I have a rather simple EA that in `start` function checks if it's trade time (which is True once every hour) and then, if True, makes HTTP post request to localhost web server (implemented in python). Python is run from Linux subsystem inside Windows Server 2019 Datacenter. Server is hosted in AWS EC2. In request BODY I send last `N` candles, along with some other information (period, timestamp, etc). Python server takes this input, feeds it to TF model and returns signal (double in range 0, 1). During initial and manual testing entire flow worked fine, but after demo testing it was a different story.
During initial testing, I set trades to every 1 minute, and EA was working as expected for several minutes. Switching back to trade time once every hour, I've started receiving errors during API calls (5203 error). As error itself isn't informative, I added retry mechanism to try and fix the issue, which helps only sometimes. Other times however, even after 20 retries I'm stuck on 5203. Only reliable solution was to reboot entire machine, which isn't acceptable in my use case (I want fully autonomous system).
Any ideas on how to fix or at least debug the issue are very welcomed.