id string | topic string | question string | answer string | sources list | license string | messages list |
|---|---|---|---|---|---|---|
de-021 | streaming | What does at-least-once delivery mean for a consumer? | At-least-once delivery permits a record to be processed again after a retry or restart. Make the sink idempotent or deduplicate with a stable event identifier. Commit offsets only when the corresponding work has reached the durability point your design promises. Document how duplicates are detected and how long dedupli... | [
"https://kafka.apache.org/documentation/#design"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What does at-least... |
de-013 | sql-modeling | Why can a join inflate a metric? | A join inflates a metric when the joined side has multiple matches for a key that was assumed to be unique. Check row counts and key cardinality before and after the join, then test the declared uniqueness. Resolve duplicates or aggregate the many-side to the required grain before joining; do not hide the problem with ... | [
"https://www.postgresql.org/docs/current/using-explain.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "Why can a join inf... |
de-017 | spark | Why do too many small files hurt a lakehouse table? | Many tiny files increase listing, planning, and per-file open overhead, and can reduce scan efficiency. Compact files periodically while preserving the table format's transaction semantics. Choose a target size based on workload and storage; do not compact while an incompatible writer is active. Track file counts and q... | [
"https://iceberg.apache.org/docs/latest/evolution/",
"https://spark.apache.org/docs/latest/sql-performance-tuning.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "Why do too many sm... |
de-024 | streaming | What is a watermark used for? | A watermark bounds how long a streaming system waits for late event-time data before finalizing certain stateful results. It trades completeness for bounded state and latency. Set it from observed lateness and the business correction window, then monitor records arriving beyond it. A watermark is not a guarantee that n... | [
"https://spark.apache.org/docs/latest/streaming/index.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What is a watermar... |
de-047 | security-delivery | What should a data pipeline CI check? | Run formatting and static checks, unit tests for transformations, contract tests for representative inputs, and a small integration test when practical. Validate DAG or job configuration without contacting production systems. Use sanitized fixtures, protect secrets, and make deployment require a reviewed artifact rathe... | [
"https://airflow.apache.org/docs/apache-airflow/stable/best-practices.html",
"https://docs.getdbt.com/docs/build/incremental-models"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What should a data... |
de-046 | security-delivery | What does least privilege mean for a pipeline? | Give each workload only the read and write access required for its specific sources, destinations, and actions. Separate development and production identities, and avoid shared administrator credentials. Review grants when ownership changes. A narrow identity limits the impact of a bug or leaked token and makes audit t... | [
"https://airflow.apache.org/docs/apache-airflow/stable/best-practices.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What does least pr... |
de-016 | spark | When should I broadcast a join side? | Broadcast a small relation when it fits safely in executor memory and avoids shuffling a much larger relation. Check the actual post-filter size, not only the source table size. Watch executor memory and confirm the physical plan. A forced broadcast can cause memory pressure or failure if the supposedly small side grow... | [
"https://spark.apache.org/docs/latest/sql-performance-tuning.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "When should I broa... |
de-011 | sql-modeling | How do I make an incremental model safe to rerun? | Define the change-detection column, unique key, and lookback window explicitly. Reprocess a bounded overlap to catch late updates, then merge or replace rows by a stable key. Test an empty input, a repeated run, an update, and a delete if supported. Compare the incremental output with a full-refresh result on a small f... | [
"https://docs.getdbt.com/docs/build/incremental-models"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How do I make an i... |
de-030 | orchestration | What is a data interval in a scheduled DAG? | A data interval identifies the logical period a run is responsible for; it is different from the wall-clock time when the task starts. Use the interval to select source data and build deterministic output paths. This keeps delayed runs and backfills reproducible. Test boundaries, time zones, and daylight-saving transit... | [
"https://airflow.apache.org/docs/apache-airflow/stable/best-practices.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What is a data int... |
de-041 | storage | What does snapshot isolation provide in a table format? | Snapshot-based table formats let readers observe a consistent table state while writers publish new metadata states. This can make concurrent reads and writes easier to reason about, but it does not remove the need to understand catalog and commit behavior. Keep snapshots and metadata according to a retention policy th... | [
"https://iceberg.apache.org/docs/latest/evolution/"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What does snapshot... |
de-037 | quality-operations | How should nulls be treated in analytics data? | A null can mean unknown, not applicable, missing, or not yet received; those meanings are not interchangeable. Preserve the distinction when the source provides it, and document defaults if a downstream model replaces nulls. Test null behavior in joins and aggregates because SQL treats null comparisons differently from... | [
"https://www.postgresql.org/docs/current/using-explain.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How should nulls b... |
de-019 | spark | What is the difference between narrow and wide transformations? | A narrow transformation can compute an output partition from a limited set of input partitions, while a wide transformation needs data exchange across partitions. Wide work usually introduces a shuffle boundary. This distinction helps explain stage breaks and performance costs, but the physical plan and runtime metrics... | [
"https://spark.apache.org/docs/latest/sql-performance-tuning.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What is the differ... |
de-006 | architecture | How should I design a reliable pipeline boundary? | Give each stage an explicit input contract, output contract, owner, and recovery method. Validate schema and key assumptions at the boundary, attach a run or source version, and publish output only after the stage succeeds. Keep transformations deterministic where possible. A clear boundary helps isolate failures and m... | [
"https://airflow.apache.org/docs/apache-airflow/stable/best-practices.html",
"https://openlineage.io/docs/"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How should I desig... |
de-045 | security-delivery | How should pipeline secrets be managed? | Keep credentials in a secret manager or the platform's protected secret store, inject them at runtime, and scope them to the minimum required actions. Never place secrets in source control, logs, task arguments, or dataset rows. Rotate credentials, audit access, and test that a failed task does not print the secret val... | [
"https://airflow.apache.org/docs/apache-airflow/stable/best-practices.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How should pipelin... |
de-025 | streaming | How should event schemas evolve? | Treat an event schema as a contract with producers and consumers. Prefer compatible additive changes, define defaults and nullability deliberately, and test old and new versions together. Breaking changes need versioning or a migration plan. Record semantic changes too: keeping the same field name while changing its me... | [
"https://kafka.apache.org/documentation/#design"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How should event s... |
de-028 | orchestration | How should a pipeline retry a failed task? | Retry transient failures with a bounded attempt count and backoff. Do not automatically retry deterministic data errors such as an invalid schema without changing the input or code. Make side effects idempotent, emit a clear failure reason, and alert after retries are exhausted. Test a retry after partial completion. | [
"https://airflow.apache.org/docs/apache-airflow/stable/best-practices.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How should a pipel... |
de-027 | orchestration | What belongs in an Airflow DAG? | A DAG should describe dependencies and schedule for work, while task code performs bounded units of computation. Keep DAG parsing lightweight, avoid doing data work at import time, and pass small references rather than large datasets between tasks. Make dependencies explicit and ensure tasks can be retried safely. | [
"https://airflow.apache.org/docs/apache-airflow/stable/best-practices.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What belongs in an... |
de-038 | quality-operations | What is a practical data contract? | A data contract states fields, types, semantics, ownership, delivery expectations, and compatibility rules between a producer and consumers. Include examples and validation, not just a schema file. Agree how breaking changes are announced and rolled out. A contract is useful only when both sides can detect and respond ... | [
"https://kafka.apache.org/documentation/#design",
"https://docs.getdbt.com/docs/build/incremental-models"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What is a practica... |
de-036 | quality-operations | Why is lineage useful during an incident? | Lineage connects a dataset to its upstream inputs, transformations, and downstream consumers. During an incident it helps identify the likely blast radius and who needs notice. Keep lineage tied to real executions and stable dataset names; manually maintained diagrams become stale unless someone owns them. | [
"https://openlineage.io/docs/"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "Why is lineage use... |
de-035 | quality-operations | How should I reconcile a new pipeline with the old one? | Run both paths over the same bounded intervals and compare row counts, key coverage, null rates, and business aggregates. Classify differences rather than relying on one total checksum. Include edge cases such as deletes, late updates, and timezone boundaries. Switch consumers only after the remaining differences are u... | [
"https://docs.getdbt.com/docs/build/incremental-models"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How should I recon... |
de-033 | quality-operations | What layers of data quality checks should a pipeline have? | Check source shape and required fields at ingestion, key and domain rules during transformation, and reconciliation or freshness at the serving boundary. Add checks that detect distinct failure modes rather than repeating the same assertion. Give each check an owner, severity, and response action so alerts lead to usef... | [
"https://docs.getdbt.com/docs/build/incremental-models"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What layers of dat... |
de-010 | sql-modeling | Should I use a natural key or a surrogate key? | Keep the source's natural identifier for traceability, but consider a warehouse surrogate key when identifiers can change, collide across systems, or vary by source. Do not discard the natural key: it is essential for reconciliation and repeatable loads. Test both uniqueness and the intended mapping between source and ... | [
"https://docs.getdbt.com/docs/build/incremental-models"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "Should I use a nat... |
de-020 | spark | How should I choose Spark partition counts? | Start from the input size, expected task parallelism, and target work per task, then measure. Too few partitions limit parallelism and create large tasks; too many add scheduling and file overhead. Inspect task-size distribution, spill, and duration. Revisit partitioning after filters and joins because their data volum... | [
"https://spark.apache.org/docs/latest/sql-performance-tuning.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How should I choos... |
de-048 | security-delivery | How do I make a production backfill safe? | Define the exact intervals and expected output before starting, use a separate bounded run if possible, and cap concurrency to protect sources. Ensure writes replace or merge deterministic partitions. Monitor source load and downstream freshness, then compare results before marking the repair complete. Keep a stop and ... | [
"https://airflow.apache.org/docs/apache-airflow/stable/best-practices.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How do I make a pr... |
de-039 | storage | Why is Parquet common for analytical workloads? | Parquet is a columnar format that lets readers skip unneeded columns and can compress similar values efficiently. It is usually a good fit for analytical scans, but layout still matters: file size, row-group statistics, partitioning, and query filters affect performance. Benchmark on representative queries rather than ... | [
"https://spark.apache.org/docs/latest/sql-performance-tuning.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "Why is Parquet com... |
de-026 | streaming | What does exactly-once processing actually guarantee? | Exactly-once claims are scoped to a system boundary and a failure model. A framework may coordinate source offsets and a transactional sink, but an external side effect can still duplicate unless it participates in the transaction or is idempotent. Write down the source, processing, and sink guarantees separately and t... | [
"https://kafka.apache.org/documentation/#design",
"https://spark.apache.org/docs/latest/streaming/index.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What does exactly-... |
de-012 | sql-modeling | How should I handle late-arriving facts? | Separate event time from ingestion time and decide how far back corrections may arrive. A bounded lookback can update recent partitions; older changes may need targeted backfills. Preserve enough source metadata to explain the correction. Measure lateness so the chosen window is evidence-based, and tell consumers wheth... | [
"https://docs.getdbt.com/docs/build/incremental-models",
"https://spark.apache.org/docs/latest/streaming/index.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How should I handl... |
de-007 | sql-modeling | Why should I declare the grain before building a fact table? | The grain states what one row represents, such as one order line or one daily account balance. It determines valid keys, measures, joins, and aggregation behavior. Write the grain in the model description and test its uniqueness. If a query mixes grains, measures can be duplicated even when every individual join is syn... | [
"https://docs.getdbt.com/docs/build/incremental-models"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "Why should I decla... |
de-005 | architecture | How do I decide whether to use CDC or periodic snapshots? | Use CDC when consumers need changes with low latency or need deletes and update history. Use snapshots when the source lacks reliable change logs, the table is small, or periodic state is sufficient. CDC requires stable ordering, offsets, schema evolution, and delete semantics. A snapshot is simpler but can miss short-... | [
"https://kafka.apache.org/documentation/#design"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How do I decide wh... |
de-008 | sql-modeling | When is a star schema useful? | A star schema is useful when analysts need predictable joins between descriptive dimensions and measurable facts. Keep facts at a declared grain and dimensions focused on descriptive attributes. It can make reporting easier than a highly normalized operational schema. It is not a substitute for understanding source sem... | [
"https://docs.getdbt.com/docs/build/incremental-models"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "When is a star sch... |
de-029 | orchestration | What is a backfill, and how can it go wrong? | A backfill reruns historical intervals to repair or recompute data. It can overload source systems, collide with current writes, or change historical results. Bound the interval and concurrency, make writes idempotent, and compare outputs before publishing. Announce whether downstream consumers should expect corrected ... | [
"https://airflow.apache.org/docs/apache-airflow/stable/best-practices.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What is a backfill... |
de-015 | spark | How do I investigate data skew in a Spark job? | Compare task durations and input sizes: a few much slower tasks often indicate skewed keys or partitions. Identify hot keys, then consider pre-aggregation, salting, splitting exceptional keys, or a skew-aware join strategy. Validate that the transformation preserves results. Increasing the number of partitions alone ma... | [
"https://spark.apache.org/docs/latest/sql-performance-tuning.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How do I investiga... |
de-040 | storage | How should I partition a large table? | Partition on fields frequently used for selective pruning and with a manageable number of distinct values, often a date or coarse region. Avoid partitioning by high-cardinality identifiers, which can create huge numbers of small files. Measure actual query filters and file counts, and remember that partitioning is a ph... | [
"https://spark.apache.org/docs/latest/sql-performance-tuning.html",
"https://iceberg.apache.org/docs/latest/evolution/"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How should I parti... |
de-002 | architecture | When is batch processing a better fit than streaming? | Use batch when consumers can tolerate a defined delay and the work benefits from bounded, repeatable windows. Batch jobs are usually easier to replay, compare, and operate. Choose streaming when latency is part of the product requirement, such as reacting to an event within seconds. Streaming adds state, late-event han... | [
"https://spark.apache.org/docs/latest/streaming/index.html",
"https://kafka.apache.org/documentation/#design"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "When is batch proc... |
de-042 | storage | How should I handle a schema change in a lakehouse table? | Classify the change as additive, compatible, or breaking and check the table format's evolution rules. Test old and new readers, preserve field identity where the format uses it, and coordinate with downstream models. A schema change can be technically accepted while still changing business meaning, so document both st... | [
"https://iceberg.apache.org/docs/latest/evolution/"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How should I handl... |
de-009 | sql-modeling | How does a Type 2 slowly changing dimension work? | A Type 2 dimension keeps a new row when a tracked attribute changes, with effective start and end times and usually a current-row flag. Facts can then join to the dimension version valid at event time. Define which changes create a version, how late corrections work, and how open-ended current records are represented. | [
"https://docs.getdbt.com/docs/build/incremental-models"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How does a Type 2 ... |
de-034 | quality-operations | How do I choose a freshness SLO? | Start with the consumer's actual decision window, then include source arrival, processing, and publication time. Define the measurement point and percentile or maximum that matters. A freshness SLO should have an alert threshold and an owner. Do not promise a cadence the source system cannot reliably meet. | [
"https://docs.getdbt.com/docs/build/incremental-models",
"https://airflow.apache.org/docs/apache-airflow/stable/best-practices.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How do I choose a ... |
de-014 | spark | What is a shuffle in Spark, and why can it be expensive? | A shuffle redistributes data across partitions, commonly for joins, grouping, or sorting. It can require network transfer, serialization, disk spill, and synchronization. Reduce input early when semantics allow, use an appropriate join strategy, and inspect the physical plan and task metrics. Avoid assuming that adding... | [
"https://spark.apache.org/docs/latest/sql-performance-tuning.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What is a shuffle ... |
de-043 | storage | When should old table snapshots be expired? | Expire snapshots only after considering the recovery window, audit obligations, active readers, and time-travel use cases. Retention should be long enough to recover from a bad write but finite enough to control storage. Test the cleanup policy in a non-production table and make sure orphan-file cleanup cannot remove d... | [
"https://iceberg.apache.org/docs/latest/evolution/"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "When should old ta... |
de-001 | architecture | How should I choose between ETL and ELT? | Choose based on where transformations are safest and cheapest. ETL transforms before loading, which helps when the destination should receive only curated data or source data is sensitive. ELT loads raw data into a capable warehouse and transforms there, improving traceability and reuse. Compare data volume, privacy ru... | [
"https://docs.getdbt.com/docs/build/incremental-models"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "How should I choos... |
de-044 | storage | What is compaction, and when should it run? | Compaction rewrites many small data files into fewer larger files to reduce metadata and scan overhead. Schedule it based on file counts and workload, not just elapsed time. Coordinate with concurrent writers and preserve table-format commits. Measure query planning time and read throughput before and after to verify t... | [
"https://iceberg.apache.org/docs/latest/evolution/"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "What is compaction... |
de-031 | orchestration | When should I use a sensor? | Use a sensor when work should wait for an external condition such as a file or upstream dataset. Prefer a deferrable or rescheduling mode when the wait may be long so a worker slot is not held idle. Set a timeout, define what happens on timeout, and expose the waited-on resource in logs. | [
"https://airflow.apache.org/docs/apache-airflow/stable/best-practices.html"
] | CC BY 4.0 | [
{
"role": "system",
"content": "You are a practical data engineering mentor. Explain trade-offs, include validation and recovery steps, and flag version-sensitive advice. Do not invent current product behavior; point to official documentation."
},
{
"role": "user",
"content": "When should I use ... |
Data Engineering Mentor Q&A Dataset
An original, source-linked starter dataset of practical data engineering questions and answers. It covers architecture, SQL modeling, Spark, Kafka, orchestration, quality, lakehouse storage, and production operations.
Load with Python
pip install datasets
from datasets import load_dataset
data = load_dataset("LoNebula/data-engineering-mentor-dataset")
print(data)
print(data["train"][0]["question"])
The repository has 50 examples split into 42 training and 8 test records. Each row includes an ID, topic, question, answer, official-documentation references, license, and chat messages for instruction tuning.
Provenance and license
All Q&A explanations were originally composed for this project; no third-party Q&A corpus or documentation passages were copied. Each answer links to relevant primary documentation for readers to check product-specific details. The dataset is released under CC BY 4.0. You may reuse and adapt it with attribution to LoNebula and a link to this dataset.
References include the official documentation for Apache Airflow, Apache Kafka, Apache Spark, dbt, Apache Iceberg, PostgreSQL, and OpenLineage. Links are references, not claims that those projects reviewed or endorsed this dataset.
Scope and limitations
This is a small learning corpus focused on durable concepts and safe operational habits. It is not a complete data engineering curriculum, a benchmark, or a substitute for current product manuals and organization-specific standards. Some answers may omit context; verify version-sensitive behavior in the linked official documentation.
Citation
LoNebula/data-engineering-mentor-dataset, version 1.0, CC BY 4.0. Original Q&A explanations and dataset packaging by the dataset maintainers; official documentation links are listed per record.
- Downloads last month
- -