The mobile app shows “Network error” at random: a few 502s per minute on the API, with no error on the server side

The mobile app of the MasterLinux school (video courses, quizzes, progress tracking) talks to api.masterlinux.lab, a Node.js API running on two EC2 instances (ml-api-01, ml-api-02) behind an Application Load Balancer alb-api.

Since last Tuesday, support has been receiving screenshots of "Network error, try again": a quiz that is not saved, lost progress. CloudWatch confirms it: 0.4 to 0.6% of 502s, spread throughout the day, slightly more during off-peak hours than at peak times. Nobody finds any error in the API logs. The developers blame "the AWS network"; a colleague suggests adding a third instance.

Last week, the team simplified the architecture (CHG-2231): the nginx that ran on each instance in front of Node.js was removed, and the ALB now sends traffic directly to port 3000.

You are the operations engineer in charge of the ticket, on ml-poste-01 (AWS CLI v2), with SSH access to the API instances (account ops).