Senior Site Reliability Engineer Interview Experience at Myntra – 2025

The first on-site round is dedicated entirely to Linux fundamentals, going well beyond command familiarity. Candidates are asked to explain what happens internally when a process is sent a SIGTERM versus a SIGKILL, and why a well-behaved application should handle SIGTERM to flush in-flight writes before exiting. A common follow-up scenario: a server's load average is high but CPU utilization looks normal - the expected reasoning walks through I/O wait as the likely cause, and how to confirm it using iostat. File descriptor limits and what happens to a process when it exhausts them under high connection load are also covered.

This coding round is conducted in Java, since Myntra's backend stack is heavily Java-based. The problem asks for designing and implementing a simple in-memory rate limiter (token bucket style) that can be applied per API key, including how to handle concurrent requests safely without a global lock becoming a bottleneck. The expected approach uses a ConcurrentHashMap keyed by API key with per-key synchronization rather than a single lock across all keys, and discusses the tradeoff of eventual consistency in a distributed version of the same rate limiter across multiple app servers.

Candidates are asked to design the infrastructure for handling Myntra's End of Reason Sale (EORS)-style traffic spike, where traffic can jump 10-15x within minutes. A strong answer structures around pre-scaling compute ahead of the known sale window rather than relying purely on reactive autoscaling, queueing checkout requests during the peak rather than letting them hit the database directly, and a circuit breaker on non-critical services (like recommendation widgets) so they degrade gracefully rather than taking down checkout. Interviewers typically push on how to test this capacity plan beforehand without risking a real production incident; a solid answer proposes a controlled load test against a scaled-down replica environment with production-like data volume.

This advanced troubleshooting round is built around a live-style scenario: a Java service's response latency is gradually climbing over several hours, memory usage looks normal, but GC pause times are increasing. The expected reasoning checks whether the old generation is filling up faster than expected (pointing to a memory leak or an object retention issue) versus a change in traffic pattern causing more allocation churn, and how a heap dump comparison across two points in time can confirm which it is. A common follow-up asks how to roll out a fix safely given the service is customer-facing during business hours.

This engineering manager round dives into a past project in real depth, asking candidates to justify specific technical decisions rather than just describe outcomes, plus a scenario-based question on how to handle a disagreement with a peer SRE about whether a given incident warranted a full postmortem.

This closing culture-fit conversation covers compensation expectations, notice period, and confirming willingness to be part of Myntra's on-call rotation given the role's operational nature.