The idle timeout passes
No connections for the configured time, between one minute and one hour. Compute stops billing; storage stays online and billed.
LaravelCloud
Ready for the audit
A decider story
One data layer.
Chosen by context.
Eloquent does not care which engine sits under it, but your bill, your recovery plan and your auditor do. Laravel Cloud offers managed MySQL, Postgres and a Redis-compatible cache on every plan, AWS-native stores inside Private Cloud, and a path for the database you already have.
Open the rack
The idea in one line Pick the store for the shape of the work, not for the brand of the engine.
The store rackWhat you can attach
Serverless Postgres · autoscaling, sleeps when idleEvery plan
Laravel MySQL · Flex sleeps, Pro stays onEvery plan
Laravel Valkey · cache, sessions, queuesEvery plan
RDS MySQL and Postgres · replicas, Multi-AZPrivate Cloud
ElastiCache · Valkey or Redis OSSPrivate Cloud
Your own database · Atlas, PlanetScale, anythingBring it
SQLite · on your laptop and in testsLocal only
Injected on attach: DB_HOST, DB_DATABASE, REDIS_HOST
Resources attached
The options, one at a time
Eight stores.
Each has a context.
For every option: when to choose it, what to watch for, how it grows, and which plan carries it. Sizes and limits are taken from the Laravel Cloud docs; prices live there too and are not repeated here.
Laravel Serverless Postgres
Managed Postgres, powered by Neon. Compute moves within a range you set and storage grows with your data. It supports pgvector for embeddings.
- Choose it when
- Load is uneven or unknown, you want Postgres features such as pgvector, or the database should sleep with the rest of a quiet environment
- Watch for
- Connection limits under many replicas and workers: connect through the pgbouncer pooler (append -pooler to the host). Changing the compute range restarts compute and drops connections.
- Scales by
- Compute units from 0.25 (1 GB RAM) to 10 (40 GB), autoscaled within your range; storage autoscales; up to 500 databases per cluster; pooler up to 10,000 connections
- Available on
- Every plan. Compute ceiling: Starter up to 1 CU, Growth up to 4, Business up to 16. Also offered to Private Cloud customers, outside the dedicated VPC.
Source: Serverless Postgres docs, Plans and Pricing, laravel.com/cloud/private-cloud FAQ.
Laravel MySQL
Managed MySQL in two compute classes. Flex is light and, since July 2026, can scale to zero. Pro is always on, for sustained heavy use.
- Choose it when
- The application already speaks MySQL, you want a fixed size you can reason about, or you want Flex to sleep on a side project and Pro to hold steady in production
- Watch for
- Connections follow RAM: 51 at 512 MB up to 3,264 at 32 GB, though the proxy multiplexes reads up to 10,000. Resizing compute restarts the cluster, under a minute. Binary logs count toward storage.
- Scales by
- Choosing a larger Flex or Pro size; storage from 5 GB to 1,000 GB, auto-upsized by 50% at 90% full, at most once every six hours, never shrinking; a maximum can be set
- Available on
- Every plan. Starter is Flex only; Growth and Business have Flex and Pro. Managed database users (read-only or read-write, scoped to schemas) since September 2026.
Source: Laravel MySQL docs; Monitor 472 (scale to zero, 20 July 2026), 197 (storage auto-upsize, 30 September 2026), 467 (database users, 3 September 2026).
RDS MySQL and Postgres
Managed Amazon RDS, provisioned inside your Private Cloud's dedicated VPC. Read replicas, Multi-AZ failover and point-in-time recovery.
- Choose it when
- The database must sit inside your isolated network, you need a standby in a second availability zone, or reads outgrow a single primary
- Watch for
- Creation takes up to 20 minutes. A compute change restarts the instance; on Multi-AZ it fails over, typically 60 to 120 seconds. Read replicas need PITR of at least a day. Storage never shrinks.
- Scales by
- m7g or m8g instances from large (8 GiB) to 16xlarge (256 GiB); up to 15 read replicas; storage 20 GB to 1,000 GB; PITR up to 30 days
- Available on
- Private Cloud only, since March 2026. Pricing by quote. Managed database users supported.
Source: RDS docs; Monitor 494 (RDS replicas, Postgres and MySQL on Private Cloud, 2 March 2026), 467.
PlanetScale Postgres Through Laravel
A partnership offer, not a resource in the Cloud dashboard. Taylor Otwell has said Laravel works directly with PlanetScale to bring its Postgres to Private Cloud customers running Postgres at scale.
- Choose it when
- You are a Private Cloud customer with a very large or very busy Postgres estate and want PlanetScale's operation of it
- Watch for
- Not self-serve: the Laravel team provisions and bills it, so start with the Cloud team rather than the dashboard (Laravel Cloud team, 8 October 2026).
- Scales by
- PlanetScale's own model; not described in Laravel's documentation
- Available on
- Private Cloud, by conversation with Laravel
Source: Monitor 236 (Taylor Otwell on X, 11 September 2026). Sportmonks runs a 450 GB PlanetScale primary on Private Cloud (Monitor 993).
Laravel Valkey
A managed, Redis-compatible store built by Laravel. One attachment serves Laravel's cache, session, queue and Redis integrations.
- Choose it when
- You need a shared cache or sessions across replicas, rate limiting and locks, or a Redis queue for a team keeping Horizon with its own workers (Horizon does not run on managed queues)
- Watch for
- Each size has a message size limit; an oversized message force-closes the connection. Restarts drop TCP connections, so configure retry and backoff. Redis by Upstash takes no new caches; Valkey is its successor.
- Scales by
- Sizes from 250 MB; automatic upsizing to the next size; up to 10,000 concurrent connections; Flex sizes can opt in to scale to zero, with contents snapshotted to disk
- Available on
- Every plan, generally available since January 2026; automatic upsizing stops at your plan's maximum
Source: Laravel Valkey docs, Queues docs, Plans and Pricing; Monitor 513 (early access, 7 October 2025), 604 (GA, January 2026 releases).
ElastiCache
Managed AWS ElastiCache, in Valkey or Redis OSS, inside your Private Cloud. Replicas, sharding and automatic failover.
- Choose it when
- The cache must live in your private network next to RDS, or a hot path needs read replicas and Multi-AZ failover
- Watch for
- Not publicly reachable: only from inside the Private Cloud network. Creation takes up to 20 minutes. Multi-AZ needs at least one replica.
- Scales by
- cache.r7g node sizes; up to 5 read replicas; Multi-AZ partitions data across shards; online resize with Multi-AZ; daily backups kept 1 to 35 days
- Available on
- Private Cloud only, since June 2026. Pricing by quote.
Source: ElastiCache docs; Monitor 479 (10 June 2026).
Your own database
Cloud has no managed MongoDB and no rule that the data must move. Set the connection variables yourself and the framework connects as it would anywhere.
- Choose it when
- The data is too large or too entangled to move with the application, the engine is one Cloud does not run, or another team owns the database
- Watch for
- Backups, users and upgrades stay yours. Check the network path: Trybe met overlapping CIDR blocks and an IPv6 and NAT64 mismatch, both resolved by the Cloud team.
- Scales by
- The provider's own model
- Available on
- Any plan over the public internet. Private connectivity (VPC peering, PrivateLink, VPC Lattice, Transit Gateway) and static outbound IPs for allowlisting come with Private Cloud.
Source: Private Cloud docs; Monitor 990 (Trybe, MongoDB Atlas by VPC peering), 993 (Sportmonks, PlanetScale and ElastiCache).
SQLite, on your machine
The default in a new Laravel application, and a fine one for local work and tests. It is not supported in Cloud production: the filesystem resets on deploy, sleep and host moves.
- Choose it when
- Developing locally, running a fast test suite, or prototyping before the first deploy
- Watch for
- SQLite is permissive about types and may not enforce foreign keys. Run migrations and representative queries against the engine you will deploy on before you need to.
- Scales by
- It does not, on Cloud. Move to Serverless Postgres (pgloader or psql) or Laravel MySQL.
- Available on
- Local and CI only
Source: SQLite Support; Monitor 5 (Pinkary's move from SQLite to MySQL, 30 September 2026).
Sources: Laravel Cloud docs, database, cache and Private Cloud sections, read 8 October 2026; laravel.com/cloud/private-cloud FAQ; Monitor ids as given in each panel.
If you are
Five situations.
Five starting points.
Starting points, not verdicts. Each can be changed later: a database can be resized, a cache upsized, and an environment can attach a different store on the next deploy.
If you are
Prototyping
Use stores that sleep with the environment. Serverless Postgres or MySQL Flex, a Valkey Flex cache, and Flex compute, all scaling to zero, mean a quiet project is not billed for idle compute. Storage is still billed.
Serverless Postgres or MySQL FlexValkey FlexStarter plan or above. Laravel MySQL and Laravel Valkey docs.
If you are
A steady SaaS
Predictable load rewards a fixed size. MySQL Pro stays awake, so the first query of the morning is never the slow one. Give your reporting tool its own read-only database user instead of the default credentials.
Laravel MySQL ProValkey for cache and sessionsGrowth or Business. Monitor 467.
If you are
Facing spiky traffic
Let compute stretch and protect the database from the spike. Serverless Postgres autoscales within your range and pools connections; cache what can be held even briefly. Sportmonks caches anything static for 10 or 20 seconds.
Serverless Postgres, pooledValkey in front of hot readsPostgres docs; Monitor 993.
If you are
Regulated
Keep the data inside an isolated network you can describe to an auditor. Private Cloud places RDS and ElastiCache inside your dedicated VPC, with Multi-AZ, point-in-time recovery and private links to your own AWS accounts. A HIPAA BAA is available on Private Cloud.
RDS, Multi-AZ, PITRElastiCachePrivate Cloud. RDS and Private Cloud docs; Monitor 494, 479.
If you are
Running very large Postgres
Spread reads across replicas, or talk to Laravel about PlanetScale. RDS Postgres takes up to 15 read replicas on the largest instances; PlanetScale is provisioned and billed through the Laravel team.
RDS Postgres with replicasPlanetScale Postgres Through LaravelPrivate Cloud. Monitor 494, 236.
Laravel MySQL and Serverless Postgres are also offered to Private Cloud customers, but they are not provisioned inside the dedicated VPC; RDS and ElastiCache are (laravel.com/cloud/private-cloud FAQ).
One query, one sleeping database
Storage stays awake.
Compute sleeps.
Cloud
A request arrives
The application opens a connection to DB_HOST exactly as it always does. Nothing in the code knows the database was asleep.
The proxy holds the connection
The connection waits at the proxy layer while the cluster's compute resumes.
The first query answers
Waking adds a few hundred milliseconds to that first query, more as the cluster's RAM grows. The queries after it run at normal speed.
Backups keep their schedule
Scheduled backups and manual snapshots wake the cluster briefly, then let it return to sleep. Pro sizes never sleep.
Sources: Laravel MySQL docs, Scale-to-Zero; Monitor 472 (20 July 2026). Serverless Postgres and Valkey Flex sleep too (their docs). Application compute wakes in under 500 ms on the current Flex sizes only; legacy Flex sizes take 5 to 20 seconds (Compute docs).
How teams chose, in their own figures
Four teams.
Four data layers.
Each figure below is the customer's own, from their published story. Two kept a database Cloud does not run; one learned what SQLite had been forgiving.
Trybe
1.6 TBof MongoDB kept on Atlas, reached by VPC peering- Laravel opened an Ireland region so the app could sit next to it
- A DNS change forced IPv4 for Atlas
Sportmonks
24,000database queries a second at peak, on a 450 GB PlanetScale primary- Three billion cache events on ElastiCache for the API product
- Cursor pagination stored in ElastiCache replaced costly offset queries
Redberry
150,000visitors in four hours, on one managed PostgreSQL database and one Redis cache- 63,622 submissions in the same four hours
- Compute: Pro 2 vCPU, 2 GB, scaling to two instances
Pinkary
5+ hoursof live streams to move from SQLite to managed MySQL- UUIDs had sat in integer columns for over a year
- A grouped query failed under ONLY_FULL_GROUP_BY
Eloquent stays the same.
The store fits the work.
Three questions before you choose: how many hours a day does this database actually do work; which data could you not afford to lose a second of; and which store is so large or so shared that it should stay where it is.