GCP Data Architect interview Q&A – Part 1: BigQuery & storage design
1. Explain BigQuery's architecture in two minutes.
BigQuery is serverless and separates storage from compute. Data is stored in a columnar format on Google's distributed file system; queries run on the Dremel execution engine, which splits work across many workers called slots. A petabit-scale network connects the two, so compute can scale independently and you never manage servers or indexes.
2. Partitioning vs clustering – when do you use each?
Partitioning splits a table into segments by a DATE/TIMESTAMP/DATETIME column, ingestion time or an integer range, so filters on that column skip whole partitions and cost is predictable before the query runs. Clustering sorts data inside each partition by up to four columns, so filters and aggregations on those columns read fewer blocks. Typical design: partition by event date, cluster by high-cardinality filter columns such as customer_id or country. Enable require_partition_filter on large tables.
The rest of this is premium
Written by an OurLogic creator – 60% of your payment goes to them.
Log in to unlockFeedback & comments (2)
The storage decision table is gold – finally clear on Bigtable vs Spanner.
Worth every rupee. Q4 and Q10 came up almost word for word in my GCP data architect interview.