Meteor Client-side Collection Document Appears and Disappears - meteor

Question

After lots of reading, I'm starting to get a better handle on Meteor's publish/subscribe model. I've removed the autopublish training wheels from my first app and while I have most everything working, I am seeing one issue.
When the app first loads, my publish and subscribe hooks work great. I have a block of code that runs in a Tracker.autorun() block which makes the subscribe calls, I am able to sequentially wait for data from the server using ready() on my subscribe handles, etc.
One feature of my app is that it allows the user to insert new documents into a collection. More specifically, when the user performs a certain action, this triggers an insert. At that point, the client-side JS runs and the insert into MiniMongo completes. The reactive autorun block runs and the client can see the inserted documented. The client updates the DOM with the new inserted data and all is well.
Furthermore, when I peek into the server-side MongoDB, I see the inserted document which means the server-side JS is running fine as well.
Here's where it gets weird. The client-side autorun block runs a second time (I'm not sure why) and this time, the client no longer has the inserted item. When the DOM renders, the newly inserted item is now gone. If I reload the page, all is well again.
Has anyone seen this behavior before? I'm also noticing that the server-side publish call runs once on page load but then it doesn't run again after the insert. This seems wrong because how else will the client get the reconciled data from the server after the insertion (i.e. after Meteor's client-side latency compensation)?
The important functions (ComponentInstances is the collection that is bugging out):
Publish block:
Meteor.publish('allComponentInstances', function (documentId, screenIndex) {
console.log(`documentId: ${documentId} screenIndex: ${screenIndex}`)
const screens = Screens.find({ownerDocumentId: documentId})
const selectedScreen = screens.fetch()[screenIndex]
return ComponentInstances.find({_id: {$in: selectedScreen.allComponentInstanceIds}})
})
Subscription block in autorun:
// ... a bunch of irrelevant code above
const allComponentInstancesHandle = Meteor.subscribe('allComponentInstances', document._id, 0)
if (allComponentInstancesHandle.ready()) {
isReady = true
screens = Screens.find({ownerDocumentId: document._id}).fetch()
const componentInstanceObjects = ComponentInstances.find().fetch()
allComponentInstances = {}
componentInstanceObjects.map((componentInstance) => {
allComponentInstances[componentInstance._id] = componentInstance
})
}

Answer 1

This is most probably you're inserting documents from client side. And you have not set up your permission rules properly. When you remove autopublish and insecure from your app, you are not allowed to insert/update/remove documents into collection unless you have allow/deny rules set up in the server side.
Meteor has a great feature called latency compensation which tries emulate your db operations before it gets the actual write operation in the db. And when the server tries to write in the db, it looks for allow/deny rules.If the permission rules doesn't allow the db operation or Whatever the reason( either allow/deny or authentication) for not actually written in the db, then the server data gets synchronized with your client side db.
This is why i assume you are seeing your document being inserted for the first time and gets disappeared within a second.
check this section of meteor docs.
http://docs.meteor.com/#/full/allow

Answer 2

I ended up solving this a different way. The core issue, I believe, has nothing to do with accept/deny rules. In fact, their role is still hazy to me.
I realize now what I've been reading all along in the Meteor docs: the publish functions return cursors. If the cursor itself doesn't change (e.g. if you're passing specific keys you want to fetch), then it won't really work as a reactive data source in the sense that new documents in a collection will not make the data publish again. You are, after all, still requesting the same keys.
The way forward is to come up with a publish cursor that accurately reflects the reactive data you want to retrieve. This sounds abstract but in practice, it means make sure the cursor is general, not specific to the specific keys you are retrieving.