Troubleshoot Kafka consumer lag and rebalances - partition parallelism, max.poll.interval.ms, heartbeats, poison messages, offset commits and skew

Design idempotency keys - unique key table, request fingerprint, in-progress 409, replay stored response, TTL, concurrent-duplicate tests

JavaScript Promise patterns - all vs allSettled vs race vs any results, forEach async trap, timeout wrapper limits, concurrency pool of 3

Diagnose Kubernetes CrashLoopBackOff and OOMKilled pods - exit codes, previous logs, probes and limits

JavaScript gotchas run live - default sort as text, map(parseInt), == coercion, Date parsing UTC vs local, NaN includes, var closures

JavaScript regex cookbook - matchAll named groups, /g lastIndex statefulness, replaceAll TypeError, lookbehind, u flag emoji and unicode classes, ReDoS

JavaScript numbers lose precision - 0.1+0.2, MAX_SAFE_INTEGER, JSON.parse rounding big IDs, BigInt, toFixed and money handling

Handle SIGTERM and SIGINT gracefully in Python and Node - handlers, exit 143/130/137, KeyboardInterrupt, orphaned children, exec in scripts

Diagnose high load on a Linux server - CPU, memory and swap, disk iowait, pressure stall info and the runaway process

Add authentication to a remote MCP server - API keys vs OAuth 2.1, audience and scope checks, per-tool authorization, Origin checks and audit logs

Fix LLM streaming SSE parsing errors - buffer split chunks, handle [DONE], accumulate tool-call deltas and turn off proxy buffering

Design LLM provider fallback and routing - error classification, circuit breakers, equivalent models, request adapters and failover testing

Fix LLM JSON and structured output failures - schema-enforced output, truncation checks, validation and repair retries

Cut LLM API cost with prompt caching - stable-prefix prompt layout, cache-hit measurement, context trimming, model routing and batch jobs

Fix address already in use errno 98 - find the owner with ss -ltnp or lsof, kill safely, TIME_WAIT and SO_REUSEADDR, bind port 0 in tests

Debug Kubernetes ingress 404 and 503 errors - ingress class and rules, Service selectors and ports, empty endpoints, probes and backend protocol

Linux disk full - df -h vs df -i inodes, du -x sorted, deleted files held open by a process (lsof +L1), truncating logs, ext4 reserved blocks

Reduce LLM hallucinations - grounded prompts, abstention, quote-verified citations, tool-backed facts and independent verification passes

Build a small LLM evaluation dataset - real cases, edge and refusal tests, code and model graders, holdout split and per-change score tracking

Fix LLM context window overflow in long agent runs - token budgets, protected state and safe summarization