Singularity — Serverless Data Platform

// Platform Services

Three services.
Infinite possibilities.

⬡

Distributed Queue

A horizontally scalable message queue built for high-throughput workloads. Exactly-once delivery, ordered partitions, and consumer groups — zero ops overhead.

AT-LEAST-ONCEEXACTLY-ONCEFIFODLQFANOUT

◷

Scheduler & Delayed Messages

Cron-based and arbitrary-delay message scheduling with millisecond precision. Schedule jobs months in advance with guaranteed delivery and full audit trail.

CRONONE-TIMERECURRINGDELAY

◈

Horizontal Cache

A globally distributed caching layer with zero-downtime rolling updates, automatic sharding, and consistent hashing. Drop-in Redis-compatible API.

REDIS API0 DOWNTIMEAUTO-SHARDMULTI-REGION

// Product Details

Built for
serious workloads.

PRODUCER

P-01

P-02

P-03

↓

PARTITION RING

P-0

P-1

P-2

P-3

↓

CONSUMER GROUP

C-01

C-02

C-03

● 1.2M msg/s throughput

Distributed Queue

Messages that never drop.

Built on an append-only log architecture, Singularity Queue scales horizontally across partitions without any configuration. Write once, deliver anywhere.

Exactly-once delivery with idempotent producers and transactional consumers
Consumer groups with automatic rebalancing and offset management
Dead-letter queues with configurable retry policies and backoff
Fan-out patterns: broadcast single messages to unlimited subscribers
Message TTL, priority queues, and ordered partitions
Real-time lag monitoring and auto-scaling consumer groups

singularity-scheduler.ts

// Schedule a delayed message
const job = await singularity.schedule({
  queue: "billing.invoices",
  payload: { userId: "usr_xyz" },
  delay: "30d", // 30 days later
  idempotencyKey: "inv-2024-q1"
});

// Recurring cron job
await singularity.cron({
  schedule: "0 9 * * MON-FRI",
  queue: "reports.daily",
  timezone: "America/New_York"
});

// ✓ Job enqueued: job_9f3k2m
// ✓ Delivery guaranteed: 2024-04-12T09:00

Scheduler & Delayed Messages

Time is just another variable.

Schedule messages from 1 millisecond to 1 year in the future. Singularity's scheduler is durable, distributed, and resilient to node failures.

Millisecond-precision delay scheduling up to 365 days in the future
Cron expressions with timezone support and daylight saving handling
Idempotency keys prevent duplicate job creation on retries
Full audit log: created, scheduled, delivered, acknowledged
Cancel or modify scheduled jobs up until delivery
Durable across full region failures with 0-RPO cross-region replication

CLIENTS

APP
SERVER

MICROSERVICE
CLUSTER

EDGE
FUNCTION

CONSISTENT-HASH
LOAD BALANCER

PRIMARY

SHARD-1
PRIMARY

SHARD-2
PRIMARY

SHARD-3
PRIMARY

REPLICAS

SHARD-1
REPLICA

SHARD-2
REPLICA

SHARD-3
REPLICA

PRIMARY WRITE

READ REPLICA

LIVE

Horizontal Cache

Cache that never sleeps.

A Redis-compatible caching layer that scales horizontally using consistent hashing. Rolling updates guarantee zero downtime — no more maintenance windows.

100% Redis-compatible — migrate existing code with zero changes
Zero-downtime scaling: add or remove nodes with automatic rehashing
Multi-region replication with configurable consistency (async/sync)
LRU/LFU/TTL eviction policies with per-key overrides
Read replicas for geo-local reads with eventual consistency
Built-in cache warming and gradual traffic shifting during migrations

// Infrastructure

How Singularity
stays alive.

PRIMARY REGION — US-EAST-1

⬡

QUEUE
NODE-1

⬡

QUEUE
NODE-2

◈

CACHE
PRIMARY

◷

SCHED
NODE-1

◈

CACHE
REPLICA

⟵ REPLICATING TO US-WEST, EU-CENTRAL, AP-SOUTHEAST ⟶

CONSENSUS

Raft-based Leader Election

Every service uses a Raft consensus protocol for leader election. If a primary node fails, a new leader is elected in under 500ms with no data loss.

STORAGE

Append-Only Log Engine

Messages are persisted to a write-ahead log before acknowledgment. Durability is guaranteed even if the process crashes after ack but before flush.

NETWORKING

Gossip Protocol Membership

Nodes discover each other and detect failures using a gossip-based SWIM protocol. Ring membership is eventually consistent and extremely low overhead.

PARTITIONING

Consistent Hash Ring

The cache and queue use a virtual-node consistent hash ring. Adding or removing a node redistributes only ~1/N of keys, ensuring minimal disruption.

REPLICATION

Async + Sync Replication

Choose synchronous replication for strong consistency (RPO=0) or asynchronous for maximum throughput. Mixed modes supported per-topic or per-key.

OBSERVABILITY

OpenTelemetry Native

Every operation emits structured traces, metrics, and logs in OTLP format. Works with Datadog, Grafana, Honeycomb, or any OTel-compatible backend.

Replication Pipeline

How every write travels from producer to all replicas

STEP 01

Write Received

Client sends message to the partition leader via HTTP/2 or SDK. The leader assigns a monotonic sequence ID.

STEP 02

WAL Append

Leader appends to write-ahead log. The entry is fsync'd to durable storage before acknowledgment is possible.

STEP 03

Follower Replicate

Followers pull WAL segments from the leader. Sync mode waits for quorum (n/2+1) before acking. Async mode acks immediately.

STEP 04

Cross-Region Relay

A dedicated replication bus streams committed entries to remote regions over persistent multiplexed TCP tunnels (<50ms lag).

STEP 05

Client ACK

Once durability requirement is satisfied, the leader returns ACK. Consumers can now fetch the message from any replica.

// Pricing

Simple pricing,
no surprises.

Pay per message, per cache-GB, and per scheduled job. No cluster management fees.

STARTER

Free forever · no credit card

✓1M queue messages/mo
✓10K scheduled jobs/mo
✓256MB cache
✓1 region
✓Community support

PRO

$49

per month · usage-based above limits

✓100M queue messages/mo
✓Unlimited scheduled jobs
✓10GB cache
✓3 regions · <50ms replication
✓99.99% SLA · email support
✓OpenTelemetry export

ENTERPRISE

Custom

Annual contracts · volume discounts

✓Unlimited everything
✓12 global regions
✓99.999% SLA
✓Dedicated support + SRE
✓Private deployment option
✓Custom data retention

Data infrastructure
at singular scale

Three services.
Infinite possibilities.

Built for
serious workloads.

Messages that never drop.

Time is just another variable.

Cache that never sleeps.

How Singularity
stays alive.

Simple pricing,
no surprises.

Your data platform
is one command away.

Data infrastructureat singular scale

Three services.Infinite possibilities.

Built forserious workloads.

Messages that never drop.

Time is just another variable.

Cache that never sleeps.

How Singularitystays alive.

Simple pricing,no surprises.

Your data platformis one command away.

Data infrastructure
at singular scale

Three services.
Infinite possibilities.

Built for
serious workloads.

How Singularity
stays alive.

Simple pricing,
no surprises.

Your data platform
is one command away.