sync with live: true triggers "change" events twice

[gr2m]

var PouchDB = require('pouchdb').defaults({
  // memdown makes no difference, just added for sake of simplicity
  db: require('memdown')
})

var db1 = new PouchDB('db1')
var db2 = new PouchDB('db2')

db1.put({_id: 'a'})

.then(function () {
  return db2.put({_id: 'b'})
})

.then(function () {
  db1.sync(db2, {live: true}).on('change', function (change) {
    console.log(change.direction, change.change.docs)
  })
})

Output looks like this

$ node debug.js 
push [ { _id: 'a', _rev: '1-4c355ea809e3720366f3ce63e030039f' } ]
pull [ { _id: 'b', _rev: '1-30da76a9658982ad291add2de458cb7c' } ]
push [ { _id: 'a', _rev: '1-4c355ea809e3720366f3ce63e030039f' } ]
pull [ { _id: 'b', _rev: '1-30da76a9658982ad291add2de458cb7c' } ]

WIthout the live: true option, it correctly triggers the "change" events only once per document. Is this by design? Or some kind of race condition maybe?

[daleharvey]

Certainly not by design, just a straight up bug I think

[varjmes]

I poked around a bit: I'm unfamiliar with how everything fits together but I can say with confidence that your journey to fix this may start with this line: https://github.com/pouchdb/pouchdb/blob/master/lib/sync.js#L80

  if (opts.live) {
    this.push.on('complete', self.pull.cancel.bind(self.pull));
    this.pull.on('complete', self.push.cancel.bind(self.push));
  }

Not a great lot of help, but a start :)

[nolanlawson]

Mmm yeah if this only occurs with sync() but not with replicate(), then it's definitely a bug in sync.js.

[daleharvey]

Added a test here, its a bug in replicate - #4348, when we have a replication, when a document is written to a database via pull replication, we trigger a change on the push replication that is empty

[daleharvey]

Fairly trivial fix - 764ba8a, however looks like the retry tests are broken somehow (they depend on extra events being fired) and now failing

[daleharvey]

So the retry test depends on this behaviour, the retry test does rep.on('change', postNewDoc) to post docs in a loop, currently we only fire the change event once the checkpoint has successfully written, if that fails, the batch fails, replication gets retried but the next document is already written so it will fire the empty change event

[daleharvey]

We can change the behaviour for replication to fire change events before the checkpointer, however that is going to have a knock on effect (the tests fail when I do that)

[daleharvey]

Fixed in fe4cc24