webhooker

webhooker is a self-hosted webhook proxy and store-and-forward service written in Go by @sneak. It receives webhooks from external services, durably stores them, and delivers them to configured targets with retry support, logging, and observability. Category: infrastructure / web service. License: MIT.

Getting Started

Prerequisites

• Go 1.26.1+ (the version in go.mod)
• Docker (for linting, for the test stage of the CI gate, and for containerized deployment)
• curl, used by script/fetch-assets to download the third-party browser assets, which are not committed (make bootstrap installs it if missing)

golangci-lint is not a prerequisite and must not be installed on the host: script/bootstrap does not install it, and make lint runs the digest-pinned linter image via Dockerfile.lint.

Quick Start

# Clone the repo
git clone https://git.eeqj.de/sneak/webhooker.git
cd webhooker

# Install Go dependencies and the third-party browser assets.
# `make deps` alone is not enough: it only runs go mod download/tidy,
# and the checks below need the fetched assets.
make bootstrap

# Run all checks (test, lint, format check)
make check

# Run in development mode. DATA_DIR defaults to /var/lib/webhooker in
# every environment, so set it (in .env or the shell) to a writable
# directory when running from a clone.
DATA_DIR=./data make dev

# Build Docker image
make docker

Development Commands

make bootstrap  # Install all dependencies (idempotent)
make setup      # Bootstrap + install git pre-commit hook
make assets     # Fetch + verify third-party browser assets
make fmt        # Format code (gofmt + goimports)
make fmt-check  # Fail if gofmt would change anything (writes nothing)
make lint       # Run golangci-lint in Docker (Dockerfile.lint)
make test       # Run tests with race detection
make check      # test + lint + fmt-check (CI gate)
make build      # Build binary to bin/webhooker
make run        # build, then run ./bin/webhooker
make dev        # go run ./cmd/webhooker
make deps       # go mod download + go mod tidy
make docker     # Build Docker image
make hooks      # Install git pre-commit hook that runs script/precommit
make css        # Regenerate static/css/tailwind.css (needs tailwindcss)
make clean      # Remove bin/

Configuration

All configuration is via environment variables. For local development, you can place variables in a .env file in the project root (loaded automatically via godotenv/autoload).

The environment is selected by setting WEBHOOKER_ENVIRONMENT to dev or prod (default: dev). The setting controls several behaviors:

Behavior	dev	prod
CORS	Allows any origin (*)	Disabled (no-op)
Session cookie Secure	false (works over plain HTTP)	true (requires HTTPS)

The CSRF cookie's Secure flag and Origin/Referer validation mode are determined per-request based on the actual transport protocol, not the environment setting. The middleware checks r.TLS (direct TLS) and the X-Forwarded-Proto header (TLS-terminating reverse proxy) to decide:

• Direct TLS or X-Forwarded-Proto: https: Secure cookies, strict Origin/Referer validation.
• Plaintext HTTP: Non-Secure cookies, relaxed Origin/Referer checks (token validation still enforced).

This means CSRF protection works correctly in all deployment scenarios: behind a TLS-terminating reverse proxy, with direct TLS, or over plain HTTP during development. When running behind a reverse proxy, ensure it sets the X-Forwarded-Proto: https header.

All other differences (log format, security headers, etc.) are independent of the environment setting — log format is determined by TTY detection, and security headers are always applied.

Variable	Description	Default
WEBHOOKER_ENVIRONMENT	dev or prod	dev
PORT	HTTP listen port	8080
DATA_DIR	Directory for all SQLite databases	/var/lib/webhooker
DEBUG	Enable debug logging	false
MAINTENANCE_MODE	Report maintenanceMode: true in the healthcheck JSON. It does not change how any request is served — no maintenance page exists	false
METRICS_USERNAME	Basic auth username for /metrics	""
METRICS_PASSWORD	Basic auth password for /metrics	""
SENTRY_DSN	Sentry error reporting DSN	""
RETENTION_SWEEP_INTERVAL	How often the retention reaper and archive sweeper run (Go duration, must be positive)	1h
SESSION_IDLE_TIMEOUT	Idle session timeout (Go duration)	24h
RECEIVER_RATE_LIMIT	Receiver requests/minute per IP per entrypoint (10x that per IP across the route)	120
TRUSTED_PROXIES	CIDRs whose forwarded headers are trusted (unset: all clients behind a proxy share one rate-limit bucket; a correct login password is never throttled either way)	"" (none)

Trusted proxies

TRUSTED_PROXIES is a comma-separated list of CIDR blocks (a bare address such as 192.168.1.7 is accepted and treated as a single host), for example 192.168.1.7, 2001:db8::5. It decides whose X-Forwarded-For header the rate limiters believe, so it should name the addresses of your reverse proxies and nothing else.

X-Forwarded-For is honoured only when the connecting peer is inside one of these blocks; for every other peer the client identity is the connection's own address and the header is ignored. The default is the empty list, which trusts nobody — anything else would let any client pick its own rate limit bucket, minting a fresh one per request or draining someone else's. Set it to the address of your reverse proxy, and to nothing wider. A set but unparseable value aborts startup.

That default is safe against forged headers, but leaving it unset in production has a cost you must know about. Production runs behind a TLS-terminating reverse proxy, so with TRUSTED_PROXIES unset every request keys on the proxy's own address and all clients share a single bucket per limit. The receiver limits become service-wide ceilings, and the login endpoint's failure counting collapses onto one key, so a stranger's wrong passwords throttle every other client's wrong passwords.

What it cannot do is lock the operator out. The login endpoint verifies credentials before it consults any limit and charges only failures, so a correct password is never throttled no matter how full the bucket is. See Rate Limiting.

The remedy is to set TRUSTED_PROXIES to your reverse proxy's address, which restores per-client buckets. webhooker logs a warning at startup whenever TRUSTED_PROXIES is empty, in every environment — not only when WEBHOOKER_ENVIRONMENT=prod, because that variable defaults to dev and an operator who never set it is precisely the one at risk. The warning is informational when nothing proxies to the process: with no proxy in front, the peer address is the client's own and the buckets are already per-client. See Rate Limiting for what each limit shares.

X-Real-IP and True-Client-IP are never read, from any peer. Reverse proxies append to X-Forwarded-For but forward other client headers verbatim, so a single-valued header is client-controlled even behind a trusted proxy.

Within a trusted request, X-Forwarded-For is read right to left, because the rightmost entry is the one the nearest proxy appended and everything left of it may have been written by the client. The first hop that is not itself a trusted proxy is taken as the client. A hop that is not a bare IP address — ip:port, a bracketed IPv6 literal, the token unknown — ends the walk and the peer address is used instead, since past such an entry the chain is not the shape assumed here. The peer address is likewise used when the header is absent or every hop in it is a trusted proxy.

Two operator requirements follow:

• Your proxy must append the peer address to X-Forwarded-For (nginx $proxy_add_x_forwarded_for, HAProxy option forwardfor, Caddy and AWS ALB by default), and must append a bare address with no port.
• List proxy hosts only. Any address inside TRUSTED_PROXIES chooses its own rate-limit key: its X-Forwarded-For is walked, so it can name a different address on every request to get a fresh bucket each time, or name another client's address to drain that client's bucket. Never list a block that also covers clients — a broad 10.0.0.0/8 on a network where clients live in the same range makes all three limits, including the unauthenticated webhook receiver, silently bypassable by every client in the block.

Sessions

Sessions are bounded by two independent clocks, and end at whichever one runs out first:

• Idle expiry (SESSION_IDLE_TIMEOUT, default 24h) is a sliding window. Every authenticated request pushes it forward, so a session in continuous use never hits it, while an abandoned one expires a day after its last use. Any non-positive value (0, or a negative duration such as -1s) disables idle expiry entirely; the absolute cap below still applies. A set-but-unparseable value aborts startup rather than silently falling back to the default.
• Absolute expiry is a fixed 7 days from login. Activity does not extend it: after a week, every session ends and the user authenticates again.

Only requests that authenticate with the session count as activity, so an unauthenticated request carrying the cookie cannot keep a session alive. The idle timestamp is rewritten at most once per tenth of the idle window rather than on every request, which means a session may expire up to 10% early relative to the user's true last request, but never late.

Both clocks are anchored by timestamps stored in the session cookie. Sessions issued before this feature existed carry neither, so they are treated as expired: upgrading to a build that has it logs every existing session out once, and those users sign in again.

Invalid values abort startup

The defaults above apply only to variables that are unset (or set to an empty string). A variable that is set but cannot be parsed is a fatal configuration error: webhooker logs the offending variable and its value and refuses to start, rather than silently running with a substituted default. PORT=eighty, DEBUG=ture, and RETENTION_SWEEP_INTERVAL=1 hour all abort startup. PORT must additionally be a number in the range 1–65535, RECEIVER_RATE_LIMIT must be at least 1, RETENTION_SWEEP_INTERVAL must be greater than zero (it is a ticker period, so 0s or a negative value would crash the reaper after startup), and every entry in TRUSTED_PROXIES must be a CIDR block or a bare IP address. SESSION_IDLE_TIMEOUT is the exception: a non-positive value there means idle expiry is disabled, not invalid.

Boolean variables (DEBUG, MAINTENANCE_MODE) accept exactly the spellings Go's strconv.ParseBool accepts — 1, t, T, TRUE, true, True, 0, f, F, FALSE, false, False — and nothing else. yes, on, and off are rejected rather than quietly treated as false.

On first startup, webhooker automatically generates a cryptographically secure session encryption key and stores it in the database. This key persists across restarts — no manual key management is needed.

On first startup, webhooker creates an admin user with a randomly generated password and logs it to stdout. This password is only displayed once.

Running with Docker

docker run -d \
  -p 8080:8080 \
  -v /path/to/data:/var/lib/webhooker \
  -e WEBHOOKER_ENVIRONMENT=prod \
  webhooker:latest

The container runs as a non-root user (webhooker, UID 1000), exposes port 8080, and includes a health check against /.well-known/healthcheck. The /var/lib/webhooker volume holds all SQLite databases: the main application database (webhooker.db), the per-webhook event databases (events-{uuid}.db), and any archive databases written by database targets (archive-{uuid}.db). Mount this as a persistent volume to preserve data across container restarts.

Entrypoints

This repository adheres to the Scripts to Rule Them All standard: normalized scripts in script/ are the entrypoints for the development workflow. Ten of the Makefile's sixteen targets are thin shims that call them; build, run, dev, deps, clean and css are inline commands with no script behind them. We provide:

• script/bootstrap — install all dependencies (idempotent)
• script/setup — make a fresh clone ready for development (bootstrap, then install-precommit)
• script/projectname — output the project name ("webhooker")
• script/fetch-assets — download the third-party browser assets into static/, verifying each against its pinned sha256
• script/test — run the test suite
• script/lint — run golangci-lint in Docker (see Linting below)
• script/fmt — format all code (writes)
• script/fmt-check — check formatting (read-only)
• script/check — run test, lint, and fmt-check
• script/docker — build the Docker image tagged via script/projectname
• script/cibuild — CI entrypoint: docker build . (the Dockerfile runs the checks, so a green build implies a green repo)
• script/ci-mark-superseded — CI helper: mark the commits whose run a newer push cancelled (see CI gate honesty)
• script/precommit — pre-commit checks (go mod tidy guard, then script/check)
• script/install-precommit — install the git pre-commit hook that runs script/precommit

Third-party browser assets

The web UI serves one third-party script, Alpine.js. It is not committed: a minified bundle in the tree is unreviewable, and REPO_POLICIES.md bars both committed build artifacts and unpinned external references.

Instead script/fetch-assets downloads it from a pinned URL, checks the download against a hardcoded sha256, and installs it under static/. The sha256 of every installed asset is recorded in static/vendor.sha256, and static/vendor_test.go re-hashes the bytes go:embed put in the binary against that manifest — so the pin is enforced on what actually ships, not merely written down. Any mismatch fails the build.

make bootstrap runs the fetch for local development, and the Dockerfile runs it in the build stage; .gitignore and .dockerignore keep the artifact out of both the repo and the build context.

To move to a new version: update the version, URL, and tarball sha256 in script/fetch-assets and the asset sha256 in static/vendor.sha256, then run make assets && make check.

Rationale

Webhook integrations between services are inherently fragile. The receiving service must be online when the webhook fires, most webhook senders provide no built-in retry mechanism, and there is no standard way to inspect what was sent, when it was sent, or whether delivery succeeded.

webhooker solves this by acting as a durable intermediary:

1. Reliable ingestion — webhooker is always ready to accept incoming webhooks. It stores every received event before attempting any delivery, so nothing is lost if downstream targets are unavailable.

2. Guaranteed delivery — Events are queued for delivery to each configured target. Failed deliveries are retried with configurable backoff. Every delivery attempt is logged with status codes, response bodies, and timing.

3. Observability — Full request/response logging for every webhook received and every delivery attempted. Prometheus metrics expose volume, latency, and error rates. The web UI provides real-time visibility into event flow.

4. Fan-out — A single incoming webhook can be delivered to multiple targets simultaneously. This enables patterns like forwarding a GitHub webhook to both a deployment service and a Slack channel.

5. Replay (not yet implemented) — Every received event is stored in full, which is what manual redelivery for debugging or testing will be built on. No redelivery exists today, in the web UI or the API; see TODO.md.

Use Cases

• Store-and-forward with configurable retries for unreliable receivers
• Observability via Prometheus metrics on webhook frequency, payload size, and delivery performance
• Debugging and introspection of webhook payloads in the web UI
• Replay of webhook events for application testing and development (planned; not yet implemented)
• Fan-out delivery of a single webhook to multiple downstream targets
• High-availability ingestion for delivery to less reliable backend systems

Design

Architecture Overview

webhooker is structured as a standard Go HTTP server following the sneak/prompts GO_HTTP_SERVER_CONVENTIONS. It uses:

• Uber fx for dependency injection and lifecycle management
• go-chi for HTTP routing
• GORM for database access with modernc.org/sqlite as the runtime SQLite driver. Note: gorm.io/driver/sqlite transitively depends on mattn/go-sqlite3, which requires CGO at build time (see Docker section)
• slog (stdlib) for structured logging with TTY detection (text for dev, JSON for prod)
• gorilla/sessions for encrypted cookie-based session management
• gorilla/csrf for CSRF protection (cookie-based double-submit tokens)
• go-chi/httprate for sliding-window rate limiting of the password-change and webhook receiver endpoints. The bucket is per client IP only when TRUSTED_PROXIES names the reverse proxy; unset, every client behind that proxy shares one bucket per limit. The login endpoint counts failed attempts itself instead, so that a correct password is never throttled (see Rate Limiting)
• Prometheus for metrics, served at /metrics behind basic auth
• Sentry for optional error reporting

Naming Conventions

The codebase uses consistent naming throughout (rename completed in issue #12):

Entity	Description
Webhook	Top-level configuration entity grouping entrypoints and targets
Entrypoint	A receiver URL where external services POST events
Target	A delivery destination for events

Data Model

webhooker's data model has nine entities organized into two tiers: the application tier (user and webhook configuration) and the event tier (event ingestion, delivery, and logging).

┌─────────────────────────────────────────────────────────────┐
│                    APPLICATION TIER                          │
│                  (main application database)                 │
│                                                             │
│  ┌──────────┐       ┌──────────┐       ┌──────────────┐     │
│  │   User   │──1:N──│  Webhook │──1:N──│  Entrypoint  │     │
│  │          │       │          │       │              │     │
│  │          │       │          │──1:N──│   Target     │     │
│  │          │       └──────────┘       └──────────────┘     │
│  │          │──1:N──│  APIKey  │                             │
│  └──────────┘       └──────────┘                             │
│                                                             │
│  ┌──────────┐                                               │
│  │ Setting  │  (key-value application config)               │
│  └──────────┘                                               │
└─────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────┐
│                      EVENT TIER                              │
│            (per-webhook dedicated databases)                 │
│                                                             │
│  ┌──────────┐       ┌──────────┐       ┌─────────────────┐  │
│  │  Event   │──1:N──│ Delivery │──1:N──│ DeliveryResult  │  │
│  └──────────┘       └──────────┘       └─────────────────┘  │
└─────────────────────────────────────────────────────────────┘

Setting

A key-value pair for application-level configuration that is auto-managed rather than user-provided. Used to store the session encryption key and any future auto-generated settings.

Field	Type	Description
key	string	Primary key (setting name)
value	text	Setting value

Currently stored settings:

• session_key — Base64-encoded 32-byte session encryption key, auto-generated on first startup.

User

A registered user of the webhooker service.

Field	Type	Description
id	UUID	Primary key
username	string	Unique login name
password	string	Argon2id hash (never exposed via API)

Relations: Has many Webhooks. Has many APIKeys.

Passwords are hashed with Argon2id using secure defaults (64 MB memory, 1 iteration, 4 threads, 32-byte key, 16-byte salt). On first startup, an admin user is created with a randomly generated 16-character password logged to stdout.

Webhook

The top-level configuration entity. A webhook groups together one or more entrypoints (receiver URLs) and one or more targets (delivery destinations) into a logical unit. A user creates a webhook to set up event routing.

Field	Type	Description
id	UUID	Primary key
user_id	UUID	Foreign key → User
name	string	Human-readable name
description	string	Optional description
retention_days	integer	Days to retain events (default: 30; 0 means retain forever)

Relations: Belongs to User. Has many Entrypoints. Has many Targets.

The retention_days field controls how long event data is kept in the webhook's dedicated database before automatic cleanup.

Setting retention_days to 0 means "retain events forever". Because the column carries a default of 30, a literal zero cannot survive an insert, so a zero is rewritten on save to a sentinel of 365 * 1000 days (database.RetentionForeverDays). The retention reaper recognises that sentinel and skips the webhook entirely, and the web UI displays such a webhook's retention as "forever" rather than as a day count.

Submitted retention_days values therefore fall into three bands, not two:

• 1 up to database.MaxFiniteRetentionDays (106751 days, about 292 years) is accepted as a finite retention.
• Above that ceiling but below the retain-forever sentinel of 365000 (database.RetentionForeverDays) is rejected with a 400. This is the band the cap exists for.
• 0, and 365000 or above, are accepted and mean retain forever, collapsing to the sentinel — 0 in Webhook.BeforeSave, the large values in parseRetentionDays. The large values are not out of range: the edit form pre-fills the sentinel for a retain-forever webhook, so submitting that form back unchanged has to keep meaning "forever".

A negative value is in none of the three: parseRetentionDays rejects it with a 400 before BeforeSave ever sees it.

The cap is not arbitrary: the reaper computes its cutoff as a time.Duration, an int64 nanosecond count, and a longer period overflows it. An overflowed cutoff lands in the future, where it matches every row, so the sweep would delete every event the webhook has instead of none. The reaper also clamps the value it is given, so a row written by an older version cannot trigger that either.

Entrypoint

A receiver URL where external services POST webhook events. Each entrypoint has a unique UUID-based path. When an HTTP request arrives at an entrypoint's path, webhooker captures the full request and creates an Event.

Field	Type	Description
id	UUID	Primary key
webhook_id	UUID	Foreign key → Webhook
path	string	Unique bare UUID, generated at creation. The /webhook/ prefix is route only and is not stored: the receiver matches this column against the raw {uuid} path segment
description	string	Optional description
active	boolean	Whether this entrypoint accepts events (default: true)

Relations: Belongs to Webhook.

A webhook can have multiple entrypoints. This allows separate URLs for different event sources that all feed into the same processing pipeline (e.g., one entrypoint for GitHub, another for Stripe, both routing to the same targets).

Target

A delivery destination for events. Each target defines where and how events should be forwarded.

Field	Type	Description
id	UUID	Primary key
webhook_id	UUID	Foreign key → Webhook
name	string	Human-readable name
type	TargetType	One of: http, slack, database, log
active	boolean	Whether deliveries are enabled (default: true)
config	JSON text	Type-specific configuration
max_retries	integer	Maximum retry attempts for http and slack targets (0 = fire-and-forget, >0 = retries with backoff and a circuit breaker). Ignored by database and log targets
max_queue_size	integer	Stored and shown on the target's detail view, but not enforced anywhere yet: nothing in the delivery engine consults it. Queue depth is set by the two fixed 10,000-entry channels

Relations: Belongs to Webhook. Has many Deliveries.

Target types:

• http — Forward the event as an HTTP POST to a configured URL. Behavior depends on max_retries: when max_retries is 0 (the default), the target operates in fire-and-forget mode — a single attempt with no retries and no circuit breaker. When max_retries is greater than 0, failed deliveries are retried with exponential backoff up to max_retries attempts, protected by a per-target circuit breaker.
• slack — Post the event as a formatted message to a Slack-compatible incoming webhook URL (webhookUrl in config). It is built on the same HTTP core as http and honours max_retries identically, circuit breaker included. See the Slack target section under "Per-Webhook Event Databases" for the message format.
• database — Archive the full event as a row into a separate per-webhook archive database (archive-{webhookID}.db) for long-term retention, with an optional creation-validated expiry (default: keep forever). No external delivery and no retries; an archive write failure fails the delivery. See the database target section under "Per-Webhook Event Databases" for the full semantics.
• log — Write the event to the application log (stdout). Useful for debugging.

The config field stores type-specific configuration as JSON (e.g., destination URL, custom headers, timeout settings).

APIKey

A programmatic access credential for API authentication.

Field	Type	Description
id	UUID	Primary key
user_id	UUID	Foreign key → User
key	string	Unique API key value
description	string	Optional description
last_used_at	timestamp	Last time this key was used (nullable)

Relations: Belongs to User.

Event

A captured incoming webhook request. Stores the complete HTTP request data for auditing and for the planned replay capability.

Field	Type	Description
id	UUID	Primary key
webhook_id	UUID	Foreign key → Webhook
entrypoint_id	UUID	Foreign key → Entrypoint
method	string	HTTP method of the captured request. Always POST: the receiver answers every other method with 405 before an Event is created
headers	JSON	Complete request headers
body	text	Raw request body
content_type	string	Content-Type header value

Relations: Belongs to Webhook. Belongs to Entrypoint. Has many Deliveries.

When a request arrives at an entrypoint, the full request (method, headers, body) is captured as an Event. The event is then queued for delivery to every active target configured on the parent webhook.

Delivery

The pairing of an event with a target. Tracks the overall delivery status across potentially multiple attempts.

Field	Type	Description
id	UUID	Primary key
event_id	UUID	Foreign key → Event
target_id	UUID	Foreign key → Target
status	DeliveryStatus	One of: pending, delivered, failed, retrying

Relations: Belongs to Event. Belongs to Target. Has many DeliveryResults.

Delivery statuses:

• pending — Created but not yet attempted.
• retrying — At least one attempt failed; more attempts remain.
• delivered — Successfully delivered (at least one attempt succeeded).
• failed — All retry attempts exhausted without success.

DeliveryResult

The result of a single delivery attempt. Every attempt (including retries) is individually logged for full observability.

Field	Type	Description
id	UUID	Primary key
delivery_id	UUID	Foreign key → Delivery
attempt_num	integer	Attempt number (1-based)
success	boolean	Whether this attempt succeeded
status_code	integer	HTTP response status code (if applicable)
response_body	text	Response body (if applicable)
error	string	Error message (on failure)
duration	integer	Request duration in milliseconds

Relations: Belongs to Delivery.

Common Fields

Every entity except Setting includes these fields from BaseModel. Setting is a bare key-value row with no id, no timestamps and no soft delete:

Field	Type	Description
id	UUID	Auto-generated UUIDv4 primary key
created_at	timestamp	Record creation time
updated_at	timestamp	Last modification time
deleted_at	timestamp	Soft-delete timestamp (nullable; GORM soft deletes)

Database Architecture

Per-Webhook Event Databases

webhooker uses separate SQLite database files: a main application database for configuration data and per-webhook databases for event storage. All database files live in the DATA_DIR directory.

Main Application Database ({DATA_DIR}/webhooker.db) — stores configuration and application state:

• Settings — auto-managed key-value config (e.g. session encryption key)
• Users — accounts and Argon2id password hashes
• Webhooks — webhook configurations
• Entrypoints — receiver URL definitions
• Targets — delivery destination configurations
• APIKeys — programmatic access credentials

On first startup the main database is auto-migrated, a session encryption key is generated and stored, and an admin user is created.

Per-Webhook Event Databases ({DATA_DIR}/events-{webhook_uuid}.db) — each webhook gets its own dedicated SQLite file containing:

• Events — captured incoming webhook payloads
• Deliveries — event-to-target pairings and their status
• DeliveryResults — individual delivery attempt logs

Per-webhook databases are created automatically when a webhook is created (and lazily on first access for webhooks that predate this feature). They are managed by the WebhookDBManager component, which handles connection pooling, lazy opening, migrations, and cleanup.

This separation provides:

• Isolation — a high-volume webhook won't cause lock contention or journal growth affecting the main application or other webhooks.
• Independent lifecycle — event databases can be independently backed up, archived, rotated, or size-limited without impacting the application.
• Clean deletion — removing a webhook and all its history is as simple as deleting one file. Configuration is soft-deleted in the main DB; the event database file is hard-deleted (permanently removed).
• Per-webhook retention — the retention_days field on each webhook controls automatic cleanup of old events in that webhook's database only, or disables cleanup entirely when set to 0 (retain forever).
• Performance — each webhook's database has its own page cache and its own lock, so concurrent event ingestion across webhooks won't contend. No write-ahead log is involved: both DSNs are file:{path}?cache=shared&mode=rwc and no journal_mode pragma is ever issued, so every database runs on SQLite's default rollback journal.

The database target type builds on this architecture to provide long-term archiving, separate from the per-webhook event database (which may prune events under its own retention). Delivering to a database target writes the full event — body, headers, method, content type, and webhook/entrypoint/event identifiers — as a row into a dedicated archive database, archive-{webhookID}.db, stored under the data directory beside the event database. After each write the archive handle is closed and reopened, debounced to at most once per second, so an operator can move the archive file away for offline archiving without stopping the service; a moved or removed archive file is recreated automatically on the next write. An optional expiry in the target's config JSON (e.g. {"expiry":"720h"}) is validated when the target is created — the default (unset or the literal never) keeps rows forever — and rows older than the expiry are pruned each time the archive is (re)opened. An archive write failure is never silent success: the delivery records a failed attempt with the error and is marked failed.

Because reopens only happen on writes, an archive belonging to a webhook that has stopped receiving events would never be pruned. A background archive sweeper closes that gap: on the same interval as the event retention reaper (RETENTION_SWEEP_INTERVAL) it prunes every archive whose database target declares a positive expiry, whether or not the webhook is still receiving traffic. The sweep never creates an archive — a webhook whose archive file does not yet exist is skipped, not initialised — it takes the same per-webhook lock the write path uses, so it can never interleave with a write, and it leaves the archive closed afterwards so the move-the-file-away workflow keeps working. Archives with no expiry, or the expiry never, are not touched by the sweep at all.

Note that a webhook has one archive file but may carry more than one database target, each with its own expiry. The shortest expiry configured on any of them therefore governs the whole archive, and the sweep applies it whether or not the webhook is still receiving events. Configure a single database target per webhook unless you intend that.

Deleting a webhook releases its archive: the delivery engine's cached archive writer is dropped and its file handle closed, so nothing lingers after the webhook is gone. The archive file itself is deliberately left on disk. Unlike the event database — per-webhook working storage that is hard-deleted with the webhook — an archive is long-term storage an operator may still want to keep or move away for offline retention, and destroying it as a side effect of deleting a webhook would be unrecoverable. Removing archive-{webhookID}.db is the operator's call. Deleting a webhook's last database target releases the writer the same way, and for the same reason leaves the file alone.

The Slack target type sends webhook events as formatted messages to any Slack-compatible incoming webhook URL (works with Slack, Mattermost, and other compatible services). Each message includes event metadata (HTTP method, content type, timestamp, body size) and the payload pretty-printed in a code block. JSON payloads are automatically formatted with indentation for readability; non-JSON payloads are shown as raw text. Large payloads are truncated to keep messages reasonable. Config stores webhookUrl — the Slack/Mattermost incoming webhook endpoint. That is the JSON key; the error text for a missing one reads webhook_url is required, which is the message, not the key.

The database uses the modernc.org/sqlite driver at runtime, though CGO is required at build time due to the transitive mattn/go-sqlite3 dependency from gorm.io/driver/sqlite.

Request Flow

External Service
       │
       │  POST /webhook/{uuid}
       ▼
┌─────────────┐     ┌──────────────┐     ┌──────────────┐
│  chi Router │────►│  Middleware   │────►│   Webhook    │
│             │     │  Stack       │     │   Handler    │
└─────────────┘     └──────────────┘     └──────┬───────┘
                                                │
                     1. Look up Entrypoint by UUID
                     2. Capture full request as Event
                     3. Create Delivery records for each active Target
                     4. Build self-contained delivery.Task structs
                        (target config + event data inline for
                         bodies < 16 KiB)
                     5. Notify Engine via channel (no DB read needed)
                                                │
                                                ▼
                                        ┌──────────────┐
                                        │  Delivery    │◄── retry timers
                                        │  Engine      │    (backoff)
                                        │  (worker     │
                                        │   pool)      │
                                        └──────┬───────┘
                                               │
                              ┌── bounded worker pool (N workers) ──┐
                              ▼                ▼                    ▼
                       ┌────────────┐   ┌────────────┐      ┌────────────┐
                       │ HTTP Target│   │ HTTP Target│      │ Log Target │
                       │(max_retries│   │(max_retries│      │ (stdout)   │
                       │   == 0)    │   │   > 0,     │      └────────────┘
                       │ fire+forget│   │ backoff +  │
                       └────────────┘   │ circuit    │
                                        │ breaker)   │
                                        └────────────┘

Bounded Worker Pool

The delivery engine uses a fixed-size worker pool (default: 10 workers) to process all deliveries. At most N deliveries are in-flight at any time, preventing goroutine explosions regardless of queue depth.

Architecture:

• Channels as queues: Two buffered channels serve as bounded queues: a delivery channel (new tasks from the webhook handler) and a retry channel (tasks from backoff timers). Both are buffered to 10,000.
• Fan-out via channel, not goroutines: When an event arrives with multiple targets, each delivery.Task is sent to the delivery channel. Workers pick them up and process them — no goroutine-per-target.
• Worker goroutines: A fixed number of worker goroutines select from both channels. Each worker processes one task at a time, then picks up the next. Workers are the ONLY goroutines doing actual HTTP delivery.
• Retry backpressure with DB fallback: When a retry timer fires and the retry channel is full, the timer is dropped — the delivery stays in retrying status in the database. A periodic sweep (every 60s) scans for these "orphaned" retries and re-queues them. No blocked goroutines, no unbounded timer chains.
• Bounded concurrency: At most N deliveries (N = number of workers) are in-flight simultaneously. Even if a circuit breaker is open for hours and thousands of retries queue up in the channels, the workers drain them at a controlled rate when the circuit closes.

This means:

• No goroutine explosion — even with 10,000 queued retries, only N worker goroutines exist.
• Natural backpressure — if workers are busy, new tasks wait in the channel buffer rather than spawning more goroutines.
• Independent results — each worker records its own delivery result in the per-webhook database without coordination.
• Graceful shutdown — cancel the context, workers finish their current task and exit. The stop hook waits for the pool via lifecycle.WaitForShutdown, which bounds that wait by fx's stop timeout rather than blocking forever on a wedged worker. On timeout it logs at ERROR and returns an error, and the goroutines that did not finish are still running — an unclean shutdown is reported rather than hidden.

Recovery paths:

1. Startup recovery: When the engine starts, it scans all per-webhook databases for pending and retrying deliveries. Pending deliveries are sent to the delivery channel; retrying deliveries get backoff timers scheduled.
2. Periodic retry sweep (DB-mediated fallback): Every 60 seconds the engine scans for retrying deliveries whose backoff period has elapsed. This catches "orphaned" retries — ones whose in-memory timer was dropped because the retry channel was full. The database is the durable fallback that ensures no retry is permanently lost, even under extreme backpressure.

Changing a target's type does not migrate in-flight deliveries. Only http and slack targets own durable retries; database and log targets are fire-and-forget and never produce a retrying delivery. If a target's type is edited from a retrying type to a non-retrying (or unknown) one while one of its deliveries is still retrying, both recovery paths above terminally mark that delivery failed and record a DeliveryResult naming the current target type as the reason, logging it at warn level. The delivery is not re-dispatched under the new type — the operator never asked for that delivery — and while the event itself remains stored in the per-webhook event database, there is no way to redeliver it: manual redelivery is planned, not implemented (see TODO.md).

Circuit Breaker (HTTP and Slack Targets with Retries)

http and slack targets with max_retries > 0 are protected by a per-target circuit breaker that prevents hammering a down target with repeated failed delivery attempts. The circuit breaker is in-memory only and resets on restart (which is fine — startup recovery rescans the database anyway).

States:

State	Behavior
Closed	Normal operation. Deliveries flow through. Consecutive failures are counted.
Open	Target appears down. Deliveries are skipped and rescheduled for after the cooldown.
Half-Open	Cooldown expired. One probe delivery is allowed to test if the target has recovered.

Transitions:

     success              ┌──────────┐
   ┌────────────────────► │  Closed  │ ◄─── probe succeeds
   │                      │ (normal) │
   │                      └────┬─────┘
   │                           │ N consecutive failures
   │                           ▼
   │                      ┌──────────┐
   │                      │   Open   │ ◄─── probe fails
   │                      │(tripped) │
   │                      └────┬─────┘
   │                           │ cooldown expires
   │                           ▼
   │                      ┌──────────┐
   └──────────────────────│Half-Open │
                          │ (probe)  │
                          └──────────┘

Defaults:

• Failure threshold: 5 consecutive failures before opening
• Cooldown: 30 seconds in open state before probing

Scope: Circuit breakers apply to http and slack targets with max_retries > 0. The Slack target is built on the same HTTP core and hands its own max_retries to the same retry path, so it gets a breaker with the same 5-failure / 30-second defaults. Fire-and-forget targets of either type (max_retries == 0), database targets (local operations), and log targets (stdout) do not use circuit breakers.

When a circuit is open and a new delivery arrives, the engine marks the delivery as retrying and schedules a retry timer for after the remaining cooldown period. This ensures no deliveries are lost — they're just delayed until the target is healthy again.

Rate Limiting

Global blanket rate limiting middleware (e.g., a per-IP throttle shared with the web UI) must not apply to webhook receiver endpoints. Webhook endpoints receive automated traffic from external services at unpredictable rates, and blanket limits shared with other routes would cause legitimate deliveries to be dropped.

The receiver instead has its own dedicated abuse limit, scoped to the /webhook/{uuid} route only and keyed per client IP per request path (httprate.KeyByEndpoint): one misbehaving sender is throttled without affecting other senders of the same entrypoint or the same sender's other entrypoints. Keying on the path rather than on the entrypoint matters — see the aggregate limit below. The limit is RECEIVER_RATE_LIMIT requests per minute (default 120, generous for legitimate webhook senders). Requests over the limit receive HTTP 429 with a Retry-After header. A set-but-invalid RECEIVER_RATE_LIMIT value aborts startup rather than silently falling back to the default.

A second limit sits in front of that one, keyed on the client IP alone and covering the whole route at ten times RECEIVER_RATE_LIMIT requests per minute (default 1200). The per-entrypoint limit needs it: the route pattern matches any single path segment, so a client that invents a fresh path per request gets a fresh per-entrypoint bucket every time and would otherwise have no aggregate limit at all — while each of those requests still costs an entrypoint lookup before it 404s. The aggregate limit leaves room for one address to drive several entrypoints at their full rate, and it is not configurable separately.

What that aggregate limit bounds is the database work an invented path costs; log volume it caps rather than eliminates. A path that names no entrypoint is recorded by the handler at DEBUG, and the aggregate limiter logs its own rejections at DEBUG and without the path, so neither appears at all under the default level. The per-entrypoint limiter is the loud one: it logs every rejection at WARN with the request path, which on this route is attacker-controlled text. A client hammering a single invented path is served RECEIVER_RATE_LIMIT requests and has the rest of its aggregate budget rejected there, so the aggregate limit is what bounds the number of those WARN lines — to under ten times RECEIVER_RATE_LIMIT per minute per client IP, 1080 at the defaults, where before it there was no bound at all. Their width is bounded by the field budgets below, the same ones the access log spends. The access log is bounded by neither limit: every request is recorded once at INFO, served or rejected alike.

What the access log does bound is the content of those lines. A 3xx or 4xx response logs the chi route pattern — /webhook/{uuid}, /user/{username}//, or the literal (unmatched) when the request hit no route at all — in place of the concrete URL. Those are the outcomes an unauthenticated client can drive for free: 404 and 429 on any invented receiver path, a login redirect on any invented profile path. Logging the URL there would let a flood write text of its own choosing, at a length of its own choosing, into the log. 2xx and 5xx responses keep the concrete path — a success resolved against a static route or against the operator's own data (on the receiver, against a stored entrypoint UUID), and a 5xx is a bug in this service, where the exact path is the evidence and no client can provoke one at will.

The query string is never logged; it is replaced by the fixed marker ?(redacted). It is client-chosen on every route, and /.well-known/healthcheck and /s/* answer 200 to anyone with no rate limiter in front of them, so a query on a fixed 200 URL would otherwise buy the same amplification as an invented path. Nothing debuggable is lost: page, on the authenticated pagination links, is the only query parameter this service reads.

Client-supplied request content does not leave the host by the other route either. The Sentry SDK attaches the request to every event it captures, independently of the access log, and SendDefaultPII=false does not cover all of what it copies: the raw query string and the first 10 KiB of the request body are both taken unconditionally, the body precisely because these handlers call ParseForm. A BeforeSend hook therefore replaces the query string and the body with (redacted), drops cookies and the remote-address environment, and reduces the headers to a fixed allowlist — Accept, Content-Length, Content-Type, Host, Origin, Referer, User-Agent and X-Request-Id.

The same hook rewrites the request URL. The SDK builds it as scheme://host/path from the concrete path, which on the receiver route is /webhook/<uuid> in full — and that UUID is a write capability, not an identifier: anyone holding it can post events this service accepts and its targets then deliver. A tracker has its own retention, access control and deletion policy, so the rule the access log follows above does not carry across that boundary. What is sent is the chi route pattern instead: http://host/webhook/{uuid}.

The scheme and the host are kept, and everything else in the URL is discarded rather than edited, so a future SDK version that starts appending a query string cannot widen this. The scheme has to survive for the reason given below. The host is whatever the request's Host header carried — this service validates no hostname, so on a directly exposed deployment a client sets it — and that same header is on the allowlist above, so scrubbing the host out of the URL would withhold nothing that is not sent anyway.

The body, the query string and the URL are all handled on every route rather than filtered by route. For the URL that is also what keeps the event locatable: an error event is grouped by its exception and stack trace, not by its URL, so replacing the path with the pattern costs no grouping and the pattern still names the route in the UI. And an unconditional rule cannot leak on a route somebody forgets to add to it, which a route-conditional one can. For the body there is a second reason: nothing debuggable is lost, because every handler reads its fields with PostFormValue, so the body is exactly where the credentials are — the target destination URL, the login password, both password-change fields — and the one route whose body is genuine signal is the receiver, whose body is already stored on the event and served from the UI, so a tracker is not where anyone reads it.

The route is reachable from the hook only on the error dispatch. sentryhttp's recover path puts the request on the context it hands to RecoverWithContext, and the SDK carries that context through to BeforeSend as hint.Context, so hint.Context.Value(sentry.RequestContextKey) yields the live request and chi's RoutePattern() yields the matched pattern off it. The transaction dispatch has no such request: a finished span captures with a nil hint, which the client replaces with an empty one, so BeforeSendTransaction sees no context at all. Tracing is off in this service, so no transaction event is produced today, but the hook is installed on both dispatches as a floor.

Where the pattern is out of reach — the transaction dispatch, an event captured outside the router, or a request that matched no route — the fallback is never the concrete path. The path becomes the literal /(redacted), so the URL reads http://host/(redacted); a URL the rewrite cannot parse into a scheme is withheld whole. A transaction event additionally carries the SDK's own METHOD /path name, built from the concrete path as well; it is rewritten on the same terms, to POST /webhook/{uuid} where the pattern is known and POST /(redacted) where it is not.

The headers are an allowlist for the same reason the rules above are unconditional: the SDK's own filter removes four names and passes everything else, which would ship X-CSRF-Token and the shared secrets senders put on the receiver route. What survives still names the failing route — scheme, host, route pattern, method — and X-Request-Id ties the event to the local access log line that holds the rest. Nothing dropped is needed for the likeliest use, debugging a CSRF rejection. Its three inputs are the TLS decision, Origin and Referer; the latter two are kept, and the first is the scheme of the retained URL, because the SDK derives that scheme from r.TLS != nil || r.Header.Get("X-Forwarded-Proto") == "https" — byte for byte the predicate internal/middleware/csrf.go uses to choose between the csrf.Secure(true) and csrf.Secure(false) handlers. That is what the rewrite above preserves it for, and it is why dropping X-Forwarded-Proto costs nothing. The dropped provider headers (X-GitHub-Event, X-Gitlab-Event and the like) are real signal but are recorded locally on the event, and Sentry-Trace/Baggage are already reflected in the event's trace context.

The remaining client-supplied fields are truncated rather than dropped, each to a fixed budget: 512 bytes for url, useragent and referer, 128 for request_id (chi passes an inbound X-Request-Id header through), and 32 for method. A truncated User-Agent is still worth reading; an absent one is not. A cut value ends in [truncated], which is charged on top of the budget rather than inside it.

Each budget is spent in encoded bytes, not in the bytes the client sent. Every rune is charged what the wider of the two log handlers emits for it: two bytes for a quotation mark, a backslash or a tab; six for a non-printable rune below U+10000; ten for one at or above it, which the text handler spells \UXXXXXXXX. Go's header parser accepts all of them in a header value, so a budget counted raw would buy a field several times its nominal size — and the line, not the header, is what an operator has to store. Plain ASCII encodes one byte for one, so a real browser's User-Agent still fits whole; a value built out of escapes keeps a proportionally shorter prefix, which is the right trade.

Net: one INFO line per request, of at most 2,560 bytes. That ceiling is arithmetic, not an observation: 3 × (512 + 11) for url, useragent and referer, plus 128 + 11 for request_id, plus 32 + 11 for method, plus a 336-byte fixed portion (the field names, the punctuation, both timestamps at their longest, an IPv6 remoteIP with a zone, the status and the latency) — 2,087 bytes, stated at 2,560 so the figure has headroom. internal/middleware/accesslog_test.go asserts it against 8 KB of client-chosen text in the path, in the query, and in each of User-Agent, Referer and X-Request-Id, including cases built from the characters the handlers escape, and against the widest access log line the service can be made to write: a 5xx that keeps its concrete path while all three header fields are also at their budget. Every case runs through both handlers internal/logger can select — the JSON one and the text one it installs on a tty — since the two do not escape alike and the ceiling is quoted unqualified. Measured over a real connection, the widest access log line is 1,972 bytes.

Multiply that ceiling by the request rate to size log storage. Note that the rate is not bounded by the limits above on every route: /.well-known/healthcheck and /s/* sit behind no limiter, so there the multiplier is whatever the deployment will serve.

The same ceiling covers every other line the service writes through slog that carries text an unauthenticated client supplies. The access log is not the only line a client can put its own text into, and a budget that held for one line and not the others would be worse than no stated budget at all. Every slog call an unauthenticated request can reach spends the same per-field budget through internal/logfield, and each carries strictly fewer client-supplied fields than the access log does, so none of them can be wider than it:

Log line	Level	Client-chosen value	Reachable unauthenticated
request body exceeds limit (413)	WARN	path, method	yes — MaxBodySize precedes RequireAuth
csrf: token validation failed (403)	WARN	path, method	yes — CSRF precedes RequireAuth
... rate limit exceeded (429)	WARN	path	yes, on the receiver
auth middleware: unauthenticated request	DEBUG	path, method	yes, by definition
entrypoint not found	DEBUG	entrypoint UUID	yes, on the receiver
user not found / invalid password	DEBUG	username	yes, on the login form
login failure limit exceeded (429)	WARN	path	yes, on the login form
password verification capacity exhausted	WARN	path	yes, on the login form

DEBUG being off by default is not a bound. An operator turning it on to diagnose a flood must not thereby hand the flood an unbounded write, so those lines are capped too.

The last two rows are capped defensively rather than against a demonstrated width: chi routes POST /pages/login on a static pattern, so r.URL.Path there is the 12-byte constant /pages/login and each line lands near 120 bytes. RecordLoginFailure is nonetheless an exported method taking any *http.Request, and a future caller on a route with a URL parameter would widen the line. Since no request through the mux can, both caps are pinned by tests that call those two entry points directly with the path such a caller would supply. Removing either cap fails 14 subtests.

internal/middleware/logbound_test.go and internal/handlers/logbound_test.go drive 8 KB of client-chosen text at each of these — 1 KB at invalid password, whose accounts are shared with the successful-login line, where a username past 4 KB overflows the session cookie and answers 500 before that line is written — through both handlers, and through seven fills: plain text as the baseline, and then the quotation mark, backslash, tab, newline, C0 control and astral non-printable, six characters the wider of the two handlers spends more on than the client spent sending them. Every case holds each line to the 2,560-byte ceiling. That per-line ceiling is what the figure above states, and every row establishes it.

Three of the sites go further and bound the whole flood's output — the total bytes a run of distinct invented values wrote, which is the shape an operator sizing storage cares about. They are request body exceeds limit (TestMaxBodySize_FloodOfOversizePathsDoesNotGrowTheLog), entrypoint not found and user not found (the last two through assertBoundedFlood). The other rows carry no aggregate assertion; the per-line ceiling is what they establish.

internal/logfield/logfield_test.go measures the per-rune charge against what the handlers really emit, over roughly 3,000 code points on each, so an undercharged rune fails a test rather than quietly falsifying the ceiling.

It covers GORM's statement logging as well. GORM's own default logger printed the fully interpolated SQL — parameters and all — to standard output on every statement that returned an error, including a plain record-not-found, at a level no operator setting reached. Two of this service's lookups miss by design on unauthenticated routes: the entrypoint lookup behind /webhook/{uuid} and the user lookup behind the login form, whose path segment and submitted username the client picks outright. Every gorm.Open in the service now installs the adapter in internal/gormlog instead. It writes through the same slog logger as everything else, so its lines take the level the operator set and the handler internal/logger selected, and every value it emits is spent through the same internal/logfield budget. A record-not-found is not logged as an error: it is the expected outcome on both of those paths, and each handler already records its own miss at DEBUG — bounded, per the table above — without the SQL. Slow statements are kept, at WARN, above the same 200 ms threshold GORM used and with the statement bounded, because that report is the one thing GORM's logger gave an operator that nothing else here does. The adapter orders its cases exactly as GORM's own Trace orders them, so a statement that both missed and ran slow is still reported as slow, and dropping the miss costs an operator no report IgnoreRecordNotFoundError would have kept. A GORM line spends at most two of those budgets — the statement and the driver error — against a smaller fixed portion than the access log's, and internal/gormlog/gormlog_test.go asserts each line against MaxAccessLogLineBytes directly rather than leaving it as arithmetic.

What that ceiling does not cover, stated here so the figure is not read as more than it is:

• Lines carrying an authenticated operator's own input, which are not truncated at all. webhook created logs the submitted name verbatim and target URL blocked by SSRF protection logs the target host (both internal/handlers/source_management.go), as do the target_name lines in internal/delivery/engine.go and internal/delivery/target_http.go. The only bound on any of them is the 1 MB form body cap, so a 100 KB name writes a single line of roughly 600 KB — measured. This is deliberate: every one of these requires an authenticated operator on a service with no self-registration, and truncating the operator's own configuration echoed back would cost debuggability against no adversary. It does mean the 2,560-byte figure sizes unauthenticated traffic, not the operator's own administrative requests.
• The log delivery target, which writes the whole inbound event — headers and body — to the log. This one is deliberate: capping it would defeat the target, since emitting the payload is the delivery. It costs nothing unless an authenticated operator creates a target of that type on a specific webhook, and each line it writes is bounded per event by the 1 MB receiver body cap. Adding one is a decision to spend log volume on that webhook's payloads.
• Two writers that do not go through internal/logger at all, both on standard error. fx prints the dependency graph and the lifecycle hooks through its default console logger at startup and shutdown — nothing calls fx.WithLogger, and fx.New builds that logger over os.Stderr. The Go runtime writes a panic or a fatal error itself; a panic in a background worker rather than in a request handler is the case that reaches it, since nothing recovers those. Neither carries a client-chosen value at a client-chosen length: the five panic calls in this service are invariant guards over constants and over crypto/rand.
• net/http's own faults, which are not a separate writer. internal/server/http.go builds its server with a nil ErrorLog, so net/http falls back to the log package's default logger — and internal/logger calls slog.SetDefault, which redirects that logger into whichever handler it installed. Those lines therefore arrive on standard output, shaped like every other line, at INFO. They are not truncated. A handler panic is no longer one of them: the recover middleware below answers it and writes it as the bounded record described there instead, and internal/server/recoverer_test.go requires that http: panic serving appear in neither of the process's two streams when a panic is driven through the production router. The one panic still handed back to net/http is http.ErrAbortHandler, which it special-cases and does not log at all. What is left on this path is net/http's own diagnostics, whose values are the runtime's, not a client's.

Wider than that 2,560-byte ceiling, and stated separately rather than carved out of it: the record a recovered panic produces. The recover middleware in internal/middleware answers 500 and writes one ERROR record through internal/logger carrying the panic value, the stack and the request id — the same request_id the access log line for that request carries, which is how the two are joined. It replaced chi's middleware.Recoverer, which on a current Go release crashed inside its own stack pretty-printer: the connection was dropped rather than answered, and what reached the operator described that crash rather than the fault behind it.

That record is bounded the same way, in the same encoded bytes and through the same internal/logfield budget: 512 for the panic value, because a handler is free to build one out of the request, and 8,192 for the stack, cut at its far end so that the panic site survives a cut and net/http's accept frames are what is lost. Net: at most 10,240 bytes, once per recovered panic — 9,121 by the arithmetic (523 + 8,203

• 139 + a 256-byte fixed portion), stated at 10,240 for headroom. internal/middleware/recoverer_test.go measures 8,898 bytes with the stack and the panic value both driven past their budgets, over both handlers; that figure carries no source paths and reproduces across checkouts. The real case is far below it: through the shipped middleware chain the whole record measures roughly 3,960 bytes over a roughly 3,690-byte stack, taken by internal/server/recoverer_test.go from the process's own file descriptors while driving a panic through the production router over a real server in a subprocess. That pair is not an invariant — debug.Stack() embeds absolute source paths, so it moves with where the tree is checked out, and three checkouts have reported 3,959, 3,961 and 4,026. What the tests assert is the ceiling and that the stack arrived uncut, never the number.

Every limiter here — receiver, login, and password change — identifies the client the same way, through one shared key function: the connection's own address, unless the peer is listed in TRUSTED_PROXIES, in which case the forwarded client address is used instead. That address becomes a bucket by family: IPv4 keys on the full address, IPv6 on its /64 prefix. A routed /64 is the normal residential and mobile IPv6 allocation, so keying IPv6 per address would let one subscriber rotate source addresses and mint a fresh bucket per request, evading these limits at the network layer without spoofing anything; the cost is that distinct clients inside one /64 share a bucket. IPv4-mapped addresses (::ffff:1.2.3.4) key as the IPv4 address they carry. See Trusted proxies. Deployed without that variable set, a client behind a reverse proxy shares one bucket with every other client behind the same proxy. Set TRUSTED_PROXIES to the proxy's address to get per-client limits back. What the shared bucket costs is not the same for every limiter, and the two cases pull in opposite directions:

• For the receiver limits it costs throughput, which is the safe direction to be wrong in: sharing can only make a limit bind sooner, never let a sender past it. It matters more for the aggregate limit than for the per-entrypoint one: with TRUSTED_PROXIES unset behind the reverse proxy a production deployment is required to run behind, every request keys on the proxy, so the aggregate limit becomes a service-wide ceiling of 1200 requests per minute across all senders and all entrypoints, where the per-entrypoint limit's capacity still grows with the number of entrypoints. Any deployment with more than a handful of busy entrypoints must set TRUSTED_PROXIES.
• For the login and password-change limits it costs precision, not availability. Login failures from every client land in one counter, so a stranger's wrong passwords make the operator's own wrong passwords answer 429 sooner; the operator's correct password is never affected, because it is never counted. Production deployments should still set TRUSTED_PROXIES; webhooker warns at startup whenever it is empty, in any environment.

The login endpoint

The login POST is the one endpoint with no pre-emptive limiter in front of it, and that is deliberate. A limiter that spends budget on arrival is a lockout in this deployment shape: sharing one bucket, a stranger sending five POSTs a minute — about 0.08 requests per second, from anywhere — keeps it permanently full, and the operator has no second administrative path. So the handler inverts the order:

1. Credentials are verified first, and only a failed attempt spends budget. A correct password is never rate-limited, whatever the counters hold. This is what guarantees the admin UI stays reachable.
2. Failures are counted per (client bucket, submitted username), five per minute, after which further failures from that pair are answered 429 with a Retry-After. That 429 is a label on the response, not a gate in front of the work: the credential check has already run by the time the counter is consulted, so a throttled client's guess is still evaluated. See the guessing rate below. A successful login clears the counter, so mistyping a few times and then getting it right leaves you unthrottled. Because the submitted username is attacker-controlled, at most 1024 username counters and 1024 fallback address counters are tracked; past the first cap failures fall back to the address counter, and past both they are answered as throttled without being recorded. Total tracked state is under half a megabyte and does not grow with the number of usernames an attacker invents.
3. Concurrent password verifications are capped at two, and the queue for them at 16. Verifying before counting means every login request costs an Argon2id hash, and Argon2id here is 64 MB per hash — two slots is a 128 MB ceiling on password hashing. Every endpoint that hashes a password takes a slot, including the password-change endpoint, which holds one across both the verification and the new hash. A request that waits five seconds without getting a slot is answered 503 Service Unavailable and no hash is computed for it. The wait alone does not bound memory, only how long one request holds some, so the number of waiters is capped as well. Size the queue from what a parked waiter actually retains, not from the 1 MB body cap: that caps the raw body read, while the body-cap, CSRF and form-parsing middleware all run before the guard, so a waiter holds its parsed form plus its request header block for the whole wait. Measured on the pinned Go 1.26.1 toolchain, as the heap delta with 64 waiters parked in the handler, an ordinary two-field login form retains ~0 MB, a 1 MB urlencoded body at Go's 10,000-parameter parse cap retains 2.82 MB (3.09 MB with %41 escapes), and the ~0.9 MB of headers the 1 MB header cap allows takes it to 4.18 MB — the retained parse and the headers dominate, not the raw body. So the cap is 16 waiters: 16 x 4.18 MB is about 67 MB of committed queue memory, and two slots drain a full 16-deep queue in roughly 0.6 s, far inside the five-second deadline. A request arriving past the cap is shed with 503 immediately instead of joining the queue. Peak commitment for the endpoint is therefore about 203 MB: 128 MB of Argon2id, plus the 18 requests holding a parsed form — 16 queued and the 2 being hashed — at about 75 MB. That 203 MB is live commitment, not resident size: the Go collector lets the heap reach roughly twice the live set before collecting, with transient parse garbage on top of it. The independent review of this endpoint fired 18 adversarial requests at an idle guard and measured a peak HeapAlloc of 392 MB. Provision on the order of 400 MB, not for the 203 MB itemised here and not for the hashing budget alone.

An unknown username is verified against a dummy hash rather than rejected early, so a nonexistent account costs the same time as a real one and the response cannot be used to enumerate usernames.

This raises online guessing throughput by about 300x, and that is the trade. Because the credential check always precedes the counter, what bounds online brute force is the semaphore, not the failure counter. Two slots at the cost of one Argon2id verification is on the order of 27 guesses per second, about 2.3 million per day, against 5 per minute under the pre-emptive limiter this replaced. Treat that figure as a lower bound rather than a ceiling: it was measured with Go's race detector enabled, so real hardware verifies faster and guesses faster. Choose the admin password to survive millions of online guesses per day — a long random passphrase, not a memorable one. Rate-limiting POST /pages/login at the reverse proxy, where the real client address is visible, is the way to put a cheaper bound back on top.

The residual exposure is a bounded, self-clearing loss of login availability — not merely of latency. A flood can keep both verification slots busy, and a request that neither gets a slot within five seconds nor finds room in the queue is answered 503. Above roughly 27 requests per second the operator is not served slowly, it is shed: its chance per attempt is about the ratio of service rate to flood rate, so at 400 requests per second it is roughly one attempt in fourteen. A sufficiently determined flood still denies login for as long as it runs.

What changed is the price and the aftermath. Denying login used to cost an attacker 0.08 requests per second from anywhere; it now costs 30 or more sustained, about 400 times as much. Nothing accumulates while the flood runs, nothing needs resetting when it stops, and the operator's correct password succeeds on the first attempt afterwards. Restarting the service is not a remedy: a restart clears the failure counters, which are not what is saturated, and the flood re-fills both verification slots on its first two requests. The remedies are to block the source at the reverse proxy, or to rate-limit POST /pages/login there — the one place a limit can be applied without reintroducing the lockout, because the proxy sees the real client address. Setting TRUSTED_PROXIES does not stop the saturation, but it makes the source visible in the failure logs.

Finer-grained per-webhook rate limits (configured in the web UI and enforced in the webhook handler) can layer on top of this env-level abuse limit later; they are tracked as future work.

API Endpoints

Public Endpoints

Method	Path	Description
GET	/	Root redirect, 303 (authenticated → /sources, unauthenticated → /pages/login)
GET	/.well-known/healthcheck	Health check (JSON: status, now, uptimeSeconds, uptimeHuman, version, appname, maintenanceMode)
any	/s/*	Static file serving (embedded CSS, JS). Mounted for every method, not just GET/HEAD: chi's Mount registers all methods and http.FileServer special-cases only HEAD (by omitting the body), so a POST or DELETE to an asset is answered 200 with the file. Pinned by TestStaticServesEveryMethod
POST	/webhook/{uuid}	Webhook receiver endpoint. POST only — every other method is answered 405 Method Not Allowed with Allow: POST. Rate limited (see Rate Limiting)

Authentication Endpoints

Method	Path	Description
GET	/pages/login	Login page (not rate limited)
POST	/pages/login	Login form submission. Credentials are verified before any limit is consulted, so a correct password is never throttled; 5 FAILED attempts per minute per bucket per submitted username, then 429. 503 if no verification slot frees up within 5s, or immediately if 16 requests are already queued for one (see Rate Limiting)
POST	/pages/logout	Logout (destroys session)

Authenticated Endpoints

Method	Path	Description
GET	/user/{username}	User profile page
POST	/user/{username}/password	Change the user's password (5 per minute per bucket, then 429; 503 if no verification slot frees up within 5s, or immediately if 16 requests are already queued for one)
GET	/sources	List user's webhooks
GET	/sources/new	Create webhook form
POST	/sources/new	Create webhook submission
GET	/source/{id}	Webhook detail view
GET	/source/{id}/edit	Edit webhook form
POST	/source/{id}/edit	Edit webhook submission
POST	/source/{id}/delete	Delete webhook
GET	/source/{id}/logs	Webhook event logs
POST	/source/{id}/entrypoints	Add entrypoint to webhook
POST	/source/{id}/entrypoints/{entrypointID}/delete	Delete an entrypoint
POST	/source/{id}/entrypoints/{entrypointID}/toggle	Enable or disable an entrypoint
POST	/source/{id}/targets	Add target to webhook
POST	/source/{id}/targets/{targetID}/delete	Delete a target
POST	/source/{id}/targets/{targetID}/toggle	Enable or disable a target

Infrastructure Endpoints

Method	Path	Description
GET	/metrics	Prometheus metrics, behind basic auth. The route is registered only when METRICS_USERNAME is set; otherwise it does not exist and returns 404

API (Planned)

Method	Path	Description
GET	/api/v1/webhooks	List webhooks
POST	/api/v1/webhooks	Create webhook
GET	/api/v1/webhooks/{id}	Get webhook details
PUT	/api/v1/webhooks/{id}	Update webhook
DELETE	/api/v1/webhooks/{id}	Delete webhook
GET	/api/v1/webhooks/{id}/events	List events for webhook
POST	/api/v1/events/{id}/redeliver	Redeliver an event

None of these exist yet. /api/v1 is mounted with no routes, so every path under it returns 404 today. API authentication will use API keys passed via Authorization: Bearer <key> header; no Bearer middleware is implemented either.

Package Layout

All application code lives under internal/ to prevent external imports. The entry point is cmd/webhooker/main.go.

webhooker/
├── cmd/webhooker/
│   └── main.go               # Entry point: sets globals, wires fx
├── internal/
│   ├── config/
│   │   └── config.go          # Configuration loading from environment variables
│   ├── database/
│   │   ├── base_model.go      # BaseModel with UUID primary keys
│   │   ├── database.go        # GORM connection, migrations, admin seed
│   │   ├── models.go          # AutoMigrate for config-tier models
│   │   ├── model_setting.go   # Setting entity (key-value app config)
│   │   ├── model_user.go      # User entity
│   │   ├── model_webhook.go    # Webhook entity
│   │   ├── model_entrypoint.go # Entrypoint entity
│   │   ├── model_target.go    # Target entity and TargetType enum
│   │   ├── model_event.go     # Event entity (per-webhook DB)
│   │   ├── model_delivery.go  # Delivery entity (per-webhook DB)
│   │   ├── model_delivery_result.go  # DeliveryResult entity (per-webhook DB)
│   │   ├── model_apikey.go    # APIKey entity
│   │   ├── password.go        # Argon2id hashing and verification
│   │   ├── retention.go       # Retention reaper (per-webhook event expiry)
│   │   ├── testing.go         # NewTestDatabase: wrapper for tests, no fx lifecycle
│   │   └── webhook_db_manager.go  # Per-webhook DB lifecycle manager
│   ├── globals/
│   │   └── globals.go         # Build-time variables (appname, version, arch)
│   ├── gormlog/
│   │   └── gormlog.go         # GORM's logger.Interface on top of slog, bounded
│   ├── logfield/
│   │   └── logfield.go        # Encoded-byte budget for client-supplied log values
│   ├── delivery/
│   │   ├── engine.go          # Event-driven delivery engine (channel + timer based)
│   │   ├── circuit_breaker.go # Per-target circuit breaker for http/slack targets with retries
│   │   ├── target.go          # Target interface, Task, Scheduler
│   │   ├── target_http.go     # HTTP target (retries, circuit breaker)
│   │   ├── target_slack.go    # Slack/Mattermost incoming-webhook target
│   │   ├── target_database.go # Database archive target
│   │   ├── target_database_archive.go  # Archive file lifecycle and pruning
│   │   ├── target_log.go      # Log target (stdout)
│   │   ├── target_config_view.go  # Masked target config for templates
│   │   ├── archive_sweeper.go # Periodic pruning of idle archives
│   │   ├── url_mask.go        # Strips credentials from *url.Error
│   │   └── ssrf.go            # SSRF prevention (IP validation, safe HTTP transport)
│   ├── handlers/
│   │   ├── handlers.go        # Base handler struct, JSON helpers, template rendering
│   │   ├── auth.go            # Login, logout handlers
│   │   ├── event_log_view.go  # Event log projection, byte-capped in SQL
│   │   ├── healthcheck.go     # Health check handler
│   │   ├── index.go           # Index page handler
│   │   ├── profile.go         # User profile handler
│   │   ├── source_management.go  # Webhook CRUD handlers
│   │   └── webhook.go         # Webhook receiver handler
│   ├── healthcheck/
│   │   └── healthcheck.go     # Health check service (uptime, version)
│   ├── lifecycle/
│   │   └── lifecycle.go       # Shared stop-hook waiter, bounded by the stop context
│   ├── logger/
│   │   └── logger.go          # slog setup with TTY detection
│   ├── middleware/
│   │   ├── middleware.go      # Logging, CORS, Auth, Metrics, MetricsAuth, SecurityHeaders, MaxBodySize
│   │   ├── csrf.go            # CSRF protection middleware (gorilla/csrf)
│   │   ├── ratelimit.go       # Per-IP rate limiting middleware (go-chi/httprate)
│   │   ├── loginguard.go      # Login failure counters and the Argon2id verification semaphore
│   │   └── testing.go         # NewForTest: Middleware without the fx lifecycle
│   ├── server/
│   │   ├── server.go          # Server struct, fx lifecycle, signal handling
│   │   ├── http.go            # HTTP server setup with timeouts
│   │   └── routes.go          # All route definitions
│   └── session/
│       ├── session.go         # Cookie-based session management
│       └── testing.go         # NewForTest: Session without the fx lifecycle
├── static/
│   ├── static.go              # //go:embed directive
│   ├── css/input.css          # Tailwind input, source for tailwind.css (make css)
│   ├── css/tailwind.css       # Generated stylesheet the pages load
│   ├── css/style.css          # Older hand-written stylesheet, no longer loaded
│   ├── js/app.js              # Progressive-enhancement copy-to-clipboard
│   ├── js/alpine.min.js       # Alpine.js, fetched by script/fetch-assets, not committed
│   └── vendor.sha256          # Pinned hashes the fetched assets are verified against
├── templates/                 # Go HTML templates (base, login, sources, etc.)
├── script/                    # Scripts to Rule Them All entrypoints
├── Dockerfile                 # Three stages: lint, test+build, Alpine runtime
├── Dockerfile.lint            # Lint-only image built by script/lint
├── Makefile                   # 10 of 16 targets shim script/; 6 are inline
├── go.mod / go.sum
└── .golangci.yml              # Linter configuration

Dependency Injection

Components are wired via Uber fx in this order:

1. globals.New — Build-time variables (appname, version, arch)
2. logger.New — Structured logging (slog with TTY detection)
3. config.New — Configuration loading (environment variables)
4. database.New — Main SQLite connection, config migrations, admin user seed
5. database.NewWebhookDBManager — Per-webhook event database lifecycle manager
6. database.NewRetentionReaper — Per-webhook event retention sweep
7. healthcheck.New — Health check service
8. session.New — Cookie-based session manager (key from database)
9. handlers.New — HTTP handlers
10. middleware.New — HTTP middleware
11. delivery.New — Event-driven delivery engine
12. delivery.NewArchiveSweeper — Periodic pruning of idle archives
13. delivery.Engine → delivery.Notifier — interface bridge
14. delivery.Engine → delivery.WebhookEvictor — interface bridge so deleting a webhook releases its archive writer
15. server.New — HTTP server and router

The server starts via fx.Invoke(func(*server.Server, *delivery.Engine, *database.RetentionReaper, *delivery.ArchiveSweeper) {}), which triggers the fx lifecycle hooks in dependency order. The delivery.Notifier interface allows the webhook handler to send self-contained delivery.Task slices to the engine without a direct package dependency. Each task carries all target config and event data inline (for bodies under 16 KiB, delivery.MaxInlineBodySize), so the engine can deliver without reading from any database — it only writes to record results.

Middleware Stack

Applied to all routes in this order:

1. RequestID — Generate unique request IDs (chi built-in)
2. SecurityHeaders — Production security headers on every response (HSTS, X-Content-Type-Options, X-Frame-Options, CSP, Referrer-Policy, Permissions-Policy)
3. Logging — Structured request logging (method, URL, status, latency, remote IP, user agent, request ID)
4. Metrics — Prometheus HTTP metrics (if METRICS_USERNAME is set)
5. CORS — Cross-origin resource sharing headers
6. Timeout — 60-second request timeout
7. Recoverer — Panic recovery: one ERROR record through internal/logger and a 500
8. Sentry — Error reporting to Sentry (if SENTRY_DSN is set; configured with Repanic: true so panics still reach Recoverer)

Recoverer sits seventh rather than first, and both neighbours are the reason. It runs inside everything that observes the response, so the 500 it writes for a panicking handler is the status the access log records and the metrics count; registered first, as chi's own middleware.Recoverer was, the same request was logged as a 200 that the client never received. It runs outside the Sentry handler, so Repanic: true has something to re-raise into: an operator with SENTRY_DSN set keeps the report, and one without it now gets the local record instead of nothing. What that placement gives up is recovery of a panic in the six entries above it, none of which does more than set a header or start a timer.

Additionally, form endpoints (/pages, /user/*, /sources, /source/*) apply a MaxBodySize middleware that limits POST/PUT/PATCH request bodies to 1 MB. It is registered ahead of the CSRF middleware in every one of those route groups, because gorilla/csrf parses the form; if the cap were installed after it, form parsing would run under net/http's 10 MB default and the 1 MB limit would never apply. A request that declares a Content-Length over the limit is answered with 413 Request Entity Too Large without its body being read and without reaching CSRF, the route group's remaining middleware, or the handler. It is not rejected before any other middleware, though: the global entries listed above all run first, so such a request is still logged and given the security headers — and counted in the metrics, on a deployment where METRICS_USERNAME is set and the Metrics middleware is therefore registered at all. The rejection itself is logged at WARN with the method, path and declared length. A chunked request, or one that lies about its length, is hard-capped by http.MaxBytesReader and fails downstream at form-parse time.

Those same four route groups then apply CSRF and NoCache (Cache-Control: no-store, Pragma: no-cache), and every group except /pages applies RequireAuth. The rate limiters are per-route rather than global: PasswordChangeRateLimit on /user/{username}/password and ReceiverRateLimit on /webhook/{uuid}. There is deliberately none on /pages/login — that endpoint counts failures inside the handler, after the credential check, see The login endpoint.

Authentication

• Web UI: Cookie-based sessions using gorilla/sessions with encrypted cookies. Sessions are configured with HttpOnly, SameSite Lax, and Secure (in production). Absolute session lifetime is 7 days, with a sliding idle timeout on top of it (see Sessions).
• API (planned): API key authentication via Authorization: Bearer header. API keys are stored per-user with usage tracking (last_used_at).
• Metrics: Basic authentication protecting the /metrics endpoint.

Security

• Passwords hashed with Argon2id (64 MB memory cost)
• Session cookies are HttpOnly, SameSite Lax, Secure (prod only)
• Session regeneration on login to prevent session fixation attacks
• Session key is a 32-byte value auto-generated on first startup and stored in the database
• Production security headers on all responses: HSTS, X-Content-Type-Options (nosniff), X-Frame-Options (DENY), Content-Security-Policy, Referrer-Policy, and Permissions-Policy
• Request body size limits (1 MB) on all form POST endpoints, enforced by middleware that runs before CSRF parses the form
• CSRF protection via gorilla/csrf on all state-changing forms (cookie-based double-submit tokens with HMAC authentication). Applied to /pages, /sources, /source, and /user routes. Excluded from /webhook (inbound webhook POSTs) and /api (stateless API). The middleware auto-detects TLS status per-request (via r.TLS and X-Forwarded-Proto) to set appropriate cookie security flags and Origin/Referer validation mode
• SSRF prevention for HTTP delivery targets: private/reserved IP ranges (RFC 1918, loopback, link-local, cloud metadata) are blocked both at target creation time (URL validation) and at delivery time (custom HTTP transport with SSRF-safe dialer that validates resolved IPs before connecting, preventing DNS rebinding attacks)
• Login limiting is inverted, deliberately. The login POST has no pre-emptive rate limiter in front of it. Credentials are verified first and only a failed attempt spends budget, so a correct password is never throttled and no flood of wrong ones can deny the operator the only administrative path. Failures are counted per (bucket, submitted username), five per minute, after which further failures are answered 429 with a Retry-After. What bounds brute force is not that counter but the cap of two concurrent Argon2id verifications: a throttled client's guess is still evaluated, so roughly 27 guesses a second get through and the admin password has to carry that load (see The login endpoint). GET requests to the login page are not limited
• Password-change rate limiting via go-chi/httprate: sliding-window rate limiter, 5 POST attempts per minute per bucket. It runs behind session auth, so only a client already holding a valid session reaches it, and an operator throttled out of changing a password can still log in. The bucket is per client IP only when TRUSTED_PROXIES names the reverse proxy; unset, every client shares one bucket, which costs precision rather than availability (see Rate Limiting). webhooker warns at startup whenever TRUSTED_PROXIES is empty
• Prometheus metrics behind basic auth
• Static assets embedded in binary (no filesystem access needed at runtime)
• Container runs as non-root user (UID 1000)
• GORM soft deletes on every entity that carries BaseModel, which is all of them but Setting (data preserved for audit)

Shutdown

On SIGINT or SIGTERM, fx runs the registered stop hooks in reverse dependency order under a 5 second budget (fx.StopTimeout in cmd/webhooker/main.go). That budget covers the whole sequence, not each hook. The order, read off the fx stop-hook log:

1. ArchiveSweeper
2. RetentionReaper
3. server — the HTTP drain, bounded separately by server.ShutdownTimeout (3 seconds), then a Sentry flush if SENTRY_DSN is set
4. delivery.Engine
5. healthcheck
6. WebhookDBManager
7. the database close

The two components that can realistically hold the budget run first: a retention sweep or an archive prune caught mid-tick each waits on its WaitGroup bounded by the stop context, so a wedge there consumes the 5 seconds before the HTTP server hook is ever entered. The hooks after the server are microsecond-scale in normal operation.

The HTTP drain budget is deliberately shorter than the sequence budget. Were the two equal, a drain that used its whole budget would exhaust the sequence budget at the instant it finished, and every later hook — the delivery engine, the healthcheck, the webhook DB manager and the database close — would be skipped in exactly the case where the drain mattered. 3 seconds leaves 2 seconds (server.TailHookReserve) for the tail, which is far more than the microseconds it needs.

That reserve belongs to the tail hooks, not to the server hook, and the Sentry flush is what could take it: it runs after the drain inside the same hook, and sentry.Flush takes a bare duration and honours no context, so an unreachable Sentry endpoint would add its own timeout on top of a full-length drain and consume the whole sequence budget by itself. It is therefore clamped to whatever is left on the stop context minus the reserve, and skipped when that leaves too little to be worth attempting — so a full-length drain means Sentry events are dropped rather than the database close being skipped.

This does not make the database close unconditional: a wedged ArchiveSweeper or RetentionReaper still runs first and can consume the whole budget on its own.

The value is chosen to sit inside the container stop grace period. Docker's default docker stop grace is 10 seconds and the Dockerfile sets no STOPSIGNAL or grace override, so the process must be gone before that. fx's own default is 15 seconds, which is past the grace: the container would be SIGKILLed (exit 137) before the bound could fire, and nothing that depends on it — including the shutdown timed out, goroutines still running error log that tells an operator a component is wedged — would ever be reached.

Two operational consequences follow from bounding the sequence:

• A wedged component aborts the rest of the shutdown. fx checks the stop context before each remaining hook and returns outright once it has expired, skipping the hooks it has not reached. If the first-stopped component consumes the whole budget, the later hooks never run — the database close among them. SQLite is crash-safe, so this is not corruption, but it is not a clean close either.
• Lowering the grace below 5 seconds reintroduces the silent truncation. docker stop --time, Compose's stop_grace_period, or Kubernetes' terminationGracePeriodSeconds set under 5 seconds put SIGKILL back in front of the bound, and the process dies with no shutdown diagnostics at all. Keep the deployment's grace above the stop timeout.

Linting

golangci-lint never runs on the host. script/lint builds Dockerfile.lint, which copies the repo into the digest-pinned golangci-lint image and lints as a build step, so a successful build is a clean lint. A host binary would share one cache and one lock with every other checkout on the machine, which has produced both invented findings attributed to other worktrees and unearned passes.

Three properties are load-bearing:

• script/lint passes --no-cache-filter=lint. Without it an unchanged tree replays the lint layer from cache and the build exits 0 in under a second having linted nothing. The deps stage stays cacheable, so module downloads are not repeated. Invalidation is scoped to the one stage; never prune the shared build cache.
• script/lint does not trust that flag. Docker silently ignores --no-cache-filter for a stage name that does not match, so a stage rename or a one-character typo would restore the cached false green with no warning and a fast exit 0. The script therefore tees the build output and treats a run as a pass only if golangci-lint's own summary line (N issues. / N issues:) appears in it: no summary, no lint, whatever the exit code says.
• Both lint steps use RUN --network=none. golangci-lint config verify is documented as fetching its JSON schema over HTTPS, which would be an unpinned remote dependency; the pinned image resolves the schema without network access, and --network=none enforces that instead of trusting it. Verify is worth keeping because golangci-lint run silently ignores config keys it does not recognize, so a typo would disable a setting with no warning.

Docker

The Dockerfile uses a three-stage build. Each stage is pinned by digest, and the two check stages are separate images so the linter's version is fixed independently of the compiler's:

1. Lint stage (golangci/golangci-lint:v2.12.2, Debian-based) — installs make, downloads dependencies, copies the source, and runs make fmt-check, then golangci-lint config verify and golangci-lint run, both with --network=none.
2. Builder stage (golang:1.26.1-bookworm) — depends on the lint stage passing (it copies a file from it), runs script/fetch-assets to download and verify the third-party browser assets, then runs make test and make build, and finally rebuilds the binary with CGO_ENABLED=1 and static linking so it runs on musl.
3. Runtime stage (alpine:3.21) — copies the static binary, creates the /var/lib/webhooker directory for all SQLite databases, runs as the non-root webhooker user (UID 1000), exposes port 8080, and includes a health check against /.well-known/healthcheck.

The lint stage invokes golangci-lint directly rather than make lint: it is already the pinned linter image, and make lint builds Dockerfile.lint, which would need a docker daemon inside this build.

Both check stages use Debian rather than Alpine because gorm.io/driver/sqlite pulls in mattn/go-sqlite3, which needs CGO and does not compile against musl. Only the final binary is statically linked, which is what lets it run on the Alpine runtime image.

script/cibuild — docker build . — is the CI gate: the checks run inside the image, so a build that succeeds is a repo that is formatted, linted, tested and compiled. script/lint also uses Docker (Dockerfile.lint, see Linting above), so make lint and make check run the same pinned linter version the gate does; only script/test and script/fmt-check run on the host.

CI gate honesty

A layer cache lets docker build . exit 0 in seconds with the lint and test stages replayed rather than executed, which would make a green check meaningless. The check workflow therefore writes .ci-fingerprint into the build context before building. Its value is the hash of the last commit that touched the build context, so:

• Any commit that changes code (including a squash merge whose tree matches an already-built branch) gets a new fingerprint, invalidates the COPY . . layer of both check stages, and really runs make fmt-check, golangci-lint, make test, and make build. A run that reports success ran them.
• A docs-only commit leaves the fingerprint unchanged — .dockerignore excludes *.md, LICENSE and .editorconfig from the context anyway — so the image replays from cache and costs seconds.

The module download layer sits above COPY . . and stays cached either way.

A separate workflow step, run before the fingerprint is written, covers a second way the gate lied: Gitea cancels an in-flight run when a newer commit lands on the same branch and records that cancellation as a failure status, so a commit nothing ever tested reads as a test result. Cancellation is unconditional server-side for push events, so the superseding run calls script/ci-mark-superseded, which rewrites that exact status to failure / Superseded by a newer commit; never tested.

The state stays failure on purpose: Gitea's combined status folds skipped into success, so marking a never-tested commit skipped made the status API report green for it, indistinguishable from a commit that passed. Reading a commit's status on this repo therefore goes:

• success / Successful in ... — the checks ran and passed.
• failure / Failing after ... — the checks ran and failed.
• failure / Superseded by a newer commit; never tested — the run was cancelled, by a newer push or by hand, and nothing was verified about this commit. Test the commit itself before concluding anything about it.

Genuine failures and successes are never touched, and no status is left pending, which would block the commit indefinitely. The step derives its context string from the workflow name, the job id and the event. That is deliberately not byte-identical to Gitea's own rule, which uses the job's display name: where the runner exports the id, so giving the job a name: — or renaming the workflow — makes the derived context stop matching. The step fails loudly when no status on the commit carries that context, so no rename can silently disable the rewrite.

TODO

See TODO.md.

License

MIT

Author

@sneak