Use cases #
Pizza is built for latency-sensitive search over fresh, mutable data at scale. The scenarios below map to concrete features — follow the links for the mechanics.
Operational logs and events, searchable the moment they land #
Observability and security pipelines need indexed events and fresh results — “what happened in the last minute” cannot wait for a refresh window. Pizza’s realtime write path makes every acknowledged write searchable immediately, rollings absorb unbounded growth without reindexing, and high-volume ingest is served by batched WAL writes plus background segment builds that never gate reads.
E-commerce and marketplace data #
Catalogs are a wide-but-shallow update problem: prices, stock levels and flags churn constantly while titles and descriptions stay put.
- In-place columns turn stock and price updates into O(1) column writes — no document rebuild per change.
- Online resharding scales shard counts through traffic spikes without a reindex window.
- Filter-heavy faceting maps to term-level queries and terms / range aggregations.
Hybrid and semantic search #
The same query can mix full-text, filters and vector similarity:
match for text,
vector for embeddings (text-in,
embedded server-side), fused across fields with
multi_vector — RAG retrieval
and semantic cache lookup without a separate vector database.
Entity and relationship search #
Documents that reference each other — products and merchants, users and
groups, knowledge graphs — are served by
join (semi-join against another
index) and
traverse (bounded
graph traversal over relation fields), with
nested for per-element matching
inside nested arrays.
Multi-tenant SaaS backends #
Namespaces isolate tenants, API keys with roles ( Security) scope each tenant’s access, and per-tenant data can be mirrored elsewhere with the change feed (CDC) — off-site copies, secondary indexes, audit pipelines.
Hot counters and live analytics #
View counts, rankings, feature flags: frequent small updates plus
analytics over their distributions.
Partial updates with the
ops operator list apply atomically; percentiles, histograms and
composite aggregations summarize them without fighting the write rate.
When Pizza is not the fit #
- You need SQL-first analytics over immutable data lakes — the SQL endpoint is reserved but not implemented (see Limitations).
- You need multi-namespace tenant grouping — tenancy is namespace-scoped 1:1 today.