path: root/src/mongo/db/exec/collection_scan.cpp

/**
 *    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.
 */

#include "mongo/util/assert_util.h"

#include "mongo/db/exec/collection_scan.h"

#include <memory>

#include "mongo/db/catalog/collection.h"
#include "mongo/db/catalog/database.h"
#include "mongo/db/exec/collection_scan_common.h"
#include "mongo/db/exec/filter.h"
#include "mongo/db/exec/scoped_timer.h"
#include "mongo/db/exec/working_set.h"
#include "mongo/db/exec/working_set_common.h"
#include "mongo/db/query/plan_executor_impl.h"
#include "mongo/db/record_id_helpers.h"
#include "mongo/db/repl/optime.h"
#include "mongo/logv2/log.h"
#include "mongo/util/assert_util.h"
#include "mongo/util/fail_point.h"

#define MONGO_LOGV2_DEFAULT_COMPONENT ::mongo::logv2::LogComponent::kQuery


namespace mongo {

using std::unique_ptr;
using std::vector;

namespace {
const char* getStageName(const CollectionPtr& coll, const CollectionScanParams& params) {
    return (!coll->ns().isOplog() && (params.minRecord || params.maxRecord)) ? "CLUSTERED_IXSCAN"
                                                                             : "COLLSCAN";
}
}  // namespace


CollectionScan::CollectionScan(ExpressionContext* expCtx,
                               const CollectionPtr& collection,
                               const CollectionScanParams& params,
                               WorkingSet* workingSet,
                               const MatchExpression* filter)
    : RequiresCollectionStage(getStageName(collection, params), expCtx, collection),
      _workingSet(workingSet),
      _filter((filter && !filter->isTriviallyTrue()) ? filter : nullptr),
      _params(params) {
    // Explain reports the direction of the collection scan.
    _specificStats.direction = params.direction;
    _specificStats.minRecord = params.minRecord;
    _specificStats.maxRecord = params.maxRecord;
    _specificStats.tailable = params.tailable;
    if (params.minRecord || params.maxRecord) {
        // The 'minRecord' and 'maxRecord' parameters are used for a special optimization that
        // applies only to forwards scans of the oplog and scans on clustered collections.
        invariant(!params.resumeAfterRecordId);
        if (collection->ns().isOplogOrChangeCollection()) {
            invariant(params.direction == CollectionScanParams::FORWARD);
        } else {
            invariant(collection->isClustered());
        }
    }

    if (params.boundInclusion !=
        CollectionScanParams::ScanBoundInclusion::kIncludeBothStartAndEndRecords) {
        // A collection must be clustered if the bounds aren't both included by default.
        tassert(6125000,
                "Only collection scans on clustered collections may specify recordId "
                "BoundInclusion policies",
                collection->isClustered());

        if (filter) {
            // The filter is applied after the ScanBoundInclusion is considered.
            LOGV2_DEBUG(6125007,
                        5,
                        "Running a bounded collection scan with a ScanInclusionBound may cause "
                        "the filter to be overriden");
        }
    }

    LOGV2_DEBUG(5400802,
                5,
                "collection scan bounds",
                "min"_attr = (!_params.minRecord) ? "none" : _params.minRecord->toString(),
                "max"_attr = (!_params.maxRecord) ? "none" : _params.maxRecord->toString());
    tassert(6521000,
            "Expected an oplog or a change collection with 'shouldTrackLatestOplogTimestamp'",
            !_params.shouldTrackLatestOplogTimestamp ||
                collection->ns().isOplogOrChangeCollection());

    if (params.assertTsHasNotFallenOff) {
        tassert(6521001,
                "Expected 'shouldTrackLatestOplogTimestamp' with 'assertTsHasNotFallenOff'",
                params.shouldTrackLatestOplogTimestamp);
        tassert(6521002,
                "Expected forward collection scan with 'assertTsHasNotFallenOff'",
                params.direction == CollectionScanParams::FORWARD);
    }

    if (params.resumeAfterRecordId) {
        // The 'resumeAfterRecordId' parameter is used for resumable collection scans, which we
        // only support in the forward direction.
        tassert(6521003,
                "Expected forward collection scan with 'resumeAfterRecordId'",
                params.direction == CollectionScanParams::FORWARD);
    }
}

PlanStage::StageState CollectionScan::doWork(WorkingSetID* out) {
    if (_commonStats.isEOF) {
        return PlanStage::IS_EOF;
    }

    boost::optional<Record> record;
    const bool needToMakeCursor = !_cursor;

    const auto ret = handlePlanStageYield(
        expCtx(),
        "CollectionScan",
        collection()->ns().ns(),
        [&] {
            if (needToMakeCursor) {
                const bool forward = _params.direction == CollectionScanParams::FORWARD;

                if (forward && _params.shouldWaitForOplogVisibility) {
                    // Forward, non-tailable scans from the oplog need to wait until all oplog
                    // entries before the read begins to be visible. This isn't needed for reverse
                    // scans because we only hide oplog entries from forward scans, and it isn't
                    // necessary for tailing cursors because they ignore EOF and will eventually see
                    // all writes. Forward, non-tailable scans are the only case where a meaningful
                    // EOF will be seen that might not include writes that finished before the read
                    // started. This also must be done before we create the cursor as that is when
                    // we establish the endpoint for the cursor. Also call abandonSnapshot to make
                    // sure that we are using a fresh storage engine snapshot while waiting.
                    // Otherwise, we will end up reading from the snapshot where the oplog entries
                    // are not yet visible even after the wait.
                    invariant(!_params.tailable && collection()->ns().isOplog());

                    opCtx()->recoveryUnit()->abandonSnapshot();
                    collection()->getRecordStore()->waitForAllEarlierOplogWritesToBeVisible(
                        opCtx());
                }

                _cursor = collection()->getCursor(opCtx(), forward);

                if (!_lastSeenId.isNull()) {
                    invariant(_params.tailable);
                    // Seek to where we were last time. If it no longer exists, mark us as dead
                    // since we want to signal an error rather than silently dropping data from the
                    // stream.
                    //
                    // Note that we want to return the record *after* this one since we have already
                    // returned this one. This is possible in the tailing case. Notably, tailing is
                    // the only time we'd need to create a cursor after already getting a record out
                    // of it and updating our _lastSeenId.
                    if (!_cursor->seekExact(_lastSeenId)) {
                        uasserted(ErrorCodes::CappedPositionLost,
                                  str::stream() << "CollectionScan died due to failure to restore "
                                                << "tailable cursor position. "
                                                << "Last seen record id: " << _lastSeenId);
                    }
                }

                if (_params.resumeAfterRecordId && !_params.resumeAfterRecordId->isNull()) {
                    invariant(!_params.tailable);
                    invariant(_lastSeenId.isNull());
                    // Seek to where we are trying to resume the scan from. Signal a KeyNotFound
                    // error if the record no longer exists.
                    //
                    // Note that we want to return the record *after* this one since we have already
                    // returned this one prior to the resume.
                    auto& recordIdToSeek = *_params.resumeAfterRecordId;
                    if (!_cursor->seekExact(recordIdToSeek)) {
                        uasserted(ErrorCodes::KeyNotFound,
                                  str::stream()
                                      << "Failed to resume collection scan: the recordId from "
                                         "which we are "
                                      << "attempting to resume no longer exists in the collection. "
                                      << "recordId: " << recordIdToSeek);
                    }
                }
            }

            if (_lastSeenId.isNull() && _params.direction == CollectionScanParams::FORWARD &&
                _params.minRecord) {
                // Seek to the approximate start location.
                record = _cursor->seekNear(_params.minRecord->recordId());
            }

            if (_lastSeenId.isNull() && _params.direction == CollectionScanParams::BACKWARD &&
                _params.maxRecord) {
                // Seek to the approximate start location (at the end).
                record = _cursor->seekNear(_params.maxRecord->recordId());
            }

            if (!record) {
                record = _cursor->next();
            }

            return PlanStage::ADVANCED;
        },
        [&] {
            // yieldHandler
            // Leave us in a state to try again next time.
            if (needToMakeCursor)
                _cursor.reset();
            *out = WorkingSet::INVALID_ID;
        });

    if (ret != PlanStage::ADVANCED) {
        return ret;
    }

    if (!record) {
        // We hit EOF. If we are tailable and have already seen data, leave us in a state to pick up
        // where we left off on the next call to work(). Otherwise, the EOF is permanent.
        if (_params.tailable && !_lastSeenId.isNull()) {
            _cursor.reset();
        } else {
            _commonStats.isEOF = true;
        }

        // For change collections, advance '_latestOplogEntryTimestamp' to the current snapshot
        // timestamp, i.e. the latest available timestamp in the global oplog.
        if (_params.shouldTrackLatestOplogTimestamp && collection()->ns().isChangeCollection()) {
            setLatestOplogEntryTimestampToReadTimestamp();
        }
        return PlanStage::IS_EOF;
    }

    _lastSeenId = record->id;
    if (_params.assertTsHasNotFallenOff) {
        assertTsHasNotFallenOff(*record);
    }
    if (_params.shouldTrackLatestOplogTimestamp) {
        setLatestOplogEntryTimestamp(*record);
    }

    WorkingSetID id = _workingSet->allocate();
    WorkingSetMember* member = _workingSet->get(id);
    member->recordId = std::move(record->id);
    member->resetDocument(opCtx()->recoveryUnit()->getSnapshotId(), record->data.releaseToBson());
    _workingSet->transitionToRecordIdAndObj(id);

    return returnIfMatches(member, id, out);
}

void CollectionScan::setLatestOplogEntryTimestampToReadTimestamp() {
    const auto readTimestamp = opCtx()->recoveryUnit()->getPointInTimeReadTimestamp(opCtx());

    // If we don't have a read timestamp, we take no action here.
    if (!readTimestamp) {
        return;
    }

    // Otherwise, verify that it is equal to or greater than the last recorded timestamp, and
    // advance it accordingly.
    tassert(
        6663000,
        "The read timestamp must always be greater than or equal to the last recorded timestamp",
        *readTimestamp >= _latestOplogEntryTimestamp);
    _latestOplogEntryTimestamp = *readTimestamp;
}

void CollectionScan::setLatestOplogEntryTimestamp(const Record& record) {
    auto tsElem = record.data.toBson()[repl::OpTime::kTimestampFieldName];
    uassert(ErrorCodes::Error(4382100),
            str::stream() << "CollectionScan was asked to track latest operation time, "
                             "but found a result without a valid 'ts' field: "
                          << record.data.toBson().toString(),
            tsElem.type() == BSONType::bsonTimestamp);
    LOGV2_DEBUG(550450,
                5,
                "Setting _latestOplogEntryTimestamp to the max of the timestamp of the current "
                "latest oplog entry and the timestamp of the current record",
                "latestOplogEntryTimestamp"_attr = _latestOplogEntryTimestamp,
                "currentRecordTimestamp"_attr = tsElem.timestamp());
    _latestOplogEntryTimestamp = std::max(_latestOplogEntryTimestamp, tsElem.timestamp());
}

void CollectionScan::assertTsHasNotFallenOff(const Record& record) {
    auto oplogEntry = uassertStatusOK(repl::OplogEntry::parse(record.data.toBson()));
    invariant(_specificStats.docsTested == 0);

    // If the first entry we see in the oplog is the replset initialization, then it doesn't matter
    // if its timestamp is later than the timestamp that should not have fallen off the oplog; no
    // events earlier can have fallen off this oplog.
    // NOTE: A change collection can be created at any moment as such it might not have replset
    // initialization message, as such this case is not fully applicable for the change collection.
    const bool isNewRS =
        oplogEntry.getObject().binaryEqual(BSON("msg" << repl::kInitiatingSetMsg)) &&
        oplogEntry.getOpType() == repl::OpTypeEnum::kNoop;

    // Verify that the timestamp of the first observed oplog entry is earlier than or equal to
    // timestamp that should not have fallen off the oplog.
    const bool tsHasNotFallenOff = oplogEntry.getTimestamp() <= *_params.assertTsHasNotFallenOff;

    uassert(ErrorCodes::OplogQueryMinTsMissing,
            str::stream()
                << "Specified timestamp has already fallen off the oplog for the input timestamp: "
                << *_params.assertTsHasNotFallenOff
                << ", oplog entry: " << oplogEntry.getEntry().toString(),
            isNewRS || tsHasNotFallenOff);
    // We don't need to check this assertion again after we've confirmed the first oplog event.
    _params.assertTsHasNotFallenOff = boost::none;
}

namespace {
bool shouldIncludeStartRecord(const CollectionScanParams& params) {
    return params.boundInclusion ==
        CollectionScanParams::ScanBoundInclusion::kIncludeBothStartAndEndRecords ||
        params.boundInclusion == CollectionScanParams::ScanBoundInclusion::kIncludeStartRecordOnly;
}

bool shouldIncludeEndRecord(const CollectionScanParams& params) {
    return params.boundInclusion ==
        CollectionScanParams::ScanBoundInclusion::kIncludeBothStartAndEndRecords ||
        params.boundInclusion == CollectionScanParams::ScanBoundInclusion::kIncludeEndRecordOnly;
}

bool pastEndOfRange(const CollectionScanParams& params, const WorkingSetMember& member) {
    if (params.direction == CollectionScanParams::FORWARD) {
        // A forward scan ends with the maxRecord when it is specified.
        if (!params.maxRecord) {
            return false;
        }

        const auto& endRecord = params.maxRecord->recordId();
        return member.recordId > endRecord ||
            (member.recordId == endRecord && !shouldIncludeEndRecord(params));
    } else {
        // A backward scan ends with the minRecord when it is specified.
        if (!params.minRecord) {
            return false;
        }
        const auto& endRecord = params.minRecord->recordId();

        return member.recordId < endRecord ||
            (member.recordId == endRecord && !shouldIncludeEndRecord(params));
    }
}

bool beforeStartOfRange(const CollectionScanParams& params, const WorkingSetMember& member) {
    if (params.direction == CollectionScanParams::FORWARD) {
        // A forward scan begins with the minRecord when it is specified.
        if (!params.minRecord) {
            return false;
        }

        const auto& startRecord = params.minRecord->recordId();
        return member.recordId < startRecord ||
            (member.recordId == startRecord && !shouldIncludeStartRecord(params));
    } else {
        // A backward scan begins with the maxRecord when specified.
        if (!params.maxRecord) {
            return false;
        }
        const auto& startRecord = params.maxRecord->recordId();
        return member.recordId > startRecord ||
            (member.recordId == startRecord && !shouldIncludeStartRecord(params));
    }
}
}  // namespace

PlanStage::StageState CollectionScan::returnIfMatches(WorkingSetMember* member,
                                                      WorkingSetID memberID,
                                                      WorkingSetID* out) {
    ++_specificStats.docsTested;

    // The 'minRecord' and 'maxRecord' bounds are always inclusive, even if the query predicate is
    // an exclusive inequality like $gt or $lt. In such cases, we rely on '_filter' to either
    // exclude or include the endpoints as required by the user's query.
    if (pastEndOfRange(_params, *member)) {
        _workingSet->free(memberID);
        _commonStats.isEOF = true;
        return PlanStage::IS_EOF;
    }

    // For clustered collections, seekNear() is allowed to return a record prior to the
    // requested minRecord for a forward scan or after the requested maxRecord for a reverse
    // scan. Ensure that we do not return a record out of the requested range. Require that the
    // caller advance our cursor until it is positioned within the correct range.
    //
    // In the future, we could change seekNear() to always return a record after minRecord in the
    // direction of the scan. However, tailable scans depend on the current behavior in order to
    // mark their position for resuming the tailable scan later on.
    if (beforeStartOfRange(_params, *member)) {
        _workingSet->free(memberID);
        return PlanStage::NEED_TIME;
    }

    if (!Filter::passes(member, _filter)) {
        _workingSet->free(memberID);
        if (_params.shouldReturnEofOnFilterMismatch) {
            _commonStats.isEOF = true;
            return PlanStage::IS_EOF;
        }
        return PlanStage::NEED_TIME;
    }
    if (_params.stopApplyingFilterAfterFirstMatch) {
        _filter = nullptr;
    }
    *out = memberID;
    return PlanStage::ADVANCED;
}

bool CollectionScan::isEOF() {
    return _commonStats.isEOF;
}

void CollectionScan::doSaveStateRequiresCollection() {
    if (_cursor) {
        _cursor->save();
    }
}

void CollectionScan::doRestoreStateRequiresCollection() {
    if (_cursor) {
        // If this collection scan serves a read operation on a capped collection, only restore the
        // cursor if it can be repositioned exactly where it was, so that consumers don't silently
        // get 'holes' when scanning capped collections. If this collection scan serves a write
        // operation on a capped collection like a clustered TTL deletion, exempt this operation
        // from the guarantees above.
        const auto tolerateCappedCursorRepositioning = expCtx()->getIsCappedDelete();
        const bool couldRestore = _cursor->restore(tolerateCappedCursorRepositioning);
        uassert(ErrorCodes::CappedPositionLost,
                str::stream()
                    << "CollectionScan died due to position in capped collection being deleted. "
                    << "Last seen record id: " << _lastSeenId,
                couldRestore);
    }
}

void CollectionScan::doDetachFromOperationContext() {
    if (_cursor)
        _cursor->detachFromOperationContext();
}

void CollectionScan::doReattachToOperationContext() {
    if (_cursor)
        _cursor->reattachToOperationContext(opCtx());
}

unique_ptr<PlanStageStats> CollectionScan::getStats() {
    // Add a BSON representation of the filter to the stats tree, if there is one.
    if (nullptr != _filter) {
        BSONObjBuilder bob;
        _filter->serialize(&bob);
        _commonStats.filter = bob.obj();
    }

    unique_ptr<PlanStageStats> ret = std::make_unique<PlanStageStats>(_commonStats, STAGE_COLLSCAN);
    ret->specific = std::make_unique<CollectionScanStats>(_specificStats);
    return ret;
}

const SpecificStats* CollectionScan::getSpecificStats() const {
    return &_specificStats;
}

}  // namespace mongo