NaySync
"It's not async!"
The year is 2017, and many of us are fully on the log/queue bandwagon. But, much of the world still wants to speak synchronous RPC - what to do?
It's been suggested that, in such a case, an API server could accept a request, enqueue it for later processing while still holding open the original request, and then return a response when the asynchronous processing completes - within a timeout, of course. NaySync is a server that does just this.
The screencast below probably best illustrates this:
On the right we are running a consumer process that runs our application code (you can find the code in test_receiver/sleep_consumer.py. On the left we make curl requests to the server.
- In the first request, we submit a payload that is handled quickly and get back a response. We can see the consumer process it.
- In the second request, we submit a payload that takes a long time to process, and exceeds the specified timeout. We can see the server return early with a request id that we can fetch later, and the consumer finish processing after that.
- In the third request, we look up the request id returned in the previous response, and get the result
Running locally
In one window: nsqd
In another window: cd test-receiver; python consumer.py (requires requests, pynsq)
In another window sbt 'run-main Server'