We have 1M+ devices registered. Is there a way to limit how quickly the messages get delivered? Obviously it's real hard to scale if 1M+ notifications at the exact same time cause a massive spike of traffic to your backend. Would be great if instead of all the messages getting delivered immediately to all devices, you could make it only send X messages per second.
The best way to control the delivery of those message is actually by calling FCM with the token IDs yourself, preferably with the batched delivery feature from the legacy API (look for the registration_ids parameter there). You can scale this up to as many calls to the API as you need to deliver your message to all devices.
Using topics is also possible, but you lose control of the delivery performance since the fan-out happens in a process you don't control.
Alternatively: consider sending a data message that contains a timestamp on when it should be displayed. That way you separate the delivery time from the display time, removing the critical path (but of course introducing other considerations).
Related
I'm currently working on an IRC-Chat and we want to add the option to chat with other people privately (User-To-User) which works fine, but the messages aren't stored, meaning that a user loses all private messages after disconnecting. They also can't message a person once they have disconnected.
All of this isn't the case with channels, where messages are stored for X time, allowing a asynchronous communication.
Is there a way of allowing asynchronous messaging for private User-To-User messages without storing the messages in an extra system? Or is this simply a limitation of IRC?
IRC isn't designed to store messages - it's a feature, not a bug. One way to get around this is to configure the server to 'simulate' past messages (essentially pretending to be the other user to send past messages). To do this, you would need to store messages - however, it would be on the server itself, not on a separate database or a third party.
I am developing an Android app which basically does this: On the landing(home) page it shows a couple of words. These words need to be updated on daily basis. Secondly, there is an 'experiences' tab in which a list of user experiences (around 500) shows up with their profile pic, description,etc.
This basic app is expected to get around 1 million users daily who will open the app daily at least once to see those couple of words. Many may occasionally open up the experiences section.
Thirdly, the app needs to have a push notification feature.
I am planning to purchase a managed wordpress hosting, set up a website, and add a post each day with those couple of words, use the JSON-API to extract those words and display them on app's home page. Similarly for the experiences, I will add each as a wordpress post and extract them from the Wordpress database. The reason I am choosing wordpress is that it has ready made interfaces for data entry which will save my time and effort.
But I am stuck on this: will the wordpress DB be able to handle such large amount of queries ? With such a large userbase and spiky traffic, I suspect I might cross the max. concurrent connections limit.
What's the best strategy in my case ? Should I use WP, or use firebase or any other service ? I need to make sure the scheme is cost effective also.
My app is basically very similar to this one:
https://play.google.com/store/apps/details?id=com.ekaum.ekaum
For push notifications, I am planning to use third party services.
Kindly suggest the best strategy I should go with for designing the back end of this app.
Thanks to everyone out there in advance who are willing to help me in this.
I have never used Wordpress, so I don't know if or how it could handle that load.
You can still use WP for data entry, and write a scheduled function that would use WP's JSON API to copy that data into Firebase.
RTDB-vs-Firestore scalability states that RTDB can handle 200 thousand concurrent connections and Firestore 1 million concurrent connections.
However, if I get it right, your app doesn't need connections to be active (i.e. receive real-time updates). You can get your data once, then close the connection.
For RTDB, Enabling Offline Capabilities on Android states that
On Android, Firebase automatically manages connection state to reduce bandwidth and battery usage. When a client has no active listeners, no pending write or onDisconnect operations, and is not explicitly disconnected by the goOffline method, Firebase closes the connection after 60 seconds of inactivity.
So the connection should close by itself after 1 minute, if you remove your listeners, or you can force close it earlier using goOffline.
For Firestore, I don't know if it happens automatically, but you can do it manually.
In Firebase Pricing you can see that 100K Firestore document reads is $0.06. 1M reads (for the two words) should cost $0.6 plus some network traffic. In RTDB, the cost has to do with data bulk, so it requires some calculations, but it shouldn't be much. I am not familiar with the pricing small details, so you should do some more research.
In the app you mentioned, the experiences don't seem to change very often. You might want to try to build your own caching manually, and add the required versioning info in the daily data.
Edit:
It would possibly be more efficient and less costly if you used Firebase Hosting, instead of RTDB/Firestore directly. See Serve dynamic content and host microservices with Cloud Functions and Manage cache behavior.
In short, you create a HTTP function that reads your database and returns the data you need. You configure hosting to call that function, and configure the cache such that subsequent requests are served the cached result via hosting (without extra function invocations).
We have a simple application, who upon every update of an entity sends out a notification to SNS(it could very well have been any other queuing system). Clients are listening to these notifications and they do a get of updated entity based on these notifications.
The problem we are facing is, when clients do a get, sometimes data is not replicated and we return 404 or sometimes stale data(even worse).
How can we mitigate this while sending notifications?
Here are Few strategies to mitigate this with pros and cons
Instead of sending notification from application send notification using database streams
For example dynamodb streams ans aws lambda. This pattern can be useful in the case of multiregion deployment as well. where all the subscriber, publisher will subscribe to their regional database streams. And also atomicity of sending message and writing to database is preserved. And we wont loose events in the case of regional failure.
Send delayed messages to your broker
Some borkers like activemq and sqs support this functionality, but SNS does not. A workaround for that could be writing to sqs queue which then writes to sns. This might be a good option when your database does not support streams.
Send special error code for retry-able gets
Since we know that eventual consistency is there we can return special error code to clients, so that they can retry based on this error code. The retry strategy should be exponential backoff. but this may mean giving away your problems to clients. Also we should have some sort of versioning in place.
Fetch from another region
If entity is not found in the same region application can go to another region or master database to fetch it. NOTE Don't do this. as it is an anti pattern. I am mentioning it here just for the sake of completion.
Send the full entity in message
If entities to be fetched by rest service is small and there are no security constrain around who can access what, we can send the full entity in message. This is ensure that client don't have to do explicit fetch of it every time a new message is arrived.
Our Firebase server currently uses a one device per topic. However as users of the App are increasing the overhead of sending many topic sends is starting to be significant. What are our options for grouping sends, given that the targets dataset changes on every send.
We started to look at device groups, but these will be unsupported if we are forced to move to HTTP V1. We had considered just adding a group of users to a topic, but managing the lifespan of the topic could become an issue. We would have to create a new one on every send, working out at what point we could tear this subscription down with impacting any messages which have not been received may be problematic.
And suggestions welcome.
The usual approach for this is to either use topics, or implementing your own topic-like system. In the latter you'd store the relation between the Instance ID and your grouping logic in a database, and then use the batch-sending feature of FCM to deliver to up to 1000 devices at a time.
Could you validate my approach for using Firebase to manage a user notification system?
Basically I want to have user specific channels as well as more general channels which hold notifications. These notifications would appear on an intranet if the user has not viewed them before.
The idea being a server side action will update Firebase endpoints using the REST API either for a specific user or broadcast to everyone. The specific user messages I can easily mark as read and therefore not show them again, its the general broadcast I am struggling slightly with.
I could add a flag(user ID) to the general broadcast to indicate its read but I am concerned about performance as the client would have to check historic broadcast messages for the existence of this flag. I could add a user id to create a new endpoint which should be quicker.
e.g. /notification/general/ - contains the message, this triggers the client which then checks to see if /users/USERID/MessageID exists if it doesnt display the message and create this end point.
Is there something I am missing or is that the best approach?
Are the messages always consumed in-order? If so you could have each client remember the ID of the last message it read in each public channel. You could then use "startAt" on the queue to limit it to only new messages.
If they're not consumed in order, then you'll need some way of storing data about which ones were read and which ones weren't. Perhaps you can have each message get sent out to everyone's personal queue, and then have each user remove read messages.
Since there are already undividual user messages, why not just deliver the broadcasts to everyone individually (think email) rather than trying to store a single copy and figure out who read it.
To reduce bulk, you could store the message content separately, and simply store the ids in a user's queue. Then when they are viewed, you flag them per-user without any additional complexity.
At 100k of users receiving 100 messages a day including broadcasts, with a standard firebase ID (around 20 characters), that comes out to 210,000,000 characters a year (i.e. nothing for a database, and probably still far less than the actual bulk of storing the message body), assuming they never expire and get deleted.