Manjeera Chanda

Talk Title:

Beyond Alerts: Predictive Failure Management and Retry-Safe Recovery in Distributed API Systems

Abstract:

Distributed API systems are inherently vulnerable to cascading failures, retry storms, and dependency-related disruptions that can significantly affect system availability and reliability. This work presents a resilience-oriented approach that moves beyond traditional alert-driven operations toward predictive failure management and retry-safe recovery. The approach emphasizes defensive architectural practices such as timeouts, bulkheading, and dependency mapping to contain failures before they propagate across interconnected services. Safe recovery is supported through idempotent operations, exponential back-off with jitter, controlled retry budgets, and circuit breakers. Graceful degradation mechanisms, including cached responses, fallback results, and functional redundancy, further help maintain service continuity during dependency failures. The framework also highlights chaos engineering and predictive scaling as proactive strategies for identifying vulnerabilities and preparing systems for changing operational conditions. Observability, simulated failure testing, and automated recovery are incorporated to improve failure detection and reduce recovery time. Overall, the proposed perspective promotes a transition from reactive failure handling toward resilient, proactive, and antifragile distributed API architectures capable of learning from failures and improving continuously.

Profile:

Manjeera Chanda is a Senior Integration Architect and practitioner-writer focused on the intersection of AI, APIs, and distributed-system reliability. With 10 years of experience across healthcare, fintech, and enterprise platforms—including GoodRx, Block, Symantec, and Salesforce—she has built integration systems that connect cloud services, ERPs, supply-chain partners, and customer platforms at scale. Her work explores practical guardrails for AI agent tool calls, reliable automation, and preventing costly duplicate actions in real-world business systems.