# FM-KED-007 — 502 Bad Gateway Error (Application Unreachable) **Severity:** S1 — Critical **Recovery Class:** B — Standard Recovery **Covered by Monthly Support:** Yes (known causes only) --- ### Description Nginx returns a **502 Bad Gateway** error because the Django application backend becomes unreachable. In the majority of observed cases, this is caused by **Out Of Memory (OOM)** conditions on the virtual machine hosting the application worker, leading to termination of Gunicorn or equivalent application processes. --- ### Typical Symptoms - HTTP 502 responses from Nginx - Application intermittently unavailable - Gunicorn workers restarting or disappearing - Kernel logs indicating OOM events --- ### Primary Root Cause - Application requests producing excessive memory usage - Large datasets loaded into memory - Insufficient RAM on the worker virtual machine When memory limits are exceeded, the operating system terminates the application process, leaving Nginx without a valid upstream. --- ### Diagnostic Checklist #### Confirm OOM Condition ```bash dmesg | grep -i oom journalctl -k | grep -i kill ``` #### Check Available Memory ```bash free -h ``` #### Verify Application Server Status ```bash systemctl status gunicorn ``` --- ### Recovery Options Apply one or more of the following, depending on constraints. --- #### Option 1: Reduce Request Scope - Apply stricter filters to API requests - Limit requested date ranges - Reduce number of portfolios, instruments, or entities per request - Avoid bulk data retrieval in a single call This reduces memory pressure at the application level. --- #### Option 2: Increase RAM on Worker Virtual Machine - Increase memory allocation on the worker VM - Restart application services after resizing - Verify stability under previous load This addresses the issue at the infrastructure level. --- ### Escalation and Unknown Issues If the issue persists after: - request scope reduction, and - sufficient memory allocation then the incident is classified as an **unknown issue**. Such cases require investigation, profiling, or architectural analysis and are **not covered** by the standard monthly support allocation. --- ### Preventive Notes - Avoid unbounded API queries - Monitor memory usage trends - Define safe defaults and limits at API level - Prefer asynchronous processing for heavy workloads --- ### Responsibility Boundary Finmars SCSA provides best-effort diagnostics and guidance for known memory-related causes. Application design decisions and infrastructure capacity planning beyond documented scenarios require separate analysis. ---