0

I am trying to solve two situations here related to making POST requests and handling responses.

  1. Making a request and having request timed out due to no server response, as well as assuming that in this case nothing has been written in a database hosted on the server.
  2. Making a request and having request timed out, but now things are written on a server, and server could / will eventually send a response. So I am trying to figure out what should I do in each of these two situations. Situation no. 2 is most important for me, and I don't really have an idea how to solve it.

I only have a solution for the first situation, but I would be interested to hear opinion / best practices about how to solve that too.

I am using Swift 2.0 if that matters, so it is iOS, and I am using custom wrapper around NSURLSession and related classes to make requests.

As an additional info I can describe how my app works:

I create a model object, call it Data. So I send a JSON representation of that object to a server, write it into database, and get a response from a server that it is recorded.

This is a task assigned to my DataManager class (which is a singleton ), and it handles requests and responses. When it receives a response from a server, it updates an appropriate model object, and through delegation / notification updates UI.

When a user taps the send button, I show activity indicator and set an appropriate label's text to Sending... And that is where these two situations can eventually arise...

In situation no. 1, I will tell the user that request has timed out, cancel the request and show a retry button. Problem solved.

Situation no. 2 is trickier... The user will tap send button, and he will wait. But imagine if this happen:

  • data sent to a server is successfully written
  • but the server response was very slow and exceeds the timeout limit on a client app
  • I show the request failed message to a user, but the request hasn't failed actually So what can happen here, is that user will experience inconsistency between what he sees on a device and what is really on the server (at least for some short period of time).

So when next refresh occurs, the user will see the previously sent object, even if he has gotten the message which says "operation failed"

How do you handle this?

1

You could ping the server several times before you send a request and check the latency of the pings. Set your timeout greater than the maximum return time (say 2x to be generous). When your client sends a request to the server, send the latency information as well so the server doesn't timeout trying to reach the client.

You could make your pings in the background, for example 1 every 10 seconds, then use the past few pings. As it's a background process you can set the timeout incredibly long1 before you try the next ping as it should not affect the client's main usage.

If you cannot get a single ping response back to calculate latency it's pretty safe to say the server would not receive your initial message.

1While it could be long to test if you can get anything back at all you have to remember your client will probably dislike being stuck in a minute long "sending..." screen because you deemed that to be fine.

  • Thanks, interesting solution. Have to think about it a bit and see how it will fit into my project. Sorry for the late response though. – Whirlwind Oct 9 '16 at 8:53
1

The server could send a 202 status (request accepted). Which means the request has been accepted by the server, but hasn't yet been acted upon. And the action can fail obviously, but usually won't fail. That seems a reasonable approach for request where you regularly expect long processing times.

The server could give each request a serial number and have a log of accepted and processed requests, and you check from time to time of the request has been processed with the server.

  • Okay, nice idea. But I guess it is possible that even 202 response has a big delay, right? Server wont neccessarily response with 202 immediately. I mean nobody guarantees that, right? So we are back to the same problem... – Whirlwind Oct 9 '16 at 8:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.