Company A has async pooling based webservice for notifications. Company B checks for notifications. Every time when it reads new notifications A deletes them from the system. Thus subsequent read requests return only new notifications. There is also requirement for the client B to interrupt the connection if there is no response within 30 sec.
This causes one potential problem: Due to unexpected slowness it is possible for A get the request deleted a notification and send the response back while B is already interrupted the connection. Under this scenario notification gets lost. Now one can argue that the core problem lies within operation realm (the HTTP response must be delivered withing 20 sec ) still on practice it is not always feasible.
How to design B (the client) to avoid this problem?
One way I can see is to do not delete the notifications by A and make B be aware of its state, so that it knows starting from what ID it needs to process notifications, but that presumes that ID will be sequential. Which is controlled by A. Even if B defines its own sequence A still has to be altered to return it back.
Are there any other approaches?
Thanks!
Web services in general are unreliable enough that it's rarely a good idea to make a "read" request serve double-duty as a "delete" request, especially without the client's knowledge. There is just too much risk of a connection dropping or timing out. There is no way to get around this only by modifying the client, because it's the server that is at fault here - the way it's designed is fundamentally unsuited for a web service.
I think you're on the right track with the incrementing IDs idea. The client knows (or can be modified to know) which notifications it's received, so if it can supply the ID of the last message it's received when it polls for notifications, the server should be able to respond based on that ID.
It really seems like Company A's webservice should be synchronous instead of asynchronous. If that is not possible, it may be a good idea to send a "ACK"-like response to a new Company A webservice that indicates a specific notification was received (by Company B) and can be deleted.
Related
I am looking into building a simple solution where producer services push events to a message queue and then have a streaming service make those available through gRPC streaming API.
Cloud Pub/Sub seems well suited for the job however scaling the streaming service means that each copy of that service would need to create its own subscription and delete it before scaling down and that seems unnecessarily complicated and not what the platform was intended for.
On the other hand Kafka seems to work well for something like this but I'd like to avoid having to manage the underlying platform itself and instead leverage the cloud infrastructure.
I should also mention that the reason for having a streaming API is to allow for streaming towards a frontend (who may not have access to the underlying infrastructure)
Is there a better way to go about doing something like this with the GCP platform without going the route of deploying and managing my own infrastructure?
If you essentially want ephemeral subscriptions, then there are a few things you can set on the Subscription object when you create a subscription:
Set the expiration_policy to a smaller duration. When a subscriber is not receiving messages for that time period, the subscription will be deleted. The tradeoff is that if your subscriber is down due to a transient issue that lasts longer than this period, then the subscription will be deleted. By default, the expiration is 31 days. You can set this as low as 1 day. For pull subscribers, the subscribers simply need to stop issuing requests to Cloud Pub/Sub for the timer on their expiration to start. For push subscriptions, the timer starts based on when no messages are successfully delivered to the endpoint. Therefore, if no messages are published or if the endpoint is returning an error for all pushed messages, the timer is in effect.
Reduce the value of message_retention_duration. This is the time period for which messages are kept in the event a subscriber is not receiving messages and acking them. By default, this is 7 days. You can set it as low as 10 minutes. The tradeoff is that if your subscriber disconnects or gets behind in processing messages by more than this duration, messages older than that will be deleted and the subscriber will not see them.
Subscribers that cleanly shut down could probably just call DeleteSubscription themselves so that the subscription goes away immediately, but for ones that shut down unexpectedly, setting these two properties will minimize the time for which the subscription continues to exist and the number of messages (that will never get delivered) that will be retained.
Keep in mind that Cloud Pub/Sub quotas limit one to 10,000 subscriptions per topic and per project. Therefore, if a lot of subscriptions are created and either active or not cleaned up (manually, or automatically after expiration_policy's ttl has passed), then new subscriptions may not be able to be created.
I think your original idea was better than ephemeral subscriptions tbh. I mean it works, but it feels totally unnatural. Depending on what your requirements are. For example, do clients only need to receive messages while they're connected or do they all need to get all messages?
Only While Connected
Your original idea was better imo. What I probably would have done is to create a gRPC stream service that clients could connect to. The implementation is essentially an observer pattern. The consumer will receive a message and then iterate through the subscribers to do a "Send" to all of them. From there, any time a client connects to the service, it just registers itself with that observer collection and unregisters when it disconnects. Horizontal scaling is passive since clients are sticky to whatever instance they've connected to.
Everyone always get the message, if eventually
The concept is similar to the above but the client doesn't implicitly un-register from the observer on disconnect. Instead, it would register and un-register explicitly (through a method/command designed to do so). Modify the 'on disconnected' logic to tell the observer list that the client has gone offline. Then the consumer's broadcast logic is slightly different. Now it iterates through the list and says "if online, then send, else queue", and send the message to a ephemeral queue (that belongs to the client). Then your 'on connect' logic will send all messages that are in queue to the client before informing the consumer that it's back online. Basically an inbox. Setting up ephemeral, self-deleting queues is really easy in most products like RabbitMQ. I think you'll have to do a bit of managing whether or not it's ok to delete a queue though. For example, never delete the queue unless the client explicitly unsubscribes or has been inactive for so long. Fail to do that, and the whole inbox idea falls apart.
The selected answer above is most similar to what I'm subscribing here in that the subscription is the queue. If I did this, then I'd probably implement it as an internal bus instead of an observer (since it would be unnecessary) - You create a consumer on demand for a connecting client that literally just forwards the message. The message consumer subscribes and unsubscribes based on whether or not the client is connected. As Kamal noted, you'll run into problems if your scale exceeds the maximum number of subscriptions allowed by pubsub. If you find yourself in that position, then you can unshackle that constraint by implementing the pattern above. It's basically the same pattern but you shift the responsibility over to your infra where the only constraint is your own resources.
gRPC makes this mechanism pretty easy. Alternatively, for web, if you're on a Microsoft stack, then SignalR makes this pretty easy too. Clients connect to the hub, and you can publish to all connected clients. The consumer pattern here remains mostly the same, but you don't have to implement the observer pattern by hand.
(note: arrows in diagram are in the direction of dependency, not data flow)
I have a requirement to count the jetty transactions and measure the time it took to process the request and get back the response using JMX for our monitoring system.
I am using Jetty 8.1.7 and I can’t seem to find a proper way to do this. I basically need to identify when request is sent (due to Jetty Async approach this is triggered from thread A) and when the response is complete (as the oncompleteResponse is done in another thread).
I usually use ThreadLocal for such state in other areas I need similar functionality, but obviously this won’t work here.
Any ideas how to overcome?
To use jetty's async requests you basically have to subclass ContentExchange and override its methods. So you can add an extra field to it which would contain a timestamp of when the request was sent, and use it later in your onResponseComplete() method to measure the processing time. If you need to know the time when your request was actually sent to the server instead of when it was created you can override the onRequestCommitted() and onRequestComplete() methods.
If I have a web api service (Order Notification) that allows a third party client to call in (they must call in to us, not use pushing to them) periodically (every 10 minutes) and gets new orders it has not yet received, how do I deal with failures?
For example there are 10 new Orders the client has not received since they last called in. The client calls into our Order Notification service. We retrieve the orders we have not sent (10 in this case). We update these 10 Orders as sent and return the response to the client.
However the client did not receive the response (sometime happened after leaving us e.g. http time out or something else).
So now we have a problem where on our side we have marked the orders as sent but the client never received them.
Any thoughts on how to solve this?
Just an idea, can you assign the caller some sort of identifier and when the caller succeeds it replies back saying it has acknowledged the request? The server will never know if something failed on the client side unless the client reports it.
For example, when caller A calls in for the requests it may do something like this:
call -> http://server/requests
server replies back with some xml that contains the result set for this caller along with a unique identifier that it will track to know if that particular call had a response (you can time out this identifier after a reasonable period of time)
when the client gets the request it can call back again
call -> http://server/requestComplete?id=[generatedID]
and the server marks it successful.
Lots of API's require some sort of identification token so it would already lend itself well to this kind of send/ack messaging system.
If you have access to both sides of the system you could create a received request so once the client picking up the data has received it makes a request to the original host telling that it's received successfully.
Imagine 3 system components:
1. External ecommerce web service to process credit card transactions
2. Local Database to store processing results
3. Local UI (or win service) to perform payment processing of the customer order document
The external web service is obviously not transactional, so how to guarantee:
1. results to be eventually persisted to database when received from web service even in case the database is not accessible at that moment(network issue, db timeout)
2. prevent clients from processing the customer order while payment initiated by other client but results not successfully persisted to database yet(and waiting in some kind of recovery queue)
The aim is to do processing having non transactional system components and guarantee the transaction won't be repeated by other process in case of failure.
(please look at it in the context of post sell payment processing, where multiple operators might attempt manual payment processing; not web checkout application)
Ask the payment processor whether they can detect duplicate transactions based on an order ID you supply. Then if you are unable to store the response due to a database failure, you can safely resubmit the request without fear of double-charging (at least one PSP I've used returned the same response/auth code in this scenario, along with a flag to say that this was a duplicate).
Alternatively, just set a flag on your order immediately before attempting payment, and don't attempt payment if the flag was already set. If an error then occurs during payment, you can investigate and fix the data at your leisure.
I'd be reluctant to go down the route of trying to automatically cancel the order and resubmitting, as this just gets confusing (e.g. what if cancelling fails - should you retry or not?). Best to keep the logic simple so when something goes wrong you know exactly where you stand.
In any system like this, you need robust error handling and error reporting. This is doubly true when it comes to dealing with payments, where you absolutely do not want to accidentaly take someone's money and not deliver the goods.
Because you're outsourcing your payment handling to a 3rd party, you're ultimately very reliant on the gateway having robust error handling and reporting systems.
In general then, you hand off control to the payment gateway and start a task that waits for a response from the gateway, which is either 'payment accepted' or 'payment declined'. When you get that response you move onto the next step in your process and everything is good.
When you don't get a response at all (time out), or the response is invalid, then how you proceed very much depends on the payment gateway:
If the gateway supports it send a 'cancel payment' style request. If the payment cancels successfully then you probably want to send the user to a 'sorry, please try again' style page.
If the gateway doesn't support canceling, or you have no communications to the gateway then you will need to manually (in person, such as telephone) contact the 3rd party to discover what went wrong and how to proceed. To aid this you need to dump as much detail as you have to error logs, such as date/time, customer id, transaction value, product ids etc.
Once you're back on your site (and payment is accepted) then you're much more in control of errors, but in brief if you cant complete the order, then you should either dump the details to disk (such as csv file for manual handling) or contact the gateway to cancel the payment.
Its also worth having a system in place to track errors as they occur, and if an excessive number occur then consider what should happen. If its a high traffic site for example you may want to temporarily prevent further customers from placing orders whilst the issue is investigated.
Distributed messaging.
When your payment gateway returns submit a message to a durable queue that guarantees a handler will eventually get it and process it. The handler would update the database. Should failure occur at that point the handler can leave the message in the queue or repost it to the queue, or post an alternate message.
Should something occur later that invalidates the transaction, another message could be queued to "undo" the change.
There's a fair amount of buzz lately about eventual consistency and distribute messaging. NServiceBus is the new component hotness. I suggest looking into this, I know we are.
Are there any best practices that dictate the maximum time between an asynchronous call and its corresponding response.
Basically I have a process that takes a long time to run (eg: 5 minutes). Option 1: I could expose the process as an asynchronous call. In which case the user calls my service and then at some later time, I respond with a process status.
Option 2
The other way I could implement it is to setup the system such that there is a one-way operation on my web-service that begins the process and immediately returns an id for the process. I could then mandate that the consumer provide a one-way operation, that I can call and report back when the process is done.
The first option is easier as I dont have to mandate anything from the caller. The second seems better as I can report back at anytime (5 minutes to years later).
As I have complete control over the caller and its an internally available service, I am leaning towards option 2.
So I am wondering if there are any time limits imposed on async calls (can they span days? if not what is the best practice). Is option 2 a standard pattern employed?
References would be extremely useful.
Option #2 is better as it's more event driven.
However, there exists an Option #3. Client issues request to server. Server queues request and responds with the id. Client checks back every so often, passing the request id, to see if it's completed.
This way you don't have to depend on the client being available when the request is completed.
I'd probably mix options #2 and #3 and let the client choose if they want an event fired on their side or if they just want to check back later.
UPDATE
Rajah has asked about the maximum time between async request and response. For a WEB application, this is typically measured in seconds. Most servers have timeout values that are typically defaulted in the 30 second range. Personally, I think this is too long.
Consider that an Async call requires the communications channel between the client and server to be open for the duration. How many of those channels can a single server handle? More to the point, how many channels will you have to maintain as requests are made? This can become quite outrageous even if you do control both ends.
Whatever is hosting your services is going to determine the maximum amount of time to keep a request open. Again, every server I've seen measures this in seconds.