# High-Score Atlassian Principal Engineer Interview: LRU Cache, Web-Scraping & Org-Tree LCA # High-Score Atlassian Principal Engineer Interview — What Happened and What to Learn Interview cover A Bugfree user shared a detailed, 6-round remote loop for an Atlassian Principal Engineer role. Below is a concise recap of each round, the technical expectations, recommended approaches, and practical tips you can use while preparing. ## Quick summary - Format: 6 rounds (phone screen + 5 onsite-style rounds) - Phone screen: LRU cache design + quick designs (Google Docs conflict handling/WebSockets; file storage + search) - Deep dives: Values + Leadership Craft - System design: Web-scraping pipeline with APIs for submitting jobs, checking status, and retrieving results - Code design: Cinema scheduling — place a new movie into a fixed-day schedule without removing existing shows - Data structures: Employee org-tree — find the closest common parent (LCA) - Outcome: Candidate was down-leveled after one weak round and ultimately declined --- ## 1) Phone screen: LRU cache + quick designs Phone screen tested fundamentals and quick architecture intuition. LRU cache: expected approach - Typical solution: hash map + doubly linked list. - Hash map provides O(1) access by key. - Doubly linked list maintains recency order; head = most recent, tail = least recent. - On read/write, move node to head; if over capacity, evict tail. - Discuss concurrency: coarse-grained lock, segmented locks, or lock-free approaches depending on requirements. - Edge cases: zero capacity, duplicate keys, TTL vs pure LRU, metrics and eviction policy variants (LFU, size-based). Quick design prompts (good to mention these points briefly): - Google Docs conflict handling / real-time edits - High-level choices: CRDTs vs Operational Transform (OT). - Use WebSockets (or WebRTC) for low-latency updates and presence; fall back to polling for unreliable clients. - Versioning, causal ordering, and merge strategies are key. Consider offline edits and reconciliation. - File storage + search (object store + Elasticsearch) - Store blobs in an object store (S3) and index metadata/content in Elasticsearch. - Ingestion pipeline: pre-processing, text extraction, indexing, and refresh strategy. - Consider consistency (eventual), reindexing, ACLs, and query patterns. Tip: For phone screens, be concise. Explain the approach, trade-offs, and call out edge cases quickly. --- ## 2) Values + Leadership Craft These rounds probe culture fit, ownership, and leadership judgment. - Expect behavioral questions and deep dives into past decisions, trade-offs, and ambiguous situations. - Use structured storytelling (context, action, result). Emphasize impact, trade-offs, and what you learned. - For senior roles, highlight stakeholder management, hiring/growth decisions, and system-level thinking. Tip: Prepare 4–6 strong stories across technical leadership, delivery under risk, program management, and mentoring. --- ## 3) System design: Web-scraping pipeline (POST /jobs, status, results) Design goal: robust, scalable pipeline that accepts scraping jobs, reports status, and returns results. Core API surface - POST /jobs: submit job (URLs, frequency/schedule, parsing rules, credentials, max depth) - GET /jobs/{id}/status: return job state (queued, running, failed, completed), progress, and logs - GET /jobs/{id}/results: return structured results (JSON, links to raw HTML stored in object store) Architecture components - API layer: validates requests, authenticates users, enqueues jobs - Job queue: durable work queue (Kafka, RabbitMQ, SQS) - Scheduler & workers: - Workers perform fetches (HTTP client or headless browser like Puppeteer for JS-heavy sites) - Respect robots.txt, rate limits, per-host backoff, proxy pools and retry policies - Parsing/Extraction: configurable scrapers/parsers; consider schema extraction and robustness to layout changes - Storage: - Raw payloads to an object store (S3). - Structured results indexed in a DB or search engine if queries are required. - Monitoring & Observability: metrics (success/failure), per-site health, error classification - Security & Legal: credential storage encryption, access controls, and compliance with scraping policies Scaling and reliability - Horizontal scale by adding workers; use partitioning to avoid double-scraping the same host. - Deduping URLs and results to avoid wasted work. - Circuit breaker per-host to avoid hammering fragile servers. Tip: When designing, ask about SLAs, expected throughput, whether pages are JS-heavy, and how results are consumed. --- ## 4) Code design: Cinema scheduling Problem sketch: given a fixed-day schedule of shows in one or more theaters, fit a new movie into the day without removing existing shows. Interpretation and assumptions to clarify - Are there multiple auditoriums? Are show durations equal? Is there a required buffer (cleaning/turnover) between shows? - Is the goal to place a single new show (one screening), or multiple showtimes for a new movie? - Is optimization required (maximize number of shows, maximize utilization, or first-fit)? Simple approach (if single screening): - Model existing shows as closed intervals [start, end]. - Compute free intervals by scanning sorted shows and finding gaps between end_i and start_{i+1}. - If new movie duration + buffer fits into any gap, schedule into earliest gap (first-fit) or highest-utilization gap depending on objective. Multiple screenings / optimization: - This becomes an interval-packing / scheduling problem. Options: - Greedy heuristics: earliest-fit across auditoriums, or sort gaps by size and fit biggest items first. - Dynamic programming or backtracking if the search space is small and exact optimality is required. Tip: Clarify constraints first. Interviewers want to see clear assumptions, edge-case handling, and time/space complexity reasoning. --- ## 5) Data structures: Employee org-tree — closest common parent (LCA) Classic problem: given two employees in a tree representing org structure, find their lowest common ancestor (closest common manager). Common solutions - Depth + parent pointers - Bring both nodes to the same depth (move the deeper one up), then move both up until equal. Time: O(h) where h is tree height. - Binary lifting (preprocess ancestors up to log N) - Preprocess in O(N log N), query in O(log N). Good for many LCA queries. - Euler tour + RMQ - Preprocess in O(N)–O(N log N) and answer queries in O(1) with RMQ; good for static trees and many queries. Tip: Start with the simple approach (depth+parents) and discuss binary lifting if interviewer cares about multiple queries or complexity at scale. --- ## Outcome and learning The candidate was down-leveled because one round was weak; they declined the offer. Key takeaways - A single weak round can change the overall outcome for senior roles. Prepare across all areas: systems, coding, and leadership. - Clarify assumptions early. Interviewers evaluate both technical correctness and communication. - For system design, be explicit about scale, trade-offs, and failure modes. For coding/design rounds, show test cases and edge-case thinking. Practical preparation checklist - Practice LRU and common data structure patterns (hashmap + list, sliding windows, union-find). - Review LCA algorithms and be ready to explain trade-offs between simple and optimized solutions. - Build a few end-to-end system designs (ingestion pipelines, auth, search index) and practice sketching components and data flow. - Prepare leadership stories using structured formats and quantify impact. --- If you want, I can: - Expand any of the above sections into a full mock whiteboard script you can rehearse. - Provide sample code for LRU cache or a binary-lifting LCA implementation. Good luck with your prep — focus on clear assumptions, trade-offs, and calm communication.