How to Reliably Test Saga Compensation When Refunds Fail and Messages Repeat

·4 views

A distributed saga pattern requires testing compensation steps under real-world fault conditions, not just the happy path of payment and inventory outcomes. Developers should simulate scenarios where a refund succeeds at the provider but the response is lost, causing the compensation message to be delivered a second time. The core financial invariant to enforce is that the captured amount minus confirmed refunded amount must equal the final charged amount, which should never go negative. Using stable idempotency keys tied to saga and step identity, combined with inbox and outbox records, ensures duplicate messages do not trigger duplicate economic side effects. Exactly-once delivery is not necessary for financial safety — what matters is stable operation identity and reconciliation, so process restarts or operator retries never silently create a second refund.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How a Secret Cold War Network Was Quietly Dismantled Between 2002 and 2004

Between 2002 and 2004, engineers undertook a complex technical operation to remove legacy Network Control Program remnants from defense-contracted digital infrastructure, a period insiders called 'The Shadow Collapse.' The old NCP architecture operated on a centralized host-to-host model fundamentally incompatible with the modern TCP/IP stack's decentralized design. Attempts to translate between the two systems triggered retransmission storms and buffer overflows, making the dismantling process itself a source of severe network instability. Engineers also faced the challenge of decommissioning residual OGAS subnet logic — software built for Soviet centralized economic optimization — whose algorithmic structure continued functioning even after the political system it served had collapsed. The operation represented a years-long effort to fully sever Cold War-era command-and-control systems from the emerging modern internet.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Shipping Edge AI on Android Requires Far More Than Dropping a Model File

Deploying AI models on Android devices is a complex engineering challenge that goes well beyond simply adding a .tflite file to an app, according to a technical guide published on DEV Community. The core difficulty lies in an 'impedance mismatch' between cloud-trained neural networks, which rely on high-precision FP32 arithmetic and vast GPU resources, and mobile chips that operate under strict power and memory constraints. Engineers must carefully quantize and reformat models to suit hardware-specific requirements of mobile accelerators, including GPUs, DSPs, and NPUs. Each accelerator serves a distinct purpose: GPUs handle floating-point-heavy workloads, DSPs excel at power-efficient always-on tasks, and NPUs are purpose-built for tensor operations like those used in large language models. Failing to account for these differences can result in high latency, excessive battery drain, device overheating, and poor user experience.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Marketing Agencies Are Deploying Autonomous AI Agents in Production in 2026

Marketing agencies in 2026 are moving beyond basic AI text generation toward autonomous agent systems that can plan, execute, and optimize multi-step campaigns across platforms like Meta, Google Ads, and TikTok. According to the article, roughly 34% of enterprises and 20% of mid-market teams already run at least one marketing agent in production, reporting 4–5x ROI on automated workflows and 27% faster campaign build times. These agents operate in continuous loops — observing data, taking actions via APIs, and iterating on results — handling tasks such as creative testing, brand knowledge management, performance reporting, and audience lifecycle orchestration. Frameworks like LangChain, CrewAI, and platforms such as Relevance AI and Gumloop have made it feasible for even solo developers to build and deploy such systems. Multimodal models including GPT-4o, Gemini, and Claude further expand agent capabilities by enabling analysis of screenshots, PDFs, and video content without requiring clean API access to every data source.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Static QR Codes Outlast Dynamic Ones for Printed Materials, Here Is Why

Most QR codes themselves do not expire, but the redirect services that power dynamic QR codes can be shut down, breaking any printed material that relies on them. Static QR codes embed the destination URL directly in the pattern, removing dependency on third-party providers and making them more reliable for long-term print use. Dynamic codes offer useful features like scan tracking and editable destinations, but these come with the risk of service discontinuation or subscription lapses. Experts recommend testing any QR code on both iPhone and Android devices, ensuring sufficient contrast, and exporting high-resolution files before committing to a large print run. For added flexibility without losing control, users can point a QR code to a URL on their own domain and manage redirects themselves.

0 comments Read more at DEV Community