Skip to content
edgesync Artifact [462403e59c]
Login

Artifact [462403e59c]

Check-ins Using Download Text

Artifact 462403e59c490aff482f276618942fe03d439f87b4833bae53b00ca9267b39b8:

Hub Package

Purpose

In-process host for an EdgeSync hub: a fossil hub repo, an embedded NATS server with JetStream, auto-port allocation, and an HTTP server exposing the fossil timeline UI / xfer endpoint. Lets consumers (notably the bones project) host a hub without rolling their own libfossil + nats-server bootstrap.

Before this package, every adopter that needed a hub fossil — a central shared repo that leaves push to and that serves the timeline — had to write the same ~900-line wrapper around libfossil.Create/Open, natsserver.NewServer, JetStream config, port allocation, PID files, and shutdown supervision.

Public API

All types are libfossil-free and nats-server-free. Consumers import only github.com/danmestas/EdgeSync/hub.

type Config struct {
    RepoPath           string        // hub.fossil path; created if absent
    BootstrapUser      string        // default "hub"
    NATSStoreDir       string        // default <repo dir>/nats-store
    FossilHTTPPort     int           // 0 = auto-pick
    NATSClientPort     int           // 0 = auto-pick
    NATSLeafPort       int           // 0 = auto-pick
    CheckpointInterval time.Duration // 0 = 10s (PASSIVE WAL checkpoint cadence)
}

type Hub struct{ /* opaque */ }

func NewHub(ctx context.Context, cfg Config) (*Hub, error)
func (h *Hub) ServeHTTP(ctx context.Context) error
func (h *Hub) Stop() error

func (h *Hub) NATSURL() string       // "nats://127.0.0.1:NNNN"
func (h *Hub) LeafUpstream() string  // "nats-leaf://127.0.0.1:LLLL"
func (h *Hub) HTTPAddr() string      // "127.0.0.1:HHHH"

type User struct{ Login, Caps string }

func (h *Hub) AddUser(User) error
func (h *Hub) GetUser(login string) (User, error)
func (h *Hub) HasUser(login string) bool
func (h *Hub) ListUsers() ([]User, error)
func (h *Hub) RemoveUser(login string) error

type RevID string
type FileToCommit struct{ Name string; Content []byte }
type CommitOpts struct{ Files []FileToCommit; Message, Author string }

func (h *Hub) Commit(ctx context.Context, opts CommitOpts) (RevID, error)
func (h *Hub) Read(ctx context.Context, path string) ([]byte, error)
func (h *Hub) ReadAt(ctx context.Context, rev RevID, path string) ([]byte, error)

Lifecycle

NewHub does the bootstrap-or-open: opens an existing RepoPath if present, calls libfossil.Create if absent. It then binds the HTTP listener (so HTTPAddr() is immediately usable for downstream consumers wiring leaves to the hub) and starts the embedded NATS server, waiting for ReadyForConnections. After NewHub returns successfully, all three address accessors return live, reachable URLs.

ServeHTTP is a blocking goroutine — typical pattern is go h.ServeHTTP(ctx). It runs the fossil HTTP server (h.repo.XferHandler()) against the listener bound by NewHub until ctx is cancelled or Stop is called. Stop is idempotent.

ctx, cancel := context.WithCancel(context.Background())
defer cancel()

h, err := hub.NewHub(ctx, hub.Config{RepoPath: "/data/hub.fossil"})
if err != nil { return err }
defer h.Stop()

go h.ServeHTTP(ctx)
// Hub now serving at h.HTTPAddr(), NATS at h.NATSURL(), leaf upstream h.LeafUpstream()

SQLite Tuning

busy_timeout = 30000 is applied at hub bootstrap so concurrent operations retry on SQLITE_BUSY rather than failing fast.

The hub deliberately does not cap MaxOpenConns. An earlier iteration set SetMaxOpenConns(1) (matching what bones did before this package existed), but libfossil's clone path has internal goroutines that all need a DB connection — capping the pool deadlocked every concurrent clone (issue #120). busy_timeout alone is the right primitive for concurrent-write safety.

Vanilla Fossil Interop

hub.fossil is opened in WAL journal mode (libfossil's default). Without periodic checkpoints, the on-disk file would stay as a ~4 KiB stub plus a multi-hundred-KiB *-wal sidecar for the lifetime of the hub — and vanilla fossil ui / fossil info / third-party SQLite tooling would refuse to open it because their validity probe stats the main file before consulting the WAL.

The hub guards both ends of that contract:

There is no "disable" knob: vanilla-fossil readability is the contract this package exists to uphold. Set a long CheckpointInterval if you really want to dilute the cadence.

Layout: in-root vs. separate module

The hub/ directory lives inside the EdgeSync root module rather than as a separate go.mod sibling of leaf//bridge/. This was a deliberate choice to ship the package without a chicken-and-egg release: the package can evolve in lockstep with the leaf agent, and root tags pick it up.

If a downstream consumer needs to depend on hub independently of EdgeSync's full transitive deps, the package can be promoted to its own module without changing its public API. That migration is tracked in issue #118; triggers worth waiting for include consumers wanting to embed hub but skip the rest of EdgeSync's deps, or hub picking up dependencies the rest of EdgeSync doesn't want.

Out of Scope

Integration with the Leaf Agent

A leaf agent boots with NATSUpstream: hub.NATSURL() (or pulls the leaf-node URL via hub.LeafUpstream() for the leaf-node path). Sync goes over the HTTP-xfer transport — leaves construct an HTTP transport against hub.HTTPAddr() and call Agent.SyncTo(...). End-to-end coverage of this path is locked by sim/hub_leaf_test.go::TestHubLeafE2E (see testing-strategy.md).

Cross-references