Question about the commit

[nieyuyao]

Why write the op before snapshot, is there any consideration?

sharedb-mongo/index.js

Lines 239 to 254 in 9e88c28

	ShareDbMongo.prototype.commit = function(collectionName, id, op, snapshot, options, callback) {
	var self = this;
	var request = createRequestForMiddleware(options, collectionName, op);
	this._writeOp(collectionName, id, op, snapshot, function(err, result) {
	if (err) return callback(err);
	var opId = result.insertedId;
	self._writeSnapshot(request, id, snapshot, opId, function(err, succeeded) {
	if (succeeded) return callback(err, succeeded);
	// Cleanup unsuccessful op if snapshot write failed. This is not
	// necessary for data correctness, but it gets rid of clutter
	self._deleteOp(request.collectionName, opId, function(removeErr) {
	callback(err || removeErr, succeeded);
	});
	});
	});
	};

[alecgibson]

I think this is done like this because back in MongoDB 3, there wasn't support for transactions, so something always needed to be written first, and ops were chosen.

There's a bit of a writeup on the behaviour in the README.

[nieyuyao]

@alecgibson Thanks for your reply. But i still have a question, can i write the snapshot before op?

[alecgibson]

If you're talking about an implementation that you're doing yourself, there's no reason why not, so long as you ensure that you have some way to ensure the op being written is "canonical", and that you don't have conflicting snapshot writes.

As I mentioned above, a new implementation might use MongoDB transactions, although this may not be as performant; there's some (light) discussion on the topic in #127