A VCL change refuses to activate
A snippet that worked on a previous VCL version, a typo in a subroutine hook, or a conflict between a boilerplate snippet and a custom one that references the same variable.
VCL compiles and activates on first try
Running on Fastly?
Fastly rewards teams that treat the edge as code and punishes the ones that treat it as a dashboard. The usual trouble is a VCL change that compiled last week and will not now, a surrogate key purge that leaves a stale object on a different POP, or a Compute@Edge service that builds locally and refuses to activate. We read the service version diff, the compile output and the real-time log stream until the cause is on screen.
What it is
Fastly is a content delivery and edge compute network launched in 2011, built on Varnish-derived VCL at the edge rather than the simpler rule-based config of its competitors. It runs an any-cast network of points of presence around the world, with instant surrogate key purging as one of its early differentiators.
On top of the cache sits Compute@Edge, a WebAssembly runtime that lets teams write edge code in Rust, Go, JavaScript and AssemblyScript and deploy it in seconds. The platform is popular with news, commerce and streaming workloads that need both aggressive caching and programmable behaviour, and is often chosen by teams who have outgrown a simpler CDN or Cloudflare's rule model.
The honest version
No technology is all upside. This is where Fastly earns its place and where it tends to cause trouble.
Sound familiar?
The Fastly problems that bring people to us, and what you have once each one is fixed.
A snippet that worked on a previous VCL version, a typo in a subroutine hook, or a conflict between a boilerplate snippet and a custom one that references the same variable.
VCL compiles and activates on first try
An API token that was created with narrower scopes than the client assumes, or a token that was rotated and never updated in the CMS or publishing pipeline.
Purge works from the right systems
A WebAssembly binary that grew past the service limit because of a heavy dependency, a static asset that was bundled by accident, or a release build that was never stripped.
Module under limit, deploys cleanly
A POP that briefly dropped the purge, or surrogate keys that are different between the publisher and the fetcher, leaving one page refreshed and another unchanged.
Content consistent across every POP
A healthcheck that is too strict for a slow but working origin, an origin that returns the wrong status on the health URL, or a probe interval set too aggressively.
Healthcheck honest, failover sensible
A CAA record missing, DNS not yet delegated, or a certificate that was generated with the wrong SAN set and has to be reissued before activation will succeed.
Domain live on TLS first time
A Rust unwrap on an unexpected request, or a Go module hitting a nil pointer under traffic that never appeared in staging.
Panics caught and logged
A shield POP that was moved, or a cache-vary configuration that splits the key differently from what the origin expects, defeating the shield entirely.
Shield hit rate back where it should be
What we do
Get us on a stuck activation, or hand over the entire edge configuration.
How it works
Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.
We fix the cause, not the symptom. If something is down, getting it back up comes first.
Backups, monitoring, updates and access, so the same problem can't quietly come back.
You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.
Often found together
We work on the whole stack, so the problem can be wherever it actually is.
Usually a snippet that references a variable in the wrong subroutine, a boilerplate directive that was removed in a newer Fastly VCL version, or two custom snippets that each try to set the same response header. The error points at the line but not always the cause, and the quickest path is to diff the service against the last version that compiled.
Either the purge was accepted on a token that is scoped out of some surrogate keys, two publishers are emitting different surrogate keys for the same object, or a rare POP returned a 503 on the purge and nobody retried. Real-time logs show which POPs confirmed the purge; the fix is usually tightening key emission on the publishing side.
The CDN parts move cleanly if you port rules into VCL rather than translating them one to one. The harder parts are WAF semantics, Workers that must become Compute@Edge functions, and DNS cutover windows. We do these with a parallel setup on a test hostname first, then flip TTLs and DNS on a scheduled window.
When the work is small, latency sensitive, and benefits from running next to the cache: auth checks, header rewrites, feature flags, and personalised fragment assembly. Heavy work, database access and anything that needs consistent state is better left at the origin or in a backend service.
Bandwidth is the lever, not requests. Tune image optimiser quality, enable Brotli, set sensible TTLs so repeat views are hits, and think hard before running video through the same account without dedicated pricing. We give a before and after on the invoice as part of the engagement.
A written estimate is free. Share the service, the traffic shape and the issue you are seeing, and a number comes back before any work is booked.
Free estimate
Tell us what you're running and what's going wrong. A person reads it and comes back with an estimate.
Your message is with a real person, and we'll get back to you soon.
Rather talk now? Call us: