a different router.
     */
    void appendRecoveryToken(BSONObjBuilder* builder) const;

private:
    // Shortcut to obtain the id of the session under which this transaction router runs
    const LogicalSessionId& _sessionId() const;

    /**
     * Run basic commit for transactions that touched a single shard.
     */
    Shard::CommandResponse _commitSingleShardTransaction(OperationContext* opCtx);

    Shard::CommandResponse _commitWithRecoveryToken(OperationContext* opCtx,
                                                    const TxnRecoveryToken& recoveryToken);

    /**
     * Run two phase commit for transactions that touched multiple shards.
     */
    Shard::CommandResponse _commitMultiShardTransaction(OperationContext* opCtx);

    /**
     * Sets the given logical time as the atClusterTime for the transaction to be the greater of the
     * given time and the user's afterClusterTime, if one was provided.
     */
    void _setAtClusterTime(const boost::optional<LogicalTime>& afterClusterTime,
                           LogicalTime candidateTime);

    /**
     * Returns true if the current transaction can retry on a stale version error from a contacted
     * shard. This is always true except for an error received by a write that is not the first
     * overall statement in the sharded transaction. This is because the entire command will be
     * retried, and shards that were not stale and are targeted again may incorrectly execute the
     * command a second time.
     *
     * Note: Even if this method returns true, the retry attempt may still fail, e.g. if one of the
     * shards that returned a stale version error was involved in a previously completed a statement
     * for this transaction.
     *
     * TODO SERVER-37207: Change batch writes to retry only the failed writes in a batch, to allow
     * retrying writes beyond the first overall statement.
     */
    bool _canContinueOnStaleShardOrDbError(StringData cmdName) const;

    /**
     * Returns true if the current transaction can retry on a snapshot error. This is only true on
     * the first command recevied for a transaction.
     */
    bool _canContinueOnSnapshotError() const;

    /**
     * Throws NoSuchTransaction if the response from abortTransaction failed with a code other than
     * NoSuchTransaction. Does not check for write concern errors.
     */
    void _assertAbortStatusIsOkOrNoSuchTransaction(
        const AsyncRequestsSender::Response& response) const;

    /**
     * Returns all participants created during the current statement.
     */
    std::vector<ShardId> _getPendingParticipants() const;

    /**
     * Removes all participants created during the current statement from the participant list and
     * sends abortTransaction to each. Waits for all responses before returning.
     */
    void _clearPendingParticipants(OperationContext* opCtx);

    /**
     * Creates a new participant for the shard.
     */
    Participant& _createParticipant(const ShardId& shard);

    /**
     * If the transaction's read concern level is snapshot, asserts the participant's atClusterTime
     * matches the transaction's.
     */
    void _verifyParticipantAtClusterTime(const Participant& participant);

    /**
     * Returns a string with the active transaction's transaction number and logical session id
     * (i.e. the transaction id).
     */
    std::string _txnIdToString() const;

    // The currently active transaction number on this transaction router (i.e. on the session)
    TxnNumber _txnNumber{kUninitializedTxnNumber};

    // Whether the router has initiated a two-phase commit by handing off commit coordination to the
    // coordinator. If so, the router should no longer implicitly abort the transaction on errors,
    // since the coordinator may independently make a commit decision.
    bool _initiatedTwoPhaseCommit{false};

    // Indicates whether this is trying to recover a commitTransaction on the current transaction.
    bool _isRecoveringCommit{false};

    // Map of current participants of the current transaction.
    StringMap<Participant> _participants;

    // The id of coordinator participant, used to construct prepare requests.
    boost::optional<ShardId> _coordinatorId;

    // The read concern the current transaction was started with.
    repl::ReadConcernArgs _readConcernArgs;

    // The cluster time of the timestamp all participant shards in the current transaction with
    // snapshot level read concern must read from. Only set for transactions running with snapshot
    // level read concern.
    boost::optional<AtClusterTime> _atClusterTime;

    // The statement id of the latest received command for this transaction. For batch writes, this
    // will be the highest stmtId contained in the batch. Incremented by one if new commands do not
    // contain statement ids.
    StmtId _latestStmtId = kUninitializedStmtId;

    // The statement id of the command that began this transaction. Defaults to zero if no statement
    // id was included in the first command.
    StmtId _firstStmtId = kUninitializedStmtId;
};

}  // namespace mongo