How the interviewer opens it
“Design a collection layer that receives logs, metrics, and traces from applications. It must send this data to several observability backends.”
The prompt is vague on purpose. The first minutes are for you. Reduce the prompt to the core features and agree on the numbers before you draw.
What to clarify first
- Sources and destinations
- Traffic volume
- Tenants and access mix
- Resource budget
- Latency
- Availability, delivery, and durability
- Destination outages
- Sample policy
- Overload policy
- Scope exclusions
The interviewer has the answers. In a SysRound round, Alex gives these answers, as a real interviewer does.
Scope of a strong answer
Core requirements
- Receive logs, metrics, and traces from application agents.
- Route records to destinations according to tenant rules.
- Apply log filters and consistent trace sample decisions.
- Bound resource use during traffic peaks and destination failures.
Usually out of scope
- Store and query telemetry for end users.
- Build dashboards.
- Instrument application code.
- Recover data after a regional disaster.
What interviewers ask about
- A slow destination
- One input record can require several deliveries. A failed destination can consume all capacity while healthy destinations still need low latency.
- Record delivery
- An acknowledgement crosses a durability boundary. A crash or lost response can leave the collector unsure whether a destination accepted a record.
- Trace sample decisions
- Spans for one trace arrive at different times and collectors. Independent decisions can produce incomplete traces or consume unbounded memory.
- Capacity across tenants
- A large tenant can exhaust shared resources before fleet capacity runs out. Small records, large payloads, and many trace IDs stress different limits.
Common mistakes
- Treat all three signal types as interchangeable when the service applies sample rules.
- Acknowledge data before the collector reaches the required durability boundary.
- Let one destination block every destination.
- Use retries or queues without byte limits and a full-state policy.
- Claim loss-free delivery during an unlimited outage with finite resources.
- Assign one worker per tenant without checking tenant size.
The bar by level
- Senior (L5/E5)
- Calculates byte budgets and outage capacity. Defines acknowledgement boundaries, independent destination progress, tenant limits, and consistent trace sample decisions. Handles partial responses and late spans.
- Staff (L6+/E6+)
- Defines loss guarantees across every resource boundary. Explains distributed quota enforcement, worker ownership, recovery capacity, and safe route changes. Compares stronger durability against its cost.