Skip to content
Tech Interview Prep
Technical interview guide

Layer 2 Scaling & Rollups

How rollups and other Layer 2 systems scale transaction throughput while still inheriting the base chain's security guarantees.

Read
47 min
Practice MCQs
25
Interview Q&A
25
Edition
v3
Editorial status
Reviewed

Scope: Ethereum rollup architecture and EIP-4844; OP Stack derivation, deposits, withdrawals, fault proofs and interoperability specifications reviewed 2026-09-04; Arbitrum Nitro architecture (protocol-specific parameters must be verified at deployment).

Curated: · Written: · Reviewed:

25 questions

MCQ-3A rollup posts one 32-byte root per batch to L1. A bridge contract releases funds when a withdrawal proof verifies against the root of the batch that included the withdrawal. What does that root commit to?
MCQ-4A rollup posts a state root to its L1 inbox contract. The contract records the root as pending and only finalizes it after a fixed window, during which the proposer's bond can be slashed. Why is the assertion provisional?
MCQ-6A team ships on an optimistic rollup. Deposits from L1 are credited on L2 within minutes, but withdrawals back to L1 take days. Which mechanism accounts for the gap?
MCQ-11A bridge service credits deposits only after an L2 block is irreversible under the rollup's own confirmation rules. Its rollup node exposes three heads: unsafe, safe, and finalized. The bridge must pick the head whose blocks it will treat as having the weakest confirmation. Which head is that?
MCQ-13A rollup team is deciding which failure modes its L1 contracts must cover. They are weighing a forced-transaction path: an L1 inbox that can push a transaction into L2 without the sequencer's cooperation. Which problem does that path exist to address?
MCQ-16A rollup's settlement contract on L1 finalizes withdrawals by verifying a Merkle proof that the withdrawal was included in an L2 output root. The contract stores `finalizedWithdrawals[withdrawalHash]` and sets it to `true` before transferring tokens to the recipient. An attacker watches the mempool, sees a legitimate withdrawal transaction, and submits the identical proof and calldata again in the next block. The second transaction reverts. Which mechanism caused the revert?
MCQ-7A validity rollup's L1 verifier contract holds verifying key VK for circuit C. A batch posts public inputs oldRoot=R0, newRoot=R1, dataCommitment=H and a 1.2 KB proof. The verifier call costs 280k gas, returns true, and the contract stores R1. An off-chain watcher re-executes the batch data that hashes to H from R0 and gets R2, not R1; the proof still verifies under VK with public inputs (R0, R1, H). What is the most likely explanation?
MCQ-5A rollup settles state-root disputes by interactive bisection on L1, and the settlement call must fit inside one block's gas. A challenger disputes a batch covering 2^26 = 67,108,864 VM steps. Both parties respond inside every clock and both post the same 1 ETH bond. In round 26 the disputed interval is one VM instruction; the two parties now submit identical pre-states (pc, registers, memory root) but different post-state hashes. What does the contract do?
MCQ-19An L1→L2 bridge mints 100 wrapped ETH on L2 to the depositor address encoded in a DepositFinalized message from L1. The relayer submits message id 0xabc. The L2 RPC returns a timeout after the transaction is included, so the relayer resubmits the identical message. The handler checks only that the L1 block header is valid and that the message signature is from the relayer set; it does not record message ids. What happens on the second execution?
MCQ-18A rollup's L2 escrow credits balances when the canonical L1→L2 messenger delivers a message (Solidity 0.8): ```solidity function onMessage(bytes calldata data) external { require(msg.sender == MESSENGER, "not messenger"); (address user, uint256 amount) = abi.decode(data, (address, uint256)); credit[user] += amount; } ``` `MESSENGER` is the rollup's canonical messenger, which relays any message an L1 contract pays to send. A day after launch, an attacker with no L1 deposit deploys a contract on L1 that calls the messenger with `abi.encode(attacker, 1_000_000e18)`, and the L2 escrow credits the attacker 1,000,000 tokens. Which change closes the hole?
MCQ-1Two L2s each hold user ETH in a bridge contract on Ethereum L1 and each runs a single sequencer. Rollup-R posts every batch's state root together with a validity proof that the L1 bridge contract verifies before accepting the root. Chain-S posts a state root every 30 minutes, and its L1 bridge contract accepts a root once 5 of its 8 registered validators have signed it. Both sequencers halt permanently at block 4,000,000; the last root accepted on L1 for each chain is from block 3,999,900, and a given user's balance is included in that root on both chains. Which statement about that user withdrawing their bridged ETH is correct?
MCQ-8A rollup posts each batch's transaction payload as EIP-4844 blobs, archives nothing else, and calls Ethereum's consensus layer its data-availability layer. Consensus clients serve blob sidecars for MIN_EPOCHS_FOR_BLOB_SIDECARS_REQUESTS = 4096 epochs (32 slots per epoch, 12 s per slot, about 18.2 days), then prune them. On day 40 the team finds that a fresh verifier node cannot replay the chain from genesis: its node fetches nothing for the old batches. Which statement identifies the design defect?
MCQ-9A rollup posts its batch data as EIP-4844 blobs. A user submits a fraud proof 20 days after the batch was posted, and the rollup's verifier needs the original blob data to re-execute the batch. The verifier node has been running continuously since the batch was posted but has no separate archive of blob data. What happens?
MCQ-20A team ships an appchain rollup that posts every batch to Ethereum as calldata. During a mainnet congestion spike the sequencer's L2 execution gas price is unchanged, yet users report their total transaction fees roughly tripled. The team's dashboard shows the batch's compressed calldata size and the L2 gas used are both flat versus the prior week. What most plausibly explains the fee increase?
MCQ-21A team deploys RateVest to an OP Stack rollup that produces one L2 block every 2 seconds. The bytecode is identical to its mainnet deployment, same solc settings, and the contract releases 0.5 tokens per elapsed block.number step - a rate sized so 648,000 tokens vest over 180 days against Ethereum's 7,200 blocks a day. Deployment and the first release calls succeed, the mainnet-fork tests pass, but the contract's token balance reaches zero after about 30 days, with releases spread evenly over that window. Which mechanism explains the early drain?
MCQ-15A user sends 5 ETH through the canonical L1 portal at L1 block 21,000,000. The sequencer then goes down for three hours; when it restarts it keeps building blocks from its own mempool and never includes the deposit transaction. The user's L2 balance is still 0. What happens to the 5 ETH?
MCQ-14A rollup runs a single centralized sequencer that posts batches to Ethereum. During a 40-minute sequencer outage, users' transactions sit unbatched, and once it recovers the operator orders a large DEX swap ahead of the pending user trades at the same block. The rollup's fraud proofs still work and no invalid state root was ever accepted. Which statement best characterizes what this incident demonstrates about the sequencer's power?
MCQ-10A rollup node has derived L2 through block 5,000, whose L1 origin is block 1,200. Its L1 provider then reports that block 1,200 was reorged out; the canonical chain now has block 1,200' at the same height with a different deposit log. The node's derivation pipeline is deterministic. What must it do before deriving L2 block 5,001?
MCQ-12An L2 rollup node derives its safe head by scanning canonical L1. L1 reorgs out the block that the node's current safe head was derived from; the replacement block at that height is already canonical and carries a different deposit. The unsafe head is unaffected. What should the node do?
MCQ-2An exchange credits a user's deposit on an OP Stack rollup as soon as the sequencer's RPC returns a receipt (median 2 s). During a sequencer failover the rollup reorged 12 blocks and dropped 340 deposits the exchange had already credited, forcing a $1.2M clawback. Batches kept posting to L1 throughout and L1 itself did not reorg. Which change removes the exposure?
MCQ-23Rollup A produces 2-second blocks and posts a batch to L1 every 12 minutes; a posted batch reaches L1 finality about 13 minutes later. A low-latency bridge executes a message on rollup B as soon as the source block appears in a posted batch, so the destination acts roughly 13 minutes before the source block is final. After a failed batch submission, rollup A's sequencer reorgs 3 blocks and drops the block that carried the message. Rollup B has already minted 50 ETH of wrapped A-assets to the recipient. What must the bridge do about that 50 ETH mint?
MCQ-17A treasury team must move 500,000 USDC from Optimism to Ethereum mainnet to meet a payment deadline 30 minutes away. A canonical Optimism withdrawal would not be claimable on L1 for about 7 days. The team instead uses a fast bridge that quotes a 6 bps fee and delivers USDC on mainnet in about 2 minutes: the bridge's liquidity provider fronts the mainnet USDC and is repaid later from the L2 escrow. The USDC lands on mainnet in 2 minutes. Which statement describes the state of this transfer?
MCQ-22A rollup posts state roots to Ethereum L1 and runs a live fraud-proof contract. Its bridge proxy admin is a 3-of-5 multisig with no timelock, and the verifier implementation slot is upgradeable by that admin. TVL is $600M. An auditor must name the system's top trust assumption. Which conclusion follows from these facts?
MCQ-24An OP-Stack rollup produces 2-second L2 blocks and posts one batch per 12-second L1 block, each batch carrying the six L2 blocks from that window. A recoverability drill starts from this state: the batch inbox's last batch, already in a finalized L1 block, covers L2 blocks 1,204,845-1,204,850; five further L1 blocks pass with no rollup batch or deposit transaction; the sequencer is then killed at L2 block 1,204,880, and its host plus both replica hosts are wiped, databases included. An empty node boots with an L1 archive only, no sequencer and no peer reachable. What L2 head does it reach?
MCQ-25An L2 rollup produces blocks every 2 s. Its batcher posts one L1 batch per 300 L2 blocks, and its proposer submits an output root whenever new batches land. Overnight, the L2 head advances steadily, block hashes agree across two independent nodes, and RPC latency is flat, but the newest batch in the L1 inbox covers L2 block 41,200 while the L2 head is 59,200; the newest output root covers L2 block 41,150. Which diagnosis does the evidence support?