150 questions from 25 interviews · updated from reports 2019-2025
About
Atlassian is a software company that makes tools for software development, project management, and team collaboration, including Jira, Confluence, Trello, and Bitbucket. In India, it is known for hiring software engineers and software engineering interns, with roles focused on product engineering and cloud platforms.
The roles that come up most are Software Engineer, Software Engineer Intern and Software Engineering Intern. This covers 25 candidate interviews reported from 2019 to 2025. The largest group sat it at internship level (11 of 24 that recorded a level). Among the 18 that recorded either route, arrivals split between campus drives (8, 44%) and off-campus applications (10, 56%). Most questions fall under DSA and Behavioural.
Ans. WhatsApp works as an end-to-end encrypted messaging system where clients keep long-lived connections to backend servers that authenticate users, route messages, and deliver notifications. The key detail is store-and-forward delivery: if a recipient is offline, encrypted messages are queued temporarily and delivered when their device reconnects, while servers cannot read the content.
Ans. The TLS handshake lets a client and server agree a secure session, authenticate the server, and derive shared encryption keys. The client sends supported versions, ciphers, and randomness; the server replies with choices and a certificate. They perform key exchange, usually ephemeral Diffie-Hellman, verify handshake messages, then switch to encrypted communication.
Ans. Use relational tables for users, groups, spaces, pages, page_versions, attachments, comments, labels and permissions. Pages belong to a space and have a parent_page_id for hierarchy, while page_versions stores immutable content revisions with author and timestamp. The key detail is separating the current page row from version history, permissions and large content for scale.
Ans. Google Search crawls the web, indexes pages, ranks matching documents for a query, and serves results from distributed systems in milliseconds. Crawlers discover URLs and fetch content, indexing extracts terms, links and signals, ranking combines relevance, authority, freshness, location and quality, then serving systems retrieve top candidates, personalise lightly, and return snippets.
Ans. HDFS is a distributed file system where a NameNode manages metadata and many DataNodes store actual file blocks. Files are split into large blocks, replicated across DataNodes for fault tolerance, and streamed to clients. Clients ask the NameNode for block locations, then read or write data directly to DataNodes.
Ans. Defend a website with layered controls: TLS, secure coding, input validation, strong authentication, least privilege, rate limiting, WAF or CDN protection, patching, logging and monitoring. The most important detail is to reduce attack surface and fail safely, because no single control stops SQL injection, XSS, credential stuffing and DDoS together.
Ans. Start by identifying the main entities, their key attributes, and primary keys from the problem statement. Then connect entities with relationships, marking cardinality such as one-to-one, one-to-many, or many-to-many, and participation as mandatory or optional. Resolve many-to-many relationships using junction entities and include foreign keys and important constraints.
Ans. Database normal forms are rules for structuring tables to reduce duplication and avoid update, insert and delete anomalies. 1NF makes values atomic, 2NF removes partial dependency on a composite key, and 3NF removes dependency on non-key columns. They improve consistency, though excessive normalisation can make queries more complex.
Ans. Choose a real work failure with clear consequences, not a personal flaw or blame story. Emphasise ownership, what you learned, and the specific changes you made afterwards. Interviewers listen for self-awareness, resilience, honesty, and evidence that your behaviour improved, such as better planning, communication, risk management, or asking for help earlier.
Ans. The relational model represents data as tables, rows, columns and relationships, while normalization is the process of organising those tables to reduce redundancy and avoid update, insert and delete anomalies. The key idea is to split data into well-designed relations using keys and dependencies, commonly up to third normal form.
Ans. A phonebook system should expose APIs to add, update, delete and search contacts, backed by a primary database and a search index. Store canonical contact records in a relational or key value store, partitioned by user or region. The key detail is indexing names and phone numbers for fast prefix and exact lookup.
Ans. Java HashMap is implemented as an array of buckets, where each bucket stores key-value entries based on the key’s hash code. The hash is spread and mapped to an index in the array. Collisions are handled by linked lists, or red-black trees when a bucket grows large. It resizes when the load factor threshold is exceeded.
Ans. Use a sliding window with two pointers to count valid subarrays in O(n) time and O(1) space. Keep multiplying the right element into the product, and while product is at least K, divide out the left element and move left. Add right minus left plus one for each right position.
Ans. Threading in Java lets multiple paths of execution run within one process, usually created with Thread, Runnable, Callable, or an ExecutorService. Concurrency is handled by scheduling threads and controlling shared state using synchronized blocks, locks, volatile, atomics, and concurrent collections. The key issue is visibility and race prevention under the Java Memory Model.
Ans. Build a tag service with APIs to create tags, attach them to entities, remove them, and query entities by tag. Store tags, entities, and a many-to-many taggings table with indexes on tag_id and entity_id. The most important detail is fast reverse lookup, so cache popular tag queries and update counts asynchronously.
Ans. A Red-Black Tree is a self-balancing binary search tree that keeps operations such as search, insertion and deletion in O(log n) time. Each node is red or black, the root is black, red nodes cannot have red children, and every path from a node to a leaf has the same number of black nodes.
Ans. Answer with the main design choice first, then give the key trade-off or bottleneck. For example, state the component, data store, scaling method, or consistency model immediately. If unsure, make one reasonable assumption and continue. Prioritise latency, availability, data volume, failure handling, and operational simplicity over exhaustive detail.
Ans. Pick a real conflict where the stakes mattered but you stayed professional. Emphasise how you listened to both sides, separated facts from assumptions, found shared goals, and agreed a practical next step. Interviewers listen for maturity, ownership, calm communication, and evidence that the team outcome improved rather than someone simply “won”.
Ans. Use a token bucket per client or API key, stored in a concurrent map from key to bucket state. Each request refills tokens based on elapsed time, caps at capacity, then consumes one token or rejects. Guard updates with per-key locking or atomic compare-and-swap. Lookup and update are O(1) per request.
Ans. Security vulnerabilities are usually found where untrusted input reaches a sensitive operation without proper controls. Look for SQL or command injection, XSS, path traversal, insecure deserialisation, broken authentication, hardcoded secrets, weak cryptography, and missing authorisation checks. The key is tracing input to sinks and checking validation, encoding, parameterisation, and access control.
Ans. XSS lets an attacker run malicious JavaScript in another user’s browser, while SQL injection lets an attacker alter database queries by supplying unsafe input. XSS is prevented with output encoding, sanitisation and CSP. SQL injection is prevented by parameterised queries, prepared statements and avoiding string-built SQL.
Ans. Common application security vulnerabilities are mitigated by defence in depth: validate input, encode output, use parameterised queries, enforce strong authentication and authorisation, apply least privilege, keep dependencies patched, and log security events. The most important detail is to treat all external input as untrusted and handle it safely at every boundary.
Ans. Join the four tables on their foreign keys, filter the rows needed, then group by the output columns and aggregate any measures. For example, use customers, orders, order_items, and products to return each customer’s total spend by product category. The key detail is using indexed join columns, giving roughly linear performance over matched rows.
Ans. I would not provide an XSS attack payload, but in authorised testing I would use a harmless test marker to check whether input is reflected or executed. The key detail is context: HTML, attribute, JavaScript, and URL contexts require different validation and output encoding to prevent script execution.
Ans. Scan from the right to find the first index where a[i] < a[i+1], swap it with the smallest larger element to its right, then reverse the suffix. If no such index exists, reverse the whole sequence to get the smallest permutation. This works in place in O(n) time and O(1) space.
Ans. Virtual memory is an abstraction that gives each process its own address space, which the operating system maps to actual RAM and sometimes disk. The key mechanism is paging: virtual addresses are split into page numbers and offsets, and page tables translate page numbers to physical frames, with the TLB caching recent translations.
Ans. Scan every cell, and when you find unvisited land, start a DFS and increment the island count. The DFS visits all connected land cells, usually using a stack or recursion and four directions for up, down, left and right. Mark cells as visited to avoid repeats. Time complexity is O(rows × columns).
Ans. A web or application system can have vulnerabilities such as injection, cross-site scripting, broken authentication, broken access control, insecure deserialisation, sensitive data exposure, security misconfiguration and vulnerable dependencies. The key issue is that untrusted input, weak identity checks or poor configuration can let attackers steal data, change behaviour or gain unauthorised access.
Ans. Use a tree of inode-like nodes where files store their byte size and directories store a cached total size of all descendant files. When creating, deleting, moving, or resizing a file, update the size delta up the parent chain to the root. Directory size lookup is O(1), while updates cost O(depth).
Ans. Represent the snake as a queue of body cells and a set for fast collision checks. On each move, compute the new head from the direction, reject wall hits, remove the tail unless food is eaten, then check self-collision and add the new head. Each move is O(1) time and O(length) space.
Ans. Choose a real disagreement about priorities, design, scope, or risk where you stayed respectful and outcome focused. Emphasise how you listened, used evidence, explained trade-offs, and accepted the final decision if needed. Interviewers listen for maturity, low ego, clear communication, and whether the relationship and project improved rather than suffered.
Ans. Indexes speed up reads by letting the database find rows without scanning the whole table, but they slow writes and use storage. Good schema design normalises data to reduce duplication, then denormalises only for proven performance needs. Efficient queries filter early, join on indexed keys, select only needed columns, and are checked with an execution plan.
Ans. Expose tags as resources and tag assignments as separate resources: GET/POST /tags, GET/PATCH/DELETE /tags/{id}, GET /tags?prefix=&limit=&cursor=, PUT/DELETE /resources/{type}/{id}/tags/{tagId}, and POST /tag-assignments:bulk for batch changes. The key detail is making writes idempotent and reads paginated, with indexed tag name and resource mapping tables for scale.
Ans. Rank pages using unique visits, repeat visits, dwell time, likes, comments, edits, backlinks, mentions, search clicks, and freshness. The most important detail is to weight these signals with time decay and normalise within a space or team, so old popular pages do not always dominate and niche pages are not unfairly buried.
Ans. A process is an independent running program with its own memory space, while a thread is a lighter execution path inside a process that shares that process’s memory. Threads are cheaper to create and communicate between, but need careful synchronisation. A deadlock happens when tasks wait forever for resources held by each other.
Ans. Pick a leadership example where your team’s work changed outcomes beyond your immediate group. Emphasise the business problem, who you influenced across functions, how you aligned priorities, and what changed because of your leadership. Interviewers listen for ownership, communication, measurable impact, handling of conflict, and whether others adopted or benefited from your work.
Ans. A monolithic kernel runs most OS services, such as drivers, file systems and networking, inside kernel space, while a microkernel keeps only core functions there and moves services to user space. Monolithic kernels are usually faster due to fewer context switches, but microkernels offer better isolation, fault tolerance and easier maintenance.
Ans. Choose a real decision where you had reasonable authority but others had valid concerns. Emphasise how you listened, tested assumptions, explained the rationale, and adjusted if needed. Interviewers listen for calm leadership, respect, evidence-based decision-making, and accountability, not stubbornness or blaming teammates for “non-compliance.”
Ans. Choose a real situation where values cost you something, such as time, comfort, or an easy win. Emphasise the specific value, the pressure you faced, the action you took, and the result. Interviewers listen for integrity, judgement, consistency under stress, and whether your behaviour matches the company’s culture.
Ans. Group anagrams by using a canonical key for each word, usually its letters sorted alphabetically, and store words with the same key together in a hash map. For each word, compute the key, append it to that key’s list, then return all map values. Time is O(n k log k), space is O(n k).
Ans. Implement it with a hash map from content id to score and a second structure ordered by score, such as a balanced tree of scores to sets of ids. On increase or decrease, remove the id from its old score bucket and insert it into the new one. Updates are O(log n), and max lookup is O(1) or O(log n).
Ans. Real-world operating systems use preemptive, priority-based scheduling, often as a multilevel feedback queue or fair-share variant. They need to respond quickly to interactive tasks while still giving CPU time to background jobs. This hybrid approach balances fairness, responsiveness, throughput and priority handling better than simple FCFS or round-robin.
Ans. Google push notifications work through Firebase Cloud Messaging, formerly GCM, where the app registers for a device token and the server sends messages to that token via Google’s servers. The device keeps a persistent connection to Google, so messages can be delivered efficiently. The enabling application is Google Play services.
Ans. CPU scheduling decides which ready process gets the CPU next, aiming for good responsiveness, throughput, and low waiting time. Round Robin gives each process a fixed time slice, so it is fair and good for interactive systems. Shortest Job First runs the shortest burst first, giving low average waiting time, but can starve long jobs.
Ans. I would keep tags global as canonical IDs, and localise only their display names, descriptions, synonyms and search matching. The most important detail is not to treat translated words as separate tags, because that fragments content, analytics, moderation and recommendations. Locale-specific tags can exist only where concepts genuinely differ by region.
Ans. Sort all arrival times and departure times separately, then scan them with two pointers, counting active trains. If the next arrival is before or at the next departure, one more platform is needed; otherwise one is freed. Track the maximum active count. This uses arrays only and runs in O(n log n) time.
Ans. Compare each list string with the initial string character by character and record characters until the first mismatch or one string ends. The result is that substring. This uses no extra data structure beyond the output and runs in O(total compared characters), bounded by the sum of min(length(initial), length(string)) over all strings.
Ans. Use a tree of directory and file nodes, with each directory storing its children in a hash map keyed by name. This matches the hierarchical path structure and gives fast lookup, insert and delete per path component. For reporting, maintain aggregate metadata such as size and file counts on nodes, updated on changes, so reports are efficient.
Ans. Use a postorder DFS and keep a global maximum for the best path seen anywhere. For each node, compute the best downward path starting there: node value plus the better of left or right downward gain, ignoring negative gains. Update the global answer with left gain plus node value plus right gain. Time is O(n), space is O(h).
Ans. Use a trie-like directory tree, where each node represents a folder or file and stores its children in a hash map plus metadata such as size, owner and timestamps. Path lookup, create, delete and move run in O(k), where k is path depth. Hash maps make child access near O(1).
Ans. Use a trie where each node represents a character, and store an end marker for complete words. To answer a query, traverse the prefix in O(p), then run DFS from that node to collect matching words. The key detail is result size dominates cost, so time is O(p + total output size).
Ans. The number of valid Binary Search Trees with nodes 1 to n is the nth Catalan number. Use dynamic programming where dp[i] stores the number of BSTs using i nodes, and for each root choice combine left and right subtree counts. The recurrence is dp[n] = sum(dp[left] * dp[right]). Time complexity is O(n²).
Ans. BFS explores level by level for shortest paths in unweighted graphs, DFS explores deeply for traversal, Dijkstra finds shortest paths with non-negative weights, Bidirectional BFS searches from both ends, and heuristic algorithms guide exploration or ranking. A* uses a heuristic plus path cost, while PageRank iteratively scores nodes by link importance.
Ans. Sort the arrival and departure times separately, then scan them with two pointers to find the maximum number of trains present at once. If the next arrival is before or equal to the next departure, add a platform; otherwise free one. The maximum count is the answer. Time complexity is O(n log n).
Ans. The answer is the size of the largest strongly connected component in the directed graph. Use Kosaraju’s or Tarjan’s algorithm with adjacency lists to find all strongly connected components, count their sizes, and take the maximum. This runs in O(n + m) time and uses O(n + m) space.
Ans. Find X with binary search over the non-negative long range, using a feasibility check for each candidate value. Sort the arrays, then greedily form required pairs, using two pointers if the constraint is ordered, or a balanced multiset if arbitrary matching is needed. The key is monotonicity. Time is typically O(n log n log R).
Ans. Use dynamic programming over the last character and its current run length. Let dp0[i][k] count length i strings ending with k zeros, where k < D, and dp1[i][k] similarly for ones, where k < T. Extend by same bit if allowed, or switch bit resetting run length to 1. Time is O(N(D+T)).
Ans. Use a monotonic increasing stack to compute the nearest smaller index on the left, and another pass to compute the nearest smaller index on the right. For each i, compare the two distances and choose the closer valid index. This runs in O(n) time and uses O(n) extra space.
Ans. Block the request if the IP matches an exact blocked address, falls within any blocked prefix, or the port is blocked; otherwise allow it. Store exact IPs and ports in hash sets, convert IPs to integers, and store prefixes in a binary trie or ranges. Lookup is O(1) for sets and O(32) for IPv4 prefixes.
No question matches those filters. Reset filters.
Showing 59 of 150 questions. Ranked by how often the same question came back across interviews.
A spoken interview built from these questions, scored when you finish; the feedback is yours.
Start practisingWhen you are ready, record The One: a single interview hiring teams watch, so you stop repeating first rounds.
Candidate interviews most often cover DSA (39%) and Behavioural (26%).
Candidate interviews show an average of 3.4 rounds per experience, with a typical sequence of Online test → Technical → HR. Individual interview paths can vary.
Among questions with a recorded difficulty, the mix is easy 23%, medium 64%, hard 14%.
