Querying when removing document can sometimes cause incorrect query result.

[amaghfur]

If a query occur during a routine that removes a document, the query can result in strange results. Run the code below anywhere with PouchDB. If you adjust the setTimeout delay so that 'end remove' happens in the middle of the query (after 'start query' and before 'end query'), the query result is weird.

I encountered this when designing a db whose internal logic does occasional document removal while still serving external users with querying functionality. So quite often the remove can happen mid-query.

var db = window.__testDb = new PouchDB('test')
db.destroy().then(function(){
    var ddoc = {
        _id: '_design/type',
        views: {
            type: {
                map: function mapFun(doc) {
                    if (doc.type) {
                        emit(doc.type);
                    }
                }.toString()
            }
        }
    }

    var doc1 = {
        _id: '001',
        item: 'Good Kill',
        type: 'Movie',
        pts: '2015-07-21T07:00:00.000Z'
    }
    var doc2 = {
        _id: '002',
        item: 'The Dark Knight',
        type: 'Movie',
        pts: '2015-07-21T08:00:00.000Z'
    }

    var revToRemove;

    db = new PouchDB('test')
    db.groupBy = function(type){
        return new Promise(function(resolve, reject){
            var ddocName = "groupby_" + type + "_sortedby_lastupdated"

            var mapFunction = "function(doc){ if (doc.type === '" + type + "') { emit(doc.pts) } }"
            var ddoc = {
                _id: '_design/' + ddocName,
                views: {}
            };
            ddoc.views[ddocName] = {
                map: mapFunction.toString()
            };

            db.put(ddoc).then(function(){
                resolve(ddocName)
            }).catch(function(){
                // design doc already exist
                resolve(ddocName)
            })       
        })
    }

    db.put(doc1).then(function(res){
        revToRemove = res.rev
        return db.put(ddoc)
    }).then(function(){
        return db.groupBy('Movie');
    }).then(function(ddocName){
        console.time('put')
        console.time('query')
        setTimeout(function(){
            console.log('start put')
            db.put(doc2).then(function(){
                console.log('end put')
                console.log('start remove')
                db.remove(doc1._id, revToRemove).then(function(){
                    console.log('end remove')
                    console.timeEnd('put')
                })
            })          
        }, 30) // problem extremely likely to happen if timeout set between 20-50ms ('put' time and 'query' time within 20ms of each other)
        console.log('start query')
        return db.query(ddocName, {
            startkey: '2015-07-21T09:00:00.000Z',
            include_docs: true,
            limit: 5,
            descending: true
        })
    }).then(function(res){
        console.log('end query')
        console.timeEnd('query')
        console.log('res:', res)
        console.table(res.rows)
    }).catch(function(err){
        console.log('err:', err)
    })
})

[nolanlawson]

Yeah, there is a race condition when using include_docs:true with query(). In fact, CouchDB has this same problem. The only solution, sadly, is to emit() your documents and avoid using include_docs:true.

[nolanlawson]

I'll note that if we manage to write native secondary indexes as a single transaction (#3775), then we could avoid the race condition.

[amaghfur]

Your suggestion to emit the doc instead of using include_docs achieved my desired result. Thanks!

[NickColley]

Awesome! Feel free to re-open if you've got any more questions! 👍