summaryrefslogtreecommitdiff
path: root/src/mongo/db/commands/pipeline_command.cpp
blob: 571c30872545be4af5149ea00fb93afb49437962 (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
347
348
/**
 * Copyright (c) 2011-2014 MongoDB Inc.
 *
 * This program is free software: you can redistribute it and/or  modify
 * it under the terms of the GNU Affero General Public License, version 3,
 * as published by the Free Software Foundation.
 *
 * 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
 * GNU Affero General Public License for more details.
 *
 * You should have received a copy of the GNU Affero General Public License
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 *
 * 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 GNU Affero General 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.
 */

#include "mongo/platform/basic.h"

#include <boost/scoped_ptr.hpp>
#include <boost/shared_ptr.hpp>
#include <vector>

#include "mongo/db/auth/action_set.h"
#include "mongo/db/auth/action_type.h"
#include "mongo/db/auth/privilege.h"
#include "mongo/db/catalog/database.h"
#include "mongo/db/client.h"
#include "mongo/db/commands.h"
#include "mongo/db/commands/cursor_responses.h"
#include "mongo/db/curop.h"
#include "mongo/db/db_raii.h"
#include "mongo/db/exec/pipeline_proxy.h"
#include "mongo/db/service_context.h"
#include "mongo/db/pipeline/accumulator.h"
#include "mongo/db/pipeline/document.h"
#include "mongo/db/pipeline/document_source.h"
#include "mongo/db/pipeline/expression.h"
#include "mongo/db/pipeline/expression_context.h"
#include "mongo/db/pipeline/pipeline.h"
#include "mongo/db/pipeline/pipeline_d.h"
#include "mongo/db/query/find_constants.h"
#include "mongo/db/query/get_executor.h"
#include "mongo/db/storage_options.h"

namespace mongo {

    using boost::intrusive_ptr;
    using boost::scoped_ptr;
    using boost::shared_ptr;
    using std::auto_ptr;
    using std::string;
    using std::stringstream;
    using std::endl;

    /**
     * Returns true if we need to keep a ClientCursor saved for this pipeline (for future getMore
     * requests).  Otherwise, returns false.
     */
    static bool handleCursorCommand(OperationContext* txn,
                                    const string& ns,
                                    ClientCursorPin* pin,
                                    PlanExecutor* exec,
                                    const BSONObj& cmdObj,
                                    BSONObjBuilder& result) {

        ClientCursor* cursor = pin ? pin->c() : NULL;
        if (pin) {
            invariant(cursor);
            invariant(cursor->getExecutor() == exec);
            invariant(cursor->isAggCursor());
        }

        const long long defaultBatchSize = 101; // Same as query.
        long long batchSize;
        uassertStatusOK(Command::parseCommandCursorOptions(cmdObj, defaultBatchSize, &batchSize));

        // can't use result BSONObjBuilder directly since it won't handle exceptions correctly.
        BSONArrayBuilder resultsArray;
        const int byteLimit = MaxBytesToReturnToClientAtOnce;
        BSONObj next;
        for (int objCount = 0; objCount < batchSize; objCount++) {
            // The initial getNext() on a PipelineProxyStage may be very expensive so we don't
            // do it when batchSize is 0 since that indicates a desire for a fast return.
            if (exec->getNext(&next, NULL) != PlanExecutor::ADVANCED) {
                // make it an obvious error to use cursor or executor after this point
                cursor = NULL;
                exec = NULL;
                break;
            }

            if (resultsArray.len() + next.objsize() > byteLimit) {
                // Get the pipeline proxy stage wrapped by this PlanExecutor.
                PipelineProxyStage* proxy = static_cast<PipelineProxyStage*>(exec->getRootStage());
                // too big. next will be the first doc in the second batch
                proxy->pushBack(next);
                break;
            }

            resultsArray.append(next);
        }

        // NOTE: exec->isEOF() can have side effects such as writing by $out. However, it should
        // be relatively quick since if there was no pin then the input is empty. Also, this
        // violates the contract for batchSize==0. Sharding requires a cursor to be returned in that
        // case. This is ok for now however, since you can't have a sharded collection that doesn't
        // exist.
        const bool canReturnMoreBatches = pin;
        if (!canReturnMoreBatches && exec && !exec->isEOF()) {
            // msgasserting since this shouldn't be possible to trigger from today's aggregation
            // language. The wording assumes that the only reason pin would be null is if the
            // collection doesn't exist.
            msgasserted(17391, str::stream()
                << "Aggregation has more results than fit in initial batch, but can't "
                << "create cursor since collection " << ns << " doesn't exist");
        }

        if (cursor) {
            // If a time limit was set on the pipeline, remaining time is "rolled over" to the
            // cursor (for use by future getmore ops).
            cursor->setLeftoverMaxTimeMicros( txn->getCurOp()->getRemainingMaxTimeMicros() );

            if (txn->getClient()->isInDirectClient()) {
                cursor->setUnownedRecoveryUnit(txn->recoveryUnit());
            }
            else {
                // We stash away the RecoveryUnit in the ClientCursor.  It's used for subsequent
                // getMore requests.  The calling OpCtx gets a fresh RecoveryUnit.
                txn->recoveryUnit()->abandonSnapshot();
                cursor->setOwnedRecoveryUnit(txn->releaseRecoveryUnit());
                StorageEngine* storageEngine = getGlobalServiceContext()->getGlobalStorageEngine();
                invariant(txn->setRecoveryUnit(storageEngine->newRecoveryUnit(),
                                               OperationContext::kNotInUnitOfWork)
                          == OperationContext::kNotInUnitOfWork);
            }

            // Cursor needs to be in a saved state while we yield locks for getmore. State
            // will be restored in getMore().
            exec->saveState();
        }

        const long long cursorId = cursor ? cursor->cursorid() : 0LL;
        appendCursorResponseObject(cursorId, ns, resultsArray.arr(), &result);

        return static_cast<bool>(cursor);
    }


    class PipelineCommand :
        public Command {
    public:
        PipelineCommand() :Command(Pipeline::commandName) {} // command is called "aggregate"

        // Locks are managed manually, in particular by DocumentSourceCursor.
        virtual bool isWriteCommandForConfigServer() const { return false; }
        virtual bool slaveOk() const { return false; }
        virtual bool slaveOverrideOk() const { return true; }
        virtual void help(stringstream &help) const {
            help << "{ pipeline: [ { $operator: {...}}, ... ]"
                 << ", explain: <bool>"
                 << ", allowDiskUse: <bool>"
                 << ", cursor: {batchSize: <number>}"
                 << " }"
                 << endl
                 << "See http://dochub.mongodb.org/core/aggregation for more details."
                 ;
        }

        virtual void addRequiredPrivileges(const std::string& dbname,
                                           const BSONObj& cmdObj,
                                           std::vector<Privilege>* out) {
            Pipeline::addRequiredPrivileges(this, dbname, cmdObj, out);
        }

        virtual bool run(OperationContext* txn,
                         const string &db,
                         BSONObj &cmdObj,
                         int options,
                         string &errmsg,
                         BSONObjBuilder &result) {
            NamespaceString nss(parseNs(db, cmdObj));
            if (nss.coll().empty()) {
                errmsg = "missing collection name";
                return false;
            }

            intrusive_ptr<ExpressionContext> pCtx = new ExpressionContext(txn, nss);
            pCtx->tempDir = storageGlobalParams.dbpath + "/_tmp";

            /* try to parse the command; if this fails, then we didn't run */
            intrusive_ptr<Pipeline> pPipeline = Pipeline::parseCommand(errmsg, cmdObj, pCtx);
            if (!pPipeline.get())
                return false;

            // This is outside of the if block to keep the object alive until the pipeline is finished.
            BSONObj parsed;
            if (kDebugBuild && !pPipeline->isExplain() && !pCtx->inShard) {
                // Make sure all operations round-trip through Pipeline::toBson() correctly by
                // reparsing every command in debug builds. This is important because sharded
                // aggregations rely on this ability.  Skipping when inShard because this has
                // already been through the transformation (and this unsets pCtx->inShard).
                parsed = pPipeline->serialize().toBson();
                pPipeline = Pipeline::parseCommand(errmsg, parsed, pCtx);
                verify(pPipeline);
            }

            PlanExecutor* exec = NULL;
            scoped_ptr<ClientCursorPin> pin; // either this OR the execHolder will be non-null
            auto_ptr<PlanExecutor> execHolder;
            {
                // This will throw if the sharding version for this connection is out of date. The
                // lock must be held continuously from now until we have we created both the output
                // ClientCursor and the input executor. This ensures that both are using the same
                // sharding version that we synchronize on here. This is also why we always need to
                // create a ClientCursor even when we aren't outputting to a cursor. See the comment
                // on ShardFilterStage for more details.
                AutoGetCollectionForRead ctx(txn, nss.ns());

                Collection* collection = ctx.getCollection();

                // This does mongod-specific stuff like creating the input PlanExecutor and adding
                // it to the front of the pipeline if needed.
                boost::shared_ptr<PlanExecutor> input = PipelineD::prepareCursorSource(txn,
                                                                                       collection,
                                                                                       pPipeline,
                                                                                       pCtx);
                pPipeline->stitch();

                // Create the PlanExecutor which returns results from the pipeline. The WorkingSet
                // ('ws') and the PipelineProxyStage ('proxy') will be owned by the created
                // PlanExecutor.
                auto_ptr<WorkingSet> ws(new WorkingSet());
                auto_ptr<PipelineProxyStage> proxy(
                    new PipelineProxyStage(pPipeline, input, ws.get()));
                Status execStatus = Status::OK();
                if (NULL == collection) {
                    execStatus = PlanExecutor::make(txn,
                                                    ws.release(),
                                                    proxy.release(),
                                                    nss.ns(),
                                                    PlanExecutor::YIELD_MANUAL,
                                                    &exec);
                }
                else {
                    execStatus = PlanExecutor::make(txn,
                                                    ws.release(),
                                                    proxy.release(),
                                                    collection,
                                                    PlanExecutor::YIELD_MANUAL,
                                                    &exec);
                }
                invariant(execStatus.isOK());
                execHolder.reset(exec);

                if (!collection && input) {
                    // If we don't have a collection, we won't be able to register any executors, so
                    // make sure that the input PlanExecutor (likely wrapping an EOFStage) doesn't
                    // need to be registered.
                    invariant(!input->collection());
                }

                if (collection) {
                    const bool isAggCursor = true; // enable special locking behavior
                    ClientCursor* cursor = new ClientCursor(collection->getCursorManager(),
                                                            execHolder.release(),
                                                            nss.ns(),
                                                            0,
                                                            BSONObj(),
                                                            isAggCursor);
                    pin.reset(new ClientCursorPin(collection->getCursorManager(),
                                                  cursor->cursorid()));
                    // Don't add any code between here and the start of the try block.
                }

                // At this point, it is safe to release the collection lock.
                // - In the case where we have a collection: we will need to reacquire the
                //   collection lock later when cleaning up our ClientCursorPin.
                // - In the case where we don't have a collection: our PlanExecutor won't be
                //   registered, so it will be safe to clean it up outside the lock.
                invariant(NULL == execHolder.get() || NULL == execHolder->collection());
            }

            try {
                // Unless set to true, the ClientCursor created above will be deleted on block exit.
                bool keepCursor = false;

                const bool isCursorCommand = !cmdObj["cursor"].eoo();

                // If both explain and cursor are specified, explain wins.
                if (pPipeline->isExplain()) {
                    result << "stages" << Value(pPipeline->writeExplainOps());
                }
                else if (isCursorCommand) {
                    keepCursor = handleCursorCommand(txn,
                                                     nss.ns(),
                                                     pin.get(),
                                                     exec,
                                                     cmdObj,
                                                     result);
                }
                else {
                    pPipeline->run(result);
                }

                // Clean up our ClientCursorPin, if needed.  We must reacquire the collection lock
                // in order to do so.
                if (pin) {
                    // We acquire locks here with DBLock and CollectionLock instead of using
                    // AutoGetCollectionForRead.  AutoGetCollectionForRead will throw if the
                    // sharding version is out of date, and we don't care if the sharding version
                    // has changed.
                    Lock::DBLock dbLock(txn->lockState(), nss.db(), MODE_IS);
                    Lock::CollectionLock collLock(txn->lockState(), nss.ns(), MODE_IS);
                    if (keepCursor) {
                        pin->release();
                    }
                    else {
                        pin->deleteUnderlying();
                    }
                }
            }
            catch (...) {
                // On our way out of scope, we clean up our ClientCursorPin if needed.
                if (pin) {
                    Lock::DBLock dbLock(txn->lockState(), nss.db(), MODE_IS);
                    Lock::CollectionLock collLock(txn->lockState(), nss.ns(), MODE_IS);
                    pin->deleteUnderlying();
                }
                throw;
            }
            // Any code that needs the cursor pinned must be inside the try block, above.

            return true;
        }
    } cmdPipeline;

} // namespace mongo