Data Engineering at Netflix: Maestro, Iceberg, and the Single-Band Comp Model

DE at Netflix in 2026

Netflix Data Engineering sits at the center of how the company personalizes recommendations, measures content performance, plans encoding and delivery, and runs A/B experiments across hundreds of millions of members. The Data Platform organization owns the lakehouse, the orchestration layer, the streaming substrate, and the SQL access tier; product-aligned DE teams build domain pipelines and semantic models on top of that platform.

The bar is unusually high because Netflix runs lean. Headcount in DE is small relative to the company's revenue and data volume, which means individual engineers own large surfaces. A senior DE at Netflix is typically responsible for an entire pipeline family, a platform component, or a data product end-to-end including SLAs, on-call, and stakeholder management. There is very little pure ticket-taking work; ambiguity is the default.

The 2026 hiring profile reflects this. Netflix recruits experienced engineers who can operate without close supervision, write durable systems, and contribute back to open source. New-grad pipelines for DE are minimal. Most hires come in at senior or staff levels, and internal mobility into DE from adjacent disciplines (backend, ML platform, analytics engineering) is common.

Interview process and culture

The Netflix DE loop is shorter than at most FAANG peers but denser. After a recruiter screen and a hiring-manager call, candidates typically face a technical phone screen focused on SQL and data modeling, then an onsite (now usually virtual) of four to five rounds: a system or pipeline design round, a coding round biased toward practical data manipulation rather than algorithms, a data modeling round, a behavioral round anchored to the Netflix culture memo, and a cross-functional round with a partner team.

System design at Netflix DE emphasizes choices: how would you partition this Iceberg table, where would you place the Kafka topic boundary, when would you choose Spark over Trino, how would you handle late-arriving data in a Maestro DAG. Interviewers want to hear tradeoffs, not memorized answers.

The culture round is real and load-bearing. Netflix expects engineers to give and receive direct feedback, take big bets with clear reasoning, and operate with what the culture memo calls "informed captains." Candidates who frame past decisions in terms of judgment under uncertainty tend to do well; candidates who frame decisions in terms of process compliance tend not to.

Compensation: single-band model

Netflix's compensation model is the most distinctive in the industry and shapes how DE candidates should approach offers. Instead of a leveling ladder with base, bonus, and refresh equity, Netflix offers a single annual cash number, which the candidate can split between cash and stock options at their discretion. There is no separate bonus, no annual refresh grant, and no formal level system on the engineering side.

For senior data engineers in 2026, levels.fyi data shows total compensation clustering in the mid-400s to high-500s in U.S. dollars, with staff and senior-staff equivalents reaching into the 700s and beyond. The cash-heavy structure means the offer does not depend on stock performance over a vesting cliff, which is unusual among large tech employers.

The tradeoff is that Netflix sets the number against market top-of-band and expects sustained top-of-band performance. There are no quiet years. Performance reviews are continuous and direct, and the "keeper test" — would your manager fight to keep you if you tried to leave — is applied honestly. Candidates evaluating Netflix should weigh the cash certainty and autonomy against the lack of a long-term equity upside curve and the higher performance pressure.

Tech stack: Maestro + Iceberg + Trino + Spark + Kafka

Maestro is Netflix's workflow orchestrator and the heart of the batch pipeline layer. Built to replace internal Airflow deployments that struggled at Netflix scale, Maestro supports cron, signal, and event-triggered workflows, parameterized DAGs, and step-level retries with strong observability. Netflix open-sourced Maestro in 2024 and continues to develop it in public; DE candidates should expect to author and operate Maestro flows on day one.

Apache Iceberg is the table format underneath everything. Netflix was an early Iceberg adopter and contributed many of its core features, including hidden partitioning, snapshot isolation, and schema evolution semantics. Several Netflix engineers sit on the Iceberg PMC. In practice, this means DEs at Netflix think in terms of snapshots, manifest files, and partition specs rather than Hive-style directory layouts.

Trino and Presto serve interactive and federated SQL workloads. Netflix engineers are long-time Presto contributors and now active in the Trino fork; analysts and DEs use Trino for ad-hoc queries against Iceberg tables on S3.

Apache Spark handles the heavy batch transformations, machine-learning feature pipelines, and large backfills. Apache Kafka is the streaming backbone, used for ingestion, change-data-capture, and event-driven pipeline triggers that Maestro listens to. The lakehouse storage layer is S3 with Iceberg metadata, and metadata services like Atlas and Netflix's internal Metacat tie the catalog together.

Beyond the core stack, Netflix DEs touch dbt-style transformation tooling, Jupyter and Polynote for analysis, and a deep set of internal libraries for data quality, lineage, and SLA management. Familiarity with the public components — Maestro, Iceberg, Trino, Spark, Kafka — is the table stake; the internal layer is taught on the job.