Run the destination engine
SQLite had quietly stored UUID strings in integer columns made by foreignId() for over a year. MySQL refused them. Run migrations and real queries against the database you are moving to.
LaravelCloud
The 500ms wake
A migration story
The servers worked.
The team had other work.
Eight teams moved Laravel applications to Cloud from Forge, Heroku, AWS and servers of their own. None of them was escaping a broken host. Each decided which parts of running an application it no longer wanted to do itself.
Read the ledger
The idea in one line Moving to Cloud is a choice to stop operating, not an escape.
The operating ledgerEight teams · one list of duties
Sizing a server for the launch spikeFilament
Patching the OS, PHP and MySQLCMS Max
Baking a machine image per deployDropInBlog
Working round a ten-minute schedulerRoomies
The weekend check-in with the hostSportmonks
The schema, the queries, the jobsEvery team
Every duty above is named in that team's own published story.
Your code · your call
What the teams stopped operating
Same duties.
A different name beside them.
Read across a row. On the left, the work a team did to keep a Laravel application running. On the right, who does it now. The application code on each side is the same.
Before: the team's list
Work that sat with a small engineering team, or with one contractor, on top of building the product.
CapacityServer stepped up $25, $50, $100 a month, and could not step back down (Filament)
DeploysA machine image built by hand on large releases, dropping the primary for seconds (DropInBlog)
PatchingOS, PHP and MySQL updates, WAF rules, bot filtering (CMS Max)
SchedulerA workaround for a ten-minute limit (Roomies)
On callA check-in with the hosting contractor most weekends (Sportmonks)
Who operates it
On Cloud: the platform's list
The same duties, now fulfilled by the platform. The team still owns the decisions about them.
CapacityOne or two replicas on a normal day, up to ten at a spike, then back down
DeploysZero-downtime deploys; nobody bakes images by hand
PatchingThe platform's runtime, DDoS protection and TLS
SchedulerRuns every minute, as Laravel intends
On callNo infrastructure failures since going live, in Sportmonks' account
Forge is still the right home for a team that wants its servers.
Forge and Laravel VPS are Laravel's own, and a team that wants root, a fixed topology or its own cloud account is well served there. Pinkary's maintainer put it plainly: Forge was not the problem.
Sources: Monitor 997 (Filament), 1186 (DropInBlog), 1003 (CMS Max), 994 (Roomies), 993 (Sportmonks), 5 (Pinkary). Each row is the customer's own account, from its published story.
Where they came from
Three starting points.
One reason.
Each starting point has real strengths, and the teams said so. What they had in common was a short list of duties they no longer wanted on their own plate.
From Forge
Forge gave these teams control of their own servers, and Dan Harrin notes that Forge now has one-click Octane too. They moved because they wanted the platform, not themselves, to answer for scaling and configuration.
- Who
- Filament, DropInBlog, Pinkary, Freek Van der Herten's sites
- What they ran
- A server per app, database alongside; DropInBlog had Forge provisioning autoscaling EC2 on AWS
- What they stopped
- Sizing for spikes, image builds, the autoscaling layer Forge did not handle at the time
- What stayed
- The application, and for DropInBlog its 105 GB PlanetScale database, left where it was
Monitor 997, 1186, 5, 730.
From Heroku
Both teams ran on Heroku and moved their Laravel applications to Cloud, one of them to Private Cloud.
- Who
- Superscript, Roomies
- What they ran
- Dynos, Heroku Postgres and Redis; custom buildpacks for monitoring; preview environments repackaged by hand for a monorepo
- What they stopped
- Paying for idle preview environments, the scheduler workaround, Laravel-specific hacks
- What stayed
- Postgres to Postgres for Superscript; Roomies moved its data with PlanetScale's sync utility
Monitor 998, 994.
From AWS or your own servers
These teams assembled and ran their own infrastructure. Sportmonks kept AWS services such as ElastiCache, now beside its apps in Private Cloud.
- Who
- CMS Max (AWS), Sportmonks (self-managed servers on UpCloud, run by one contractor)
- What they ran
- Their own servers, hand-written autoscaling, custom caching, 150 to 200 named Horizon queues
- What they stopped
- Outage nights, patching, the weekend check-in
- What they gave up
- Sportmonks is candid: the platform is opinionated, and the team had to let go of building its own infrastructure
Monitor 1003, 993, 267.
Heroku's enterprise contract announcement is reported in both customer stories (Monitor 998, 994). DropInBlog's remark that Forge did not handle autoscaling refers to its setup at the time.
The honest part
The new home shows you
what the old one let slide.
Schema
Say which row you mean
A feed query grouped by IFNULL(root_id, id) failed under ONLY_FULL_GROUP_BY. The rewrite ranks rows with ROW_NUMBER() and breaks ties by id.
Make ordering explicit
Second-level timestamps tied, and tests that assumed they would not began to fail. A secondary order by id fixed production; clock helpers fixed the tests.
Use the default disk
Hardcoded Storage::disk('public') calls ignored the new default. On object storage, every exists() check became a network request.
Copy data with a small command
A dump-and-convert path mangled UUID-like values. One-off Artisan commands that checked columns and counted rows did the job.
“The environment exposes every shortcut the old one let you get away with.”Pinkary core maintainer, Laravel blog, 30 September 2026 · Monitor 5
Source: Monitor 5 (Laravel blog, "Migrating Pinkary from Laravel Forge to Laravel Cloud") and 724 (Laravel on X). The team says Pinkary felt faster afterwards but ran no controlled benchmark, so no speed figure is claimed here.
Their results, in their figures
Eight teams.
Eight ledgers.
One move first, step by step, then each team's own figures. They are not averages and not a promise. Where a team said the cost went up, that is here too.
01 · Freeze writesAdmin panel only, in maintenanceThe public site stayed up throughout.
02 · ExportA 25 MB database, a 2 GB bucketFrom the Forge server and S3.
03 · ImportInto Serverless Postgres and CloudCredentials injected as env.
04 · CheckOn the preview URLBefore any traffic moved.
05 · Cut overSwap the DNSAbout an hour of admin downtime in all.
06 · Switch onOctane, one toggle2x faster became 3x.
Filament
3xfaster requests, with Octane on- About three hours, two apps, first time on Cloud
- Raw cost is higher: $150 for two weeks, against about $155 a month before. They count it as buying autoscaling and a separate database
DropInBlog
50%lower infrastructure cost- Three weeks from decision to running
- API 200 to 300 ms faster with Octane
- Three million jobs a month on managed queues
Pinkary
2attempts, rehearsed live on stream, cut over at 1 a.m.- SQLite to managed MySQL; local disk to object storage
- Faster by feel, not by benchmark, in the team's words
Freek Van der Herten
~20sites, plus two named ones, in a single day- One site first to learn the process, then the rest in parallel, with AI tools
- His caveat: they share one architecture
Superscript
30%lower infrastructure cost in the first month- Five apps; four in a couple of hours each, the 80 GB one in about six
- They expect 50% as more environments move
Roomies
$1,000 → $200compute a month- Eight sites, about 135 million requests a month
- Two N+1 queries found within 30 minutes of Nightwatch
CMS Max
500+sites live on Cloud, more onboarding- Thousands of dollars a month saved, by the CEO's account; no exact figure given
- No dedicated engineer watching the server
Sportmonks
600MAPI requests a week- 1,000+ jobs a second on match weekends
- No infrastructure failures since going live, in their account
Filament's flow from Monitor 997. Every figure is the customer's own, from its published story or post. Dates are when each story was first seen, on or before 4 October 2026, except DropInBlog (8 October 2026), Pinkary (30 September 2026) and Freek (2 October 2026).
Choose what to operate.
Hand over the rest.
Three questions for a team on its own servers: which line of the ledger would you most like to stop doing; which shortcut is your current environment letting you keep, a local disk or a forgiving database; and which one site would you move first to learn the process.