[DevLog] Engineering Resilience: Decoupling Industrial Data Streams from UI Response on a VPS

[Shinar-of-Clark]

1. Initial Observation & The "Crisis"
   After deploying the Cable Sheath Circulating Current Monitoring System to an Oracle VPS (1 vCPU / 1GB RAM), I encountered a critical availability issue.

When interacting with the Device Monitor or Alarm Management tabs, the frontend would hang indefinitely, followed by a cascade of 502 Bad Gateway errors in the Chrome DevTools.

Symptom: Complete UI paralysis upon tab switching.

Initial Assessment: This didn't look like a simple syntax bug; it felt like a Resource Exhaustion event triggered by high-frequency data serialization (1Hz) and real-time waveform rendering on minimal hardware.

2. Root Cause Analysis (Layered Troubleshooting)
   To isolate the issue, I followed a standard industrial troubleshooting workflow:

Environment Isolation: First, I verified the application on my Local Development Environment. The result: the UI was extremely fluid, and tab switching was instantaneous. This confirmed that the frontend business logic was 100% correct and efficient.

Cloud Infrastructure Audit: I shifted focus to the VPS. I audited network latency, port listening (Port 8050), and Gunicorn worker logs.

Identifying the Conflict: The root cause was a collision between Synchronous I/O Blocking and limited system resources. The backend was attempting to connect to an offline Edge Device. In a 1vCPU/1GB RAM environment, the Synchronous TCP Handshake created a "blocking wait," which quickly filled the request queue and led to Gateway timeouts (502).

3. Implementation & Optimization Strategy
   To break this deadlock without upgrading to expensive hardware, I implemented a three-tier optimization:

Defense Layer: Hierarchical Circuit Breaker
I refactored the acquisition logic. If a connection cannot be established within 1 second, the system triggers an Exponential Backoff period (5s -> 60s -> 1h). During this "cooldown," the backend intercepts invalid requests and serves an "Offline" status immediately. This freed the CPU threads from waiting for non-existent signals.

Kernel Layer: Swap Paging Control
To handle the memory spikes during Plotly chart serialization, I manually configured 2G of Virtual Memory (SWAP) at the Linux kernel level. This provided a vital "pressure relief valve" for the 1GB physical RAM.

Architecture Layer: Gthread Concurrency
I migrated the execution model from sync to gthread mode. By isolating "System Clock," "Data Acquisition," and "UI Interaction" into different logic lanes, I achieved a much higher degree of Asynchronous Decoupling.

4. Conclusion & Architectural Philosophy
   Outcome: The interaction experience on the cloud VPS has improved significantly. While the 1GB RAM "physical ceiling" prevents the UI from being as perfectly smooth as the local environment under high-frequency load, the persistent service crashes have been successfully eliminated.

Engineering Insight: The core of stability isn't about throwing hardware at a problem; it's about Graceful Degradation. In an industrial IIoT context, the cloud platform's primary role is the asynchronous presentation of persisted data, not the real-time passthrough of physical link latency.

Final Strategy: Having verified that the logic is fully functional and stable in the local environment, I've opted for "Appropriate Optimization" rather than "Over-Engineering." The focus remains on maintaining the integrity of the data decoupling logic.