Server functions
Server functions let app owners run a little JavaScript on the server: on demand from a page, on a schedule, after data changes or from a webhook. They’re on by default. Each run happens in a fresh WebAssembly sandbox inside Bunker, with a memory cap and a time limit, and acts with the rights of the person who triggered it (or the app’s owner, when the function is marked elevated).
Everything on this page is in Admin → Functions, and in the API as admin.functions.get and admin.functions.update (GET / PUT /api/v1/admin/functions).
On or off
Section titled “On or off”The main switch turns server functions off for the whole server. While it’s off:
- apps still load and pages keep working; only function calls, schedules, write triggers and webhooks that target a function stop,
bunker.fnthrowsfunctions_offwith a plain message,- function files are still never served to browsers.
Turning it back on resumes everything, with no redeploy.
Limits
Section titled “Limits”| Setting | Default | Range | |
|---|---|---|---|
memoryMb |
64 |
16–1024 | Memory per run |
timeoutMs |
30000 |
1000–300000 | Time per call from a page, the CLI or a webhook |
triggerTimeoutMs |
120000 |
1000–900000 | Time per schedule or write trigger run |
siteConcurrency |
4 |
1–256 | Runs at once for one app |
globalConcurrency |
4 or twice the CPU count, whichever is more | 1–1024 | Runs at once on the whole server |
Over capacity, calls wait at most 150 ms and then fail with 429 busy instead of queueing, so one busy app can’t slow down the rest. These are fixed for every run: input and output 1 MB each, 64 KB of logs (the first 8 KB is kept in the run history), 1,000 calls to Bunker and fetch responses up to 5 MB. Each app keeps its last 200 runs and up to 100 secrets of 16 KB each.
Outbound policy
Section titled “Outbound policy”Functions have a fetch for calling other services. You choose what it can reach:
outbound |
Functions can reach |
|---|---|
internet (default) |
Public hosts. Private (RFC 1918, unique-local), loopback, link-local and cloud-metadata addresses are always refused, checked on the address actually dialed after DNS, with the same guard as connections |
allowlist |
Only the hosts in allowlist: exact names (api.stripe.com) or a suffix (*.example.com). Private addresses are refused even when listed |
off |
Nothing. Functions can still use connections you define |
Every request the policy refuses is audited as functions.outbound.refused with the app, the person and the host. (Requests to private addresses are refused too, with an error, but not audited.)
internet is the default because the most common reason to write a function is calling a public API with a key. If your network policy says servers mustn’t call out freely, choose allowlist or off. Connections are always the better fit for internal APIs: you keep the key and the base URL, and they work the same from pages.
bunker api admin.functions.update --data '{"outbound":"allowlist","allowlist":["api.stripe.com","*.github.com"]}'Block a person or an app
Section titled “Block a person or an app”Sometimes one app misbehaves, or you’d like to pause someone’s server code while you talk to them. Block it instead of turning everything off:
- An app: only that app’s functions stop.
- A person: functions stop in every app they own, including ones they create later, until you unblock them.
Give a reason. It’s shown to the people affected, so write it for them: “Paused while we review the nightly export job. Talk to IT.”
What they see:
- a banner on their dashboard with your reason, and a badge on each affected app,
functions_offerrors frombunker.fnthat include the reason,- the app itself still loads, with its data and files; only function calls and triggers stop.
Viewers of a blocked app see the plain error message from the app, nothing more.
| Operation | Route |
|---|---|
admin.functions.blocks |
GET /api/v1/admin/functions/blocks → { items: [{ kind, id, label, reason, by, at }] } |
admin.functions.block |
PUT /api/v1/admin/functions/blocks/{kind}/{id} with { reason }. kind is user (a user ID) or site (an app address) |
admin.functions.unblock |
DELETE /api/v1/admin/functions/blocks/{kind}/{id} |
bunker api admin.functions.block --kind site --id lunch-orders --reason "Too many calls to an outside API. Talk to IT."bunker api admin.functions.unblock --kind site --id lunch-ordersAny signed-in person can check their own state with functions.status (GET /api/v1/functions/status), which is what drives the banner.
Pinning settings
Section titled “Pinning settings”Every setting here can be pinned from the environment, like any saved setting (how): BUNKER_FUNCTIONS_ENABLED=false, BUNKER_FUNCTIONS_OUTBOUND=allowlist, and so on. A pinned value wins over the dashboard.
Audit events
Section titled “Audit events”| Event | When |
|---|---|
functions.deployed |
A publish changed an app’s set of functions. meta lists the names and which are elevated, scheduled or write-triggered |
functions.secret.set, functions.secret.delete |
An editor stored, replaced or removed a secret (the name only, never the value) |
functions.settings |
An admin changed the switch, limits or outbound policy |
functions.block, functions.unblock |
An admin blocked or unblocked a person or an app, with the reason |
functions.outbound.refused |
A function’s fetch was refused by the outbound policy (meta.policy) |
Search them with event=functions.* in the audit log.
What to keep an eye on
Section titled “What to keep an eye on”- Elevated functions act as the app’s owner. They’re shown with a badge in each app’s Functions tab, and
functions.deployedtells you when one appears. - Secrets belong to each app and can’t be read back, not even by admins. When someone leaves and their apps move to a colleague, the secrets move with the app.
- Outbound traffic is where data can leave. If that matters to you, use
allowlist. The threat model covers what’s left.