Interview guides · Advanced Technology Search Group

Advanced Technology Search Group Software Engineer Interview Questions & Guide 2026

Software engineer interview prep for Advanced Technology Search Group: original coding, system design and debugging practice, each with a worked solution.

The 60-second brief

A Software Engineer at Advanced Technology Search Group sits at the intersection of high-performance computing, specialized hardware design, and scalable infrastructure. This role is critical to the company’s mission of delivering precision engineering solutions, whether through advanced IC design, GPU-accelerated systems, or robust cloud-based EDA (Electronic Design Automation) environments. You will be tasked with solving complex, low-level technical challenges that underpin the next generation of hardware and software integration.

If the seat owns a service boundary, scope your preparation toward failure behaviour rather than topology. Retrying over an at-least-once channel produces duplicates by construction, so a retry policy is only as safe as the idempotency key underneath it.

Focus areas

Bound blast radius with per-tenant concurrency limits
Evolve APIs without breaking pinned SDK clients
Build at-least-once pipelines with explicit deduplication horizons

Overview

The impact of this position is far-reaching, directly influencing how the company supports its global client base in demanding sectors like graphics, high-speed analog, and cloud infrastructure. You will work within highly specialized teams to build tools and systems that require both deep domain expertise and a disciplined approach to software development. If you are passionate about pushing the boundaries of what is possible in hardware-software co-design, this role offers a challenging and intellectually stimulating environment.

The Advanced Technology Search Group interview process

Advanced Technology Search Group candidates report 2 rounds · ≈ 2-4 weeks. The stages below are what candidates describe, not a published process.

Initial Screening

Before anything technical happens, someone has to decide which rung of the ladder your loop is calibrated to, and that decision sets the bar for every round after it. It comes from how you describe scope, not from your title, because titles do not convert cleanly between companies. The weak version of the answer is team size and years. The strong version names the largest change you shipped where nobody reviewed the design, what would have broken if you had been wrong, and what you were paged for. Get the level said out loud on this call, because the range and the loop both follow from it.

What to demonstrate

Whether the scope in your own account maps onto a level the team actually has an opening at, so a mismatch ends the process cheaply rather than after four interviewers have spent a day
Whether your title needs re-mapping: the same word describes very different amounts of independent decision-making at a twenty-person company and a ten-thousand-person one
Whether your compensation expectation can be filled at that level in the structure the role pays in, which is why the number gets asked for before any engineer is scheduled

How to prepare

Write down two changes from the last two years: the largest one you designed with nobody reviewing the design, and the largest one where someone more senior did. Lead with the first when scope comes up, and be ready to say which parts of the second were yours
Ask which level the loop is calibrated to and what changes at the level above it, then plan your weeks from that answer rather than from the posting
Settle a total-compensation range beforehand with the split named, base against bonus against equity and its vesting period, so a question about numbers gets a number instead of the word market

Engagement with Hiring Managers

This round is a resourcing decision in the shape of a conversation: how large a piece of work can be handed to you with a one-line brief and no check-in for two weeks. The manager is listening for the seams in your story, the places where you settled something yourself and the places you went back for a ruling. Most candidates describe the system and skip the decisions, which reads as having been present rather than responsible. Say who wanted the work, which option you rejected and why, and where you would have stopped and escalated.

What to demonstrate

Whether you can point at a design decision that was yours rather than the team's, and name the alternative you turned down and the constraint that killed it
What you treat as yours to settle against what you take to someone else, and how long you sit on a blocker before raising it
Whether the scope you claim survives a follow-up into the unglamorous part of it: the data migration, the backfill, the rollout to users who were already on the old path
Whether you asked what problem the request was solving before building what was literally asked for, and what changed in the design once you had the answer

How to prepare

Write out the brief for your last two projects exactly as it reached you, usually one sentence in a ticket, then list every question you had to answer yourself before code could be written. That list is most of what this round is asking for.
For one decision in each project, write down the rejected option, what you were trading against, and the piece of evidence that would have flipped you. A tradeoff you cannot argue in reverse was not really a decision.
Have an escalation ready: a blocker you took to your manager, what you had already tried, and the specific thing you were asking them to decide. If every example is something you handled alone, that reads as someone who does not ask.

Where candidates lose points

One shared connection pool for every tenant and every query class

A single tenant with a large table and a missing index can occupy every connection with slow queries, and every other tenant then waits in connection acquisition -- a queue invisible in database metrics, because the database itself looks healthy while the application starves. Containment is bulkheads: separate pools or per-tenant concurrency caps for interactive requests, background jobs and exports, a statement timeout low enough that a pathological query dies before it accumulates, and an idle-in-transaction timeout so a stuck client cannot pin a connection and its locks indefinitely. One caveat worth knowing in advance: if a transaction-pooling proxy sits in front of the database, session-scoped behaviour changes, so session-level advisory locks and settings applied outside a transaction do not survive the way they do on a direct connection.

Serialising a tenant's writes through select ... for update on a single counter row

It is the first change that makes a counter correct, and it caps that tenant's write throughput at roughly one divided by the lock hold time. A transaction that takes the lock, makes a network call and then commits holds it for the entire round trip: at 2 ms that is about 500 writes per second for the whole tenant, and the largest tenants are exactly the ones that exceed it. The damage then spreads, because every waiter holds a database connection while it queues, so one hot tenant drains the shared pool and the symptom presents as a site-wide latency incident rather than as a lock problem. The repairs are to shrink the critical section to a single statement, to shard the counter into per-(tenant, hour) or per-(tenant, bucket) rows and sum on read, or to batch in memory and flush periodically while accepting the bounded loss that batching implies.

Optimising an axis nobody named

Ask which resource is actually scarce here: wall-clock latency, throughput, memory footprint, cost per request, or engineering time. Shaving a constant factor off an in-memory step is wasted effort when the same function makes a blocking remote call inside the loop.

Arguing past a hint

When the interviewer asks what happens for a particular input or floats a different data structure, stop and take it seriously; it is almost always a correction rather than idle curiosity. Talking over it converts a recoverable wrong turn into a data point about how you handle review.

Questions & practice

Hold a tenant to a trailing sixty-second request limit

Medium · sliding-window · two-pointer · rate-limiting · tenant-skew

The gateway must hold each tenant to R requests in any trailing 60 seconds, in aggregate across three regions and every pod, within a budget of under 10 ms added p99. Peak is 30,000 requests/second across 200,000 active tenants, and traffic is heavily skewed toward a handful of them. Give an exact single-process algorithm with its amortised per-request cost and its memory per tenant, then a bounded-memory approximation and the worst-case overshoot it actually admits. Say what the distributed version does when the counter store is unreachable.

Approach

Exact, single process: a per-tenant deque of request timestamps. On arrival, pop from the front while front <= now - 60s, then admit if the remaining length is below R and push. Each timestamp is pushed once and popped once, so the cost is O(1) amortised. The O(R) version is the one that re-filters the whole deque on every request.
Quote the memory. R = 1,000 across 200,000 active tenants is up to 2 x 10^8 timestamps at 8 bytes, about 1.6 GB, and that is the worst case rather than the mean, because the long tail of small tenants holds almost nothing. Skew helps you here and hurts you in the sharding decision.
Bounded alternative, with its real bound stated: a fixed 60-second counter is O(1) memory but admits close to 2R across a 60-second span straddling a boundary. The weighted two-bucket estimate, prev * (60 - elapsed)/60 + cur, is better on smooth traffic but assumes the previous window's arrivals were uniform; an adversary packing them at the end of that window is undercounted and can still approach 2R. Say that rather than calling it exact.
Token bucket is the usual gateway answer and a different contract: O(1) state per tenant (tokens, last_refill), a sustained rate, and a deliberate burst allowance equal to the bucket size. Choose it when a burst is acceptable and the log when the limit is contractual.
Distributed: the limit is per tenant in aggregate, so a local bucket of R/N per pod is wrong in both directions under skew. A tenant landing on one pod is throttled at R/N, and a tenant spread evenly across pods exceeds R. The shared check must be a single atomic round trip, one script or one increment-and-compare, never read-then-write, and it must fit inside the 10 ms p99 budget.
Decide the unavailable case in advance and write it down. Failing open keeps the product up and lets a tenant exceed its limit for the duration; failing closed converts a counter-store outage into a full outage. Most gateways fail open on rate limits and closed on authorisation, and those are two separate decisions made separately.

Follow-up

One tenant sends 40% of all traffic. What does that do to a single counter key, and what do you shard on instead?
Quotas rather than rate limits: the check is select used; if used < limit then insert. Name the isolation level that still permits the overshoot, and the two fixes.
How do you return an accurate Retry-After from the exact algorithm without a second scan?

Find peak concurrent sandbox usage from run intervals

Medium · sweep-line · intervals · concurrency-caps · null-semantics

Given up to 5 million job_run rows for one tenant over one day, with run_id, started_at, finished_at, status and wall_clock_limit_seconds, report the maximum number of sandboxes running at once, the earliest instant that maximum is reached, and the first run_id that would breach a per-tenant cap of C. started_at is null while a run is queued; finished_at is null both for runs still executing and for runs in status lost. Treat a run as occupying [started_at, finished_at). Give the complexity and state how you handle each null.

Approach

Turn each run into two sweep events, (started_at, +1) and (end, -1), then sort the 2n events by timestamp with -1 ordered before +1 at equal timestamps. That tie-break is what makes the interval half-open, so a run finishing at 10:00:00 and one starting at 10:00:00 never overlap.
Decide each null out loud before sweeping, because each choice moves the answer. A null started_at means queued and contributes nothing. A null finished_at with status running or leased is clipped to the window end. Status lost has no observed end at all, so clip it at started_at + wall_clock_limit_seconds on the grounds that the supervisor owns the timeout, and record that you did. The table's check (finished_at is null or started_at is not null) guarantees you never see an end without a start.
Sweep once, maintaining a running counter, the maximum, and the timestamp at which the maximum was first attained (update peak_at only on a strict increase, or you will report the last such instant instead of the earliest). Capture the first run_id whose +1 takes the counter to C+1 during the same sweep rather than in a second pass.
Complexity: O(n log n) dominated by the sort, O(n) space. If rows already arrive ordered by started_at, a min-heap of end times gives O(n log k) time and O(k) space with k the peak concurrency, which is the better shape when the rows come from an index scan on (tenant_id, started_at).
If second resolution is acceptable, counting-sort the endpoints into an 86,400-slot delta array and prefix-sum it: O(n + T) time and O(T) space, which beats the comparison sort at 5 million rows. It answers only at second granularity, so state which resolution the cap is defined in.

Follow-up

Now report peak concurrency per tenant for 10,000 tenants from one globally sorted stream. What changes about memory and about the sort?
The cap has to be enforced at dispatch rather than reported afterwards. What does the admission check look like, and where does it race?
How would you answer 'peak concurrency within any 5-minute window' without re-sorting?

Locate a billing reconciliation gap without rescanning ninety million events

Hard · reconciliation · dimensional-bisection · watermarks · hypothesis-testing

A tenant's sealed invoice total is 0.4% below the sum of its raw usage_event rows for the period. That tenant has 90 million events over 30 days in a table partitioned daily on ingested_at, and its rollups carry source_max_ingested_at, revision and sealed_at. Recomputing all 30 days from raw is correct, and you are not going to do it. Give the procedure that locates the divergent (workspace, sku, hour) cell, the cost of each probe, and the one query you run before any of it.

Approach

Run the free query first. Sum raw quantity for the period restricted to ingested_at <= source_max_ingested_at of the sealed rollups, and compare that against the unrestricted sum. The rollup stores the watermark precisely so this can be answered without a scan. If the whole 0.4% sits above the watermark, nothing is broken: it is late data, it becomes an adjustment line, and the investigation ends in one query.
Only if the gap survives that test do you bisect, and you bisect by dimension rather than by rows. Compare 30 per-day totals, then inside the offending day compare the 6 SKUs, then the workspaces, then the 24 hours. That is roughly 30 + 6 + W + 24 grouped probes, each an indexed range scan over one daily partition for one tenant, against O(N) per attempt for the naive re-fold.
Quantify why naive is not merely slow but unusable mid-incident: at a generous 200,000 rows/second sequential, 90 million rows is about 7.5 minutes per attempt, you will want ten attempts, and every one competes for I/O on the same partitions live ingest is writing. The diagnostic worsens the backlog it is diagnosing.
Before fetching each comparison, state what it would look like under each hypothesis. Two adjacent hours off by equal and opposite amounts is occurred_at versus ingested_at bucketing. A whole day offset by exactly N hours is a timezone applied at the wrong layer. A gap confined to one SKU in one workspace is an environment filter. The same (tenant_id, idempotency_key) present in two ingested_day partitions is the dedup horizon losing a retry that crossed midnight.
Make the next bisection cheap by storing the aggregate you keep recomputing. A per-(tenant_id, ingested_day) count and quantity checksum turns step two from thirty probes into one read, and it is the same number the reconciliation job already produces.
Whatever you find, the sealed period does not change value. The correction is an adjustment line pointing at the line it reverses, carrying its own source_rollup_watermark, because the original invoice is the evidence of what the customer was charged.

Worked solution (35 min)

Reproduce the shape locally: generate 2 million events for one synthetic tenant over 5 days, fold them, then inject three defects, namely 0.2% of events bucketed by ingested_at, a handful of duplicate idempotency_key values whose retries cross midnight, and a block of events ingested after the seal.
Write the watermark-bounded query first and record how much of the gap it explains on its own.
Bisect by day, then SKU, then hour, recording the probe count and the rows each probe touches.
For each located cell, write down the predicted signature before querying it, then check whether the data matches the prediction.
Time a full re-fold of the 2 million rows and extrapolate to 90 million.

Expected result: The watermark-bounded query accounts for the post-seal block entirely and removes it from the investigation. The ingested_at bucketing appears as adjacent hours off by equal and opposite amounts. The midnight-crossing duplicates appear as one (tenant_id, idempotency_key) present in two ingested_day partitions. Bisection reaches each cell in fewer than 70 probes.

Verify it yourself

The three injected defects sum to the original gap with no unexplained remainder.
Each probe's row count is bounded by one day-partition for one tenant, not by the period.
Removing the source_max_ingested_at bound from the first query makes a gap appear, and grow, while ingest runs, even after every defect is removed.

Follow-up

The gap is 0.4% in one direction on one day and 0.4% the other way the next day. What does that shape rule in, and what does it rule out?
How do you distinguish a duplicate from a restatement, given revision and recomputed_at on the rollup?
Ingest is still running while you investigate. What makes your two numbers comparable at all?

Migrate a live partitioned event table without blocking ingest

Hard · online migration · ddl locks · expand and contract · partitioning

usage_event is range-partitioned daily on ingested_at, holds roughly 250M rows per day across 400 live partitions, and is written at 10-40k rows/second. Two changes are required: quantity must move from double precision to numeric(20,6), and a new environment column must become NOT NULL with a default of 'production'. Ingest cannot stop. Give the ordered plan, naming for each step the lock it takes, what that lock blocks, and roughly how long it is held. Identify the one step that cannot be rolled back cleanly once traffic depends on it.

Approach

Classify the two changes before planning anything. Adding a column with a non-volatile default has been metadata-only since PostgreSQL 11, so it is cheap. Changing double precision to numeric is not binary-coercible, so alter column ... type rewrites every partition under ACCESS EXCLUSIVE and rebuilds its indexes; on this volume that is hours of blocked ingest and is simply not an option, which is why the plan is expand-and-contract rather than one statement.
Expand: add quantity_numeric numeric(20,6) and environment with its default on the parent. Both are catalogue-only but both take a brief ACCESS EXCLUSIVE that cascades to partitions, so run each with lock_timeout set to a second or two and retry on failure. A queued ACCESS EXCLUSIVE request blocks every reader behind it, which is how a metadata-only change turns into an outage.
Dual-write: deploy producer code that populates both columns on every insert, and leave it running before anything reads the new column. This is the step that cannot be reverted cleanly. Once readers depend on quantity_numeric, reverting the writer leaves rows with a null there, and the gap is only discoverable by re-reading the old column, which the readers have stopped doing.
Backfill older partitions in batches keyed on the primary key, oldest first, committing every few thousand rows with a pause between batches, and skipping the partition still receiving writes until it rotates. Each batch is an ordinary UPDATE taking row locks only. The cost is bloat and WAL rather than blocking, so watch dead tuples and let autovacuum keep pace instead of wrapping 400 partitions in one transaction.
Make NOT NULL cheap with the three-step form: add constraint ... check (environment is not null) not valid (brief ACCESS EXCLUSIVE, no scan), then validate constraint (SHARE UPDATE EXCLUSIVE, scans while reads and writes continue), then set not null, which from PostgreSQL 12 uses the validated check and skips its own full scan. Do this per partition, then on the parent.
Switch and contract: move reads to the new column behind a flag, verify over a full period that both columns agree on freshly written rows, drop the old column (metadata-only), and only then remove the dual-write. Any index on the new column goes on with CREATE INDEX CONCURRENTLY per partition, since CIC is not supported on a partitioned parent: create the parent index with ONLY, build each child concurrently, then ALTER INDEX ... ATTACH PARTITION until the parent index becomes valid.

Follow-up

A CREATE INDEX CONCURRENTLY fails halfway through the partition list. What state is the table in, how do you detect it, and what do you run?
The producer computes quantity itself. What happens to a request already in flight when the dual-write deploy lands, and does it matter?
Give two queries that prove the backfill is complete: one cheap enough to run every minute, one authoritative.

Model credential revocation so history survives the delete

Easy · soft delete · partial indexes · check constraints · audit history

tenant_api_key stores key_id, tenant_id, workspace_id, name, key_prefix, secret_hash, scopes text[], status (active, revoked, expired, compromised), auth_version, created_at, expires_at, last_used_at, revoked_at, revoked_reason. Rotation inserts a new row and revocation never deletes, because an incident review asks which credential served a request last quarter. Write the constraints that enforce: a label is unique only among a tenant's live keys, revoked_at and status can never disagree, and scopes is never empty. Then write the authentication lookup predicate, and name one column in this table that must stay out of it.

Approach

Reach for a partial unique index rather than a plain UNIQUE: create unique index on tenant_api_key (tenant_id, name) where revoked_at is null. Any number of revoked rows may share a label, the live namespace stays unique per tenant, and the revoked majority is not in the index at all, so it stays small on a table that only grows.
Tie the nullable timestamp to the enum so the two cannot drift: check ((revoked_at is not null) = (status in ('revoked','compromised'))) and check ((revoked_at is null) = (revoked_reason is null)). A revocation that records no reason is the one an incident review cannot use.
Write the emptiness check as check (cardinality(scopes) > 0), not array_length(scopes, 1) > 0. array_length returns NULL for an empty array, a CHECK constraint passes when its expression is NULL, so the array_length version accepts exactly the value it was written to reject.
Make the lookup a single index probe with every liveness condition inside it: where secret_hash = $1 and revoked_at is null and (expires_at is null or expires_at > now()) and auth_version = $2, backed by a unique index on secret_hash. Nothing is filtered in application code, so there is no path that forgets a clause.
Keep last_used_at out of that predicate. It is written asynchronously and is allowed to lag by a minute, so it is a usage signal; feeding it into an authorisation decision makes the decision depend on a write that may be late, batched away or lost.
Flag the modelling smell while you are here: expired is derivable from expires_at < now(), so storing it as a status obliges a job to keep it true and guarantees the column is wrong between the expiry instant and that job's next run. Derive it in the predicate; keep the stored status for states that are decisions rather than clock readings.

Worked solution (20 min)

Create the table with all three constraints on a scratch database and insert two revoked rows sharing (tenant_id, name); the partial index should accept both.
Insert a second live row with that same name and confirm the violation names the partial index.
Run update tenant_api_key set revoked_at = now() leaving status = 'active' and confirm the CHECK rejects it; then try insert ... scopes = '{}' against both the cardinality and the array_length forms and note that only one rejects it.
Run explain (analyze, buffers) on the lookup predicate for a live key and confirm an index scan on secret_hash with rows removed by filter equal to zero.

Expected result: A partial unique index on (tenant_id, name) where revoked_at is null; check ((revoked_at is not null) = (status in ('revoked','compromised'))); check ((revoked_at is null) = (revoked_reason is null)); check (cardinality(scopes) > 0); a single-probe lookup on secret_hash plus revoked_at, expires_at and auth_version; last_used_at excluded, with a note that 'expired' should be derived rather than stored.

Verify it yourself

Two revoked keys with the same name insert cleanly; a second live one does not.
The array_length form of the emptiness check accepts an empty array and the cardinality form rejects it.
No statement anywhere in the design deletes or updates a revoked row.
Revoking a key touches exactly one row and takes effect on the next lookup with no application-side filtering.

Follow-up

Rotation issues a replacement while the old key stays live for a 30-day overlap. What does the uniqueness rule become, and what does the UI show to tell two same-named keys apart?
A password reset bumps the principal's auth_version. No row in this table changed. How does the next request fail, and what query counts how many keys that bump just killed?
A key turns up in a public repository. Which columns let you find it, and what do you write to the row?

Publish limit semantics that clients can back off against

Medium · rate limiting · quotas · degraded mode

The gateway enforces three separate limits: a per-tenant token bucket, a monthly plan quota, and a cap on concurrent in-flight requests. Tenants hold several credentials and call from three regions, and today they see undifferentiated 429s. Design the client-facing contract: which headers carry which limit, the status codes that distinguish slow down from out of plan allowance from too many in flight, and what a well-behaved SDK does for each. Also state what the gateway returns when the shared counter store is unreachable, and what bound that choice implies.

Approach

Separate the three objects before naming a header. A bucket refills continuously, a quota does not refill until the period rolls, and a concurrency cap clears when an in-flight request finishes. They have different remedies, so they cannot share a status code.
Map them accordingly: 429 with Retry-After for the bucket, where waiting works; 429 with a distinct code and a reset hint for concurrency, where waiting works only if something else completes; and 402 or 403 with a plan code for quota, where waiting never works and the SDK must surface it rather than sleep on it.
Emit RateLimit-Limit, RateLimit-Remaining and RateLimit-Reset for the bucket only, and document Remaining as advisory. It comes from a shared counter that other requests are changing concurrently, so a client treating it as a reservation has built a race.
State the aggregation in the docs: the limit applies per tenant across every credential and every region. The cheap per-pod bucket of rate/N is correct only when traffic spreads evenly, and a tenant whose connections land on a few pods is throttled well under its published limit while a widely spread one exceeds it.
Decide the degraded mode in advance and publish it. Failing open serves unmetered traffic during a counter outage; failing closed converts a counter outage into a total outage. Pick one, bound it, for example a local fallback bucket at a fraction of the limit for the duration, and say so in the contract.
Specify SDK behaviour: honour Retry-After over local backoff, apply full jitter otherwise, never retry 402 or 403, and cap total attempts so retries expire before the caller's own deadline.

Follow-up

Two services under the same tenant each read Remaining = 50 and each send 50 requests. What did the contract promise, and what actually happens?
How would you keep a burst from a staging workspace out of the production workspace's share of the same tenant's bucket?

Webhook fan-out with per-endpoint isolation and backoff

Medium · fan-out · backpressure · circuit breaker · ordering

One domain event fans out to every matching subscription, producing a webhook_delivery row per (subscription_id, event_id, redelivery_seq). Peak unique event rate is 20k/second; attempts run five to ten times that once fan-out and retries are counted. One customer endpoint has returned 503 for six hours and its backlog holds days of events; every other customer must be unaffected. Design the delivery system: how a worker claims work, the backoff schedule, the per-endpoint circuit breaker, the queue partitioning, and whether you offer ordering per subscription. State the delivery guarantee in one sentence.

Approach

State the guarantee first, because it determines the rest: at-least-once with a stable event_id, and the consumer documented as responsible for idempotency. Exactly-once over HTTP is not deliverable - the 200 can be lost after the customer has already committed - so any design that promises it is either lying or is really offering at-most-once.
Partition work per subscription rather than into one global pool, with a concurrency cap per subscription. With a shared pool, the endpoint that has been dead for six hours consumes workers on retries that will fail, and every other customer's delivery latency rises: head-of-line blocking across tenants is the exact failure being designed against here.
Claim by compare-and-set with a fencing token: UPDATE webhook_delivery SET status = 'in_flight', lease_token = $new, leased_until = now() + interval '60 seconds' WHERE delivery_id = $1 AND status IN ('pending','failed_retryable') AND (leased_until IS NULL OR leased_until < now()), and make the terminal write carry AND lease_token = $new so a paused worker's late write is rejected rather than overwriting a newer attempt. Find due work through the partial index on next_attempt_at WHERE status IN ('pending','failed_retryable'), so the scan is proportional to live rows rather than to the terminal rows that outnumber them by orders of magnitude.
Use full jitter: sleep uniformly in [0, min(cap, base x 2^(attempt-1))]. Plain exponential backoff hands a recovering endpoint its entire backlog as one synchronised herd and knocks it over again; full jitter de-correlates it. Then check the schedule actually spans the retention you promise - with base 1 s and a 3,600 s cap, twenty attempts have an expected total elapsed time of only about 4.6 hours, so a twenty-four-hour promise needs roughly fifty-nine attempts or a larger cap.
Trip a circuit per endpoint on consecutive failures or a failure ratio over a rolling window: stop dispatching, push next_attempt_at out or mark new deliveries dropped_circuit_open, and half-open with exactly one probe rather than a batch. Bound the backlog explicitly with a per-subscription cap or retention, and decide in advance whether a recovered endpoint receives six hours of events at full rate or a pointer telling it to fetch what it missed.
Offer ordering only as an opt-in mode of one in-flight attempt per subscription, and price it honestly: with parallel attempts a retried event overtakes a newer one, so ordering requires serialisation, and serialisation means one slow endpoint blocks its own queue entirely. That converts a shared problem into that customer's own problem, which is the right place for it, but it is still a real cost.

Worked solution (35 min)

Compute the attempt rate: 20k unique events/second x mean fan-out x retry multiplier, and size worker pools and the per-subscription concurrency cap from it.
Write the claim statement and the terminal write, and point at the clause that rejects a resumed worker's stale write.
Tabulate the backoff for attempts 1 to 20 with base 1 s and cap 3,600 s, take the expected value of each full-jitter sleep as half its ceiling, and sum to get total expected elapsed coverage.
Decide the policy for a subscription down six hours: events buffered, bytes held, and what the customer actually receives when it returns.

Expected result: At-least-once with a stable event_id; per-subscription lanes with a concurrency cap; lease-and-fence claiming; full jitter whose schedule matches the retention promised - with base 1 s and a one-hour cap, twenty attempts covers only about 4.6 hours of expected elapsed time, so twenty-four hours needs roughly fifty-nine attempts or a larger cap; a per-endpoint breaker with a single half-open probe; ordering offered only as opt-in single-in-flight with head-of-line blocking named as its price.

Verify it yourself

Sum the expected sleeps in your schedule and compare against the retention window you promised the customer. If the two numbers disagree, one of them is wrong and it is usually the promise.
Execute the terminal write with a stale lease_token and confirm it affects zero rows. If it succeeds, the lease was doing the work alone and the token is decorative.
The dead endpoint's backlog must not be visible in any queue another subscription's workers poll. If one index scan serves all subscriptions, the isolation is nominal.

Follow-up

The endpoint recovers. Does it receive six hours of events at full rate, and what does that do to it?
Trace the exact code path by which an event belonging to one tenant could be signed and sent to another tenant's endpoint.
A customer insists they never received an event your row marks delivered. What evidence do you have, and what does payload_digest let you prove?

Regional error rate explodes after a dependency merely slows

Hard · retry storm · circuit breaker · partial outage

A control-plane read replica in one region degrades from 4 ms to 120 ms. Within ninety seconds that region's gateway error rate rises from 0.01% to 40% and its p99 becomes bimodal, one mode near the old p99 and one at the client timeout. The other two regions are unaffected. The gateway retries control-plane reads three times with exponential backoff and no jitter. Give an ordered checklist that separates trigger from amplifier, the offered-load arithmetic, and the controls that break the loop.

Approach

Split the incident into three questions before touching a control: what started it, what amplified it, and what would make recovery slow. Here they are the replica slowdown, the retry policy interacting with queueing, and a synchronised unjittered herd at recovery. They are different mechanisms and each needs its own control.
Read the distribution rather than the mean. A bimodal p99 with one mode pinned at the client timeout is two populations, not one degraded path; split latency by cache hit and miss and confirm the fast mode is hits and the timeout mode is misses that reached the replica.
Do the load arithmetic. Three retries turn one client request into up to four upstream requests, so offered load reaches roughly 4x on a dependency that is already slower, and it arrives at the worst moment. With utilisation approaching one, queueing delay grows superlinearly, which is why a 30x latency increase upstream does not produce a 30x increase downstream, it produces timeouts.
Break the loop with controls that bound offered load rather than with more attempts: a concurrency limit on the control-plane client so at most N calls are in flight and the remainder fail fast, a circuit breaker scoped per dependency and region, and a retry budget capping retries at a small fraction of base traffic so amplification has a ceiling that does not depend on how many clients are retrying.
Add full jitter to whatever retries survive, sleeping uniformly in [0, min(cap, base x 2^attempt)], so attempts de-correlate instead of arriving in waves aligned to the moment of failure.
Decide the unreachable-dependency behaviour in advance, because it is the actual product decision underneath: serving from an expired credential cache keeps the product available while extending a revoked key's life past the stated bound, and failing closed converts a dependency degradation into a total outage. State the mode and the staleness number rather than letting the timeout choose.

Follow-up

The replica recovers. Describe what happens in the first ten seconds with your controls in place versus without them.
Which single metric would have paged before the error rate moved, and why is upstream latency by itself not it?
Requests that fail fast under the concurrency limit still need an answer. What does the gateway return, and what does it do to the usage event it would otherwise have emitted?

Suggested practice checklist

For a candidate senior enough that the loop turns on design and judgement rather than on whether the coding round gets finished. Five days build one system properly and then stress it; coding gets a single maintenance day, on the assumption that the risk at this level is an unexamined tradeoff rather than a missed algorithm.

Day 1 — Numbers before diagrams

Build your own reference card of the figures you will re-derive all week: bytes for a realistic record, requests per second implied by a given daily active count, and the storage that a year at a given write rate produces. Derive each one rather than copying it, because the derivation is what survives a follow-up.
Turn one product statement into capacity requirements. From ten million daily users at four writes and forty reads each, state the peak-to-average factor you are assuming and why, then produce peak write QPS, peak read QPS and a year of storage.
Write the two numbers whose order of magnitude changes the design, the read-to-write ratio and the working-set size against memory per node, and state the threshold at which each one flips your answer.

Deliverable: A one-page numbers card and one worked capacity estimate with every assumption written down.

Day 2 — One system, from requirements to schema

Spend the first ten minutes producing only functional requirements, non-functional targets with numbers attached, a p99 latency, a durability expectation, a consistency requirement, and an explicit out-of-scope list.
Define the interface before the boxes: the three or four endpoints, their parameters, what each returns, and which of them are idempotent.
Write the data model, then write the single access pattern that justifies it, and state what the schema would have to become if the dominant access pattern were the other one.

Deliverable: One design carried to endpoint-and-schema depth, with non-functional targets expressed as numbers and a written out-of-scope list.

Day 3 — The consistency you are actually buying

Write out what a client sees under asynchronous replication when its write commits on the leader and its next read is served by a lagging follower, then write the two fixes, pinning that session's reads to the leader for a bounded window or carrying a version token the replica must reach, and the cost of each.
Work the quorum arithmetic on paper for N of three with W and R of two, and separate what R + W > N does guarantee, that any read set intersects any write set, from what it does not: on its own it is not linearizability, and a sloppy quorum that accepts writes on nodes outside the preference list breaks even the intersection.
Take two storage choices with different defaults, a single-leader relational store committing synchronously and a quorum-replicated store that converges eventually, and write the specific product behaviour that would be wrong under each, rather than a general statement about which is stronger.

Deliverable: A page separating what quorum overlap guarantees from what it does not, with one concrete product misbehaviour attached to each gap.

Day 4 — Failure is the design

For one write path, work through the case where the client times out after the server has already committed, then design the idempotency key: who generates it, how long it is retained, and what the duplicate request returns.
Express the retry policy as parameters rather than as a word: maximum attempts, base delay, backoff factor, jitter, and which error classes are retried at all. Then state why retrying a non-idempotent write without a key is a correctness bug and not merely waste.
Compute the fan-out effect on tail latency. If a request waits on ten backends and each independently exceeds its p99 one percent of the time, the chance at least one is slow is 1 - 0.99^10, about ten percent. Then write why independence is the optimistic assumption and what correlates them in practice.
Name the backpressure mechanism for one queue or one dependency in the design, a bounded queue with shedding or a concurrency limit, and write what the caller is told when it engages.

Deliverable: One write path with an idempotency design, a parameterised retry policy, and a written tail-latency calculation with its assumption named.

Day 5 — Scaling the hot path

Choose cache-aside or write-through for one read path and write the staleness window each produces, then name the invalidation event and what the system does when that event is lost.
Design against the stampede: either coalesce requests so only one recomputes a missing key, or refresh early with jittered expiry, and write why identical TTLs on keys populated in the same moment produce a synchronised expiry and a thundering herd.
Shard one table by a key you choose, then answer the two questions that break the choice: which queries now require a scatter-gather, and what happens to the distribution when one tenant is a hundred times larger than the median.
Write the cost of adding a node under plain modulo placement, where nearly every key moves, against consistent hashing, where roughly one key in n+1 moves, and state what virtual nodes are for.

Deliverable: A caching and sharding decision for one path, each with its failure mode and its rebalancing cost written beside it.

Day 6 — Keep the coding hand in, at the bar that applies to you

Solve one medium problem in thirty minutes, then spend twenty more making it production-shaped: named invariants, validation at the boundary, and errors that distinguish a caller mistake from an internal fault.
Write the tests you would require of a colleague's version of that function: one for empty input, one for the boundary, and one for the case the implementation is most likely to get wrong.
Read a piece of your own code from six months ago and write the change you would ask for, phrased as you would actually phrase it in review.

Deliverable: One problem hardened to review standard, with its test list and one written review comment.

Day 7 — Defend it while being interrupted

Run a forty-five-minute design mock with an interviewer briefed to change a requirement halfway, a tenfold traffic increase or a new strict consistency requirement, and to push on one number you estimated.
Rehearse the two sentences a senior loop is listening for: naming the tradeoff you are choosing against and why, and saying what you would measure to learn that the choice was wrong.
Prepare the design you regret: a real decision, the constraint that produced it, what it cost, and what you changed afterwards.

Deliverable: Mock notes recording how the design changed under the new requirement, plus a written account of one regretted decision.

The estimate and the day you knew

A slipped date is only a bad story if you sat on it. What matters is what you believed when you gave the number, the signal that told you it was wrong, how many days passed before you said so, and what you cut rather than asking for more time. Scope you defended counts as much as scope you dropped.

How do you approach learning a new, highly specialized technology stack under a deadline?

Medium · behavioural and engineering judgement

Approach

State the situation in two sentences and spend the rest on the reasoning.
Pick a story where you made the decision, not one where you watched it.
Close with what you would do differently, concretely.

Follow-up

What would you do differently if you ran that again?
What did you decide not to do, and why?

What was the hardest time in your work experience?

Medium · behavioural and engineering judgement

Approach

State the situation in two sentences and spend the rest on the reasoning.
Pick a story where you made the decision, not one where you watched it.
Give the blast radius: what could have broken, and what you measured.

Follow-up

What did you decide not to do, and why?
How did you know your change caused the improvement?

How do you handle feedback on your code or architectural designs from senior stakeholders?

Medium · behavioural and engineering judgement

Approach

State the situation in two sentences and spend the rest on the reasoning.
Give the blast radius: what could have broken, and what you measured.
Name the disagreement and how you resolved it with evidence.

Follow-up

How did you know your change caused the improvement?
What would you do differently if you ran that again?

Describe a situation where you had to reconcile conflicting technical requirements.

Medium · behavioural and engineering judgement

Approach

Close with what you would do differently, concretely.
Pick a story where you made the decision, not one where you watched it.
Name the disagreement and how you resolved it with evidence.

Follow-up

What did you decide not to do, and why?
How did you know your change caused the improvement?