Database Internals

How databases store, index and recover data under the hood.

Backend Engineer

Own a feature end to end without hand-holding.

Own a system, its failure modes, and its trade-offs.

Snapshot IsolationEach transaction reads from a consistent snapshot of the database.
Storage EngineThe part of a database that reads and writes data on disk.
System CatalogThe tables where a database describes its own schema.
B-Tree vs LSM TreeRead-optimized pages vs write-optimized logs, and when each wins.
Buffer Pool / Page CacheKeeping hot disk pages in memory.
CheckpointFlushing in-memory changes to disk so recovery doesn't replay the whole log.
fsync and DurabilityWhat "written to disk" actually means, and when data can still be lost.
LSM TreeA write-optimized structure that buffers writes in memory and merges sorted files.
SSTables and CompactionSorted immutable files and the background process that merges them.

Set technical direction for the organization.

Two-Phase LockingAcquire all locks, then release them: the classic way to get serializability.
Pages, Tuples and File LayoutHow rows are physically laid out on disk.
Write AmplificationOne logical write turning into many physical ones.

Own pipelines and models end to end, including their quality.

Design the platform's storage, processing and modeling choices.

Snapshot IsolationEach transaction reads from a consistent snapshot of the database.
Storage EngineThe part of a database that reads and writes data on disk.
System CatalogThe tables where a database describes its own schema.
B-Tree vs LSM TreeRead-optimized pages vs write-optimized logs, and when each wins.
Buffer Pool / Page CacheKeeping hot disk pages in memory.
CheckpointFlushing in-memory changes to disk so recovery doesn't replay the whole log.
fsync and DurabilityWhat "written to disk" actually means, and when data can still be lost.
LSM TreeA write-optimized structure that buffers writes in memory and merges sorted files.
SSTables and CompactionSorted immutable files and the background process that merges them.

Set data strategy and architecture across the company.

Two-Phase LockingAcquire all locks, then release them: the classic way to get serializability.
Pages, Tuples and File LayoutHow rows are physically laid out on disk.
Write AmplificationOne logical write turning into many physical ones.

Own an app's architecture, performance, and failure modes.