Glosarium
Contents

Architecture & System Design › Cloud Design Patterns

Over-Fetching

Retrieving far more data than an operation needs.

Backend Engineer · Mid-levelFrontend Engineer · Mid-level

Also known as: over-fetching, over fetching, extraneous fetching

Over-fetching (extraneous fetching) downloads data nobody uses: full objects for a list showing two fields, nested expansions rendered nowhere, “just in case” includes that ride every response. The bytes cost latency, parsing, memory and money — multiplied by every user, every poll, every retry.

list needs {id, title} → API returns 40 fields incl. history blobs (20× waste)

Cures target the contract: sparse fieldsets (?fields=), shaped endpoints per view, GraphQL selection, pagination that pages (not “all plus metadata”), and measured payload budgets per screen. The general fix is asking what each UI actually renders — usually a humbling fraction of what’s fetched.

The classic mistakes:

  • “Might need it” includes. Speculative fields accrete forever (“someone might use history”); nobody removes them. Fetch what renders now; add on demand.
  • Unpaginated relations. Embedding full child collections per parent multiplies bytes with list size. Paginate or link relations; never embed unbounded.
  • Ignoring mobile cost. Desktop-indifferent megabytes meter mobile data plans and memory; over-fetching taxes the most constrained users most.
  • Caching the bloat. Cached oversized responses entrench waste (invalidation granularity suffers too). Shape before caching.
  • Version-frozen fields. Deprecated fields kept “for compatibility” downloaded by clients years migrated. Sunset with metrics, not sentiment.
  • Measuring requests, not bytes. Optimising call counts while payloads balloon misses the transfer cost. Budget bytes per view alongside call counts.
  • Backend-shaped responses. Serialised database rows (all columns, internal ids, audit fields) served raw leak internals and bloat simultaneously. Shape deliberately at the boundary.

The discipline: fetch what renders, page what lists, sunset what aged, budget bytes per view. Bandwidth is user time and real money — stop shipping what nobody reads.