Components · BteloDB

A set of beautiful cards you can mass-produce. Real product UI fragments + refined SVG (gradients / glow / depth), each with an infinite-loop animation, all driven by the design tokens — zero image assets, zero JS, unit-testable. The 20 below map to BteloDB's real architecture.

S3-first lakehouse

One copy of your data, two shapes, rows for sub-millisecond OLTP, columnar Parquet for analytics, manifest on S3 too — true HTAP, zero external database.

Streaming ingest

Row-based memory streams with offset subscriptions, append, pull by offset and live-tail, just like Kafka.

DataFusion SQL

Standard SQL over hot and cold data, column pruning + predicate pushdown; add a node and analytics scales out — eventual or strong (fresh) reads.

query.sql● ready

1
2
3

SELECT user, count(*)
FROM   events
WHERE  ts > '2026-06-01'

4 rowsscanned 287 B3 ms

File-level pruning

Cold-query cost = files × handshakes. Per-file min/max stats skip irrelevant Parquet — scan less, pay less.

Vector search

Vectors as first-class citizens, k-NN search embeddings for semantic search and RAG — geo (R-tree) and JSONB are built in too.

Linearizable KV

Strongly consistent single-key state, CAS / TTL / INCR for counters, locks and sessions.

user:1042 { name, plan }

lock:job acquiredttl 30s

views INCR9 471

Mutations

Batch UPDATE / DELETE, materialized file rewrites — even streaming tables can change and delete.

id	status	amount
1003	paid	42.50
1004	refunded	0.00
1005	paid	18.00

UPDATE events SET status='refunded' WHERE id=1004

Multi-tenancy

One engine, tens of thousands of tenants, data plus CPU, memory and disk hard quotas scoped per tenant; idle tenants scale to zero.

JWT auth

HS256-verified Bearer tokens, claims = sub / tier / exp; alg=none and confusion attacks rejected.

eyJhbGci… . eyJzdWIi… . 4f9a2c…

HS256 · alg=none rejected

Realtime subscriptions

Build multiplayer experiences, SSE tail pushes new writes to every subscriber in milliseconds.

Graceful shutdown

Zero-RPO exit, drain → flush → KV snapshot → checkpoint — power loss loses nothing.

Write-ahead log

Log first, then persist, sequential append + checkpoint; the tail also lands on S3 so peers see recent writes sub-second, and replay recovers a crash.

Compaction

Small files auto-merge, size-threshold merges keep query handshakes trending down.

Cluster sharding

Rendezvous-hash sharding, tenant keys map to peer nodes; leases drive zero-ops failover, and adding a node moves only ~1/N of the data.

Usage metering

Byte-level usage billing, ingest / scan / storage / KV metered live, ready for Stripe.

Observability

Native Prometheus /metrics, requests, latency histograms, error codes and live subscriptions at a glance.

Three wire protocols

One dataset, three doorways, Redis / PostgreSQL / Kafka wire protocols speak to one engine.

Manifest on S3

Metadata as objects, file lists and stats on S3 — clone a database in zero copies, restore to any checkpoint, no external database.

manifest.jsons3://

{
  "table": "events",
  "files": [
    { "rows": 10234, "min_ts": "06-01" },
    { "rows": 9871,  "min_ts": "06-08" }
  ]
}

Global edge

Read cold data nearby, peer nodes point at one bucket and cache hot segments on local NVMe; queries hit the nearest node.

Quality gates

100% TDD + 100% coverage, deterministic simulation tests, contract ladders, warnings-as-errors — engineering law.

Component gallery