Files
package-updater/internal/agent/conn/conn.go
shcizo 0e9e34db06 fix: address final review findings for master/agent phase 1 (#6)
- inventory: resolve the image reference via ContainerInspect when a moved
  tag makes ContainerList report the image ID
- store: allow reusing a revoked agent's name (partial unique index) and
  mark the revoked agent's compose/container services removed
- web: publish a services event on revoke; no-store on the one-time token page
- webauth: reject empty OIDC state; accept the CSRF form token from the body only
- agent conn: bound the WebSocket handshake to 30s
- hub: log agent_replaced when a connection replaces an older one
2026-10-12 15:16:29 +02:00

180 lines
5.0 KiB
Go

// Package conn keeps an agent connected to the master: it dials over pinned
// TLS, announces itself, sends its inventory, and streams changes until the
// connection drops, then reconnects with backoff.
package conn
import (
"context"
"fmt"
"log/slog"
"net/http"
"time"
"github.com/coder/websocket"
"github.com/coder/websocket/wsjson"
"github.com/shcizo/package-updater/internal/agent/inventory"
"github.com/shcizo/package-updater/internal/pin"
"github.com/shcizo/package-updater/internal/protocol"
)
// readLimit caps a single incoming message. The library default (32 KiB)
// is too small for the master's future job messages; matches the master.
const readLimit = 1 << 20
// dialTimeout bounds the TLS + WebSocket handshake so a master that accepts
// TCP but never answers does not hang the agent forever.
const dialTimeout = 30 * time.Second
// Source supplies what the agent reports. *inventory.Collector satisfies it.
type Source interface {
Hello(ctx context.Context) (protocol.Hello, error)
Snapshot(ctx context.Context) ([]protocol.Service, error)
}
// Config holds connection settings. Token is sent as a bearer header and must
// never be logged.
type Config struct {
URL string // wss://host:port/agent
Token string // AGENT_TOKEN
Fingerprint string // MASTER_KEY
Version string
MinBackoff time.Duration // default 1s
MaxBackoff time.Duration // default 30s
}
// Agent is the agent's side of the master connection.
type Agent struct {
cfg Config
src Source
log *slog.Logger
changes chan struct{}
}
// New returns an Agent. Call Run to connect.
func New(cfg Config, src Source, log *slog.Logger) *Agent {
if cfg.MinBackoff == 0 {
cfg.MinBackoff = time.Second
}
if cfg.MaxBackoff == 0 {
cfg.MaxBackoff = 30 * time.Second
}
return &Agent{cfg: cfg, src: src, log: log, changes: make(chan struct{}, 1)}
}
// Notify tells the agent its inventory may have changed. It never blocks;
// several calls before the agent gets to them collapse into one refresh.
func (a *Agent) Notify() {
select {
case a.changes <- struct{}{}:
default:
}
}
// Run keeps the agent connected until ctx ends, then returns nil.
func (a *Agent) Run(ctx context.Context) error {
backoff := a.cfg.MinBackoff
for {
connected, err := a.session(ctx)
if ctx.Err() != nil {
return nil
}
if connected {
backoff = a.cfg.MinBackoff
}
a.log.Warn("master_connection_lost", "error", err, "retry_in", backoff.String())
select {
case <-ctx.Done():
return nil
case <-time.After(backoff):
}
backoff = min(backoff*2, a.cfg.MaxBackoff)
}
}
// session runs one connection until it fails. connected reports whether the
// hello was delivered, so Run resets its backoff only after real progress.
func (a *Agent) session(parent context.Context) (connected bool, err error) {
httpClient := &http.Client{Transport: &http.Transport{TLSClientConfig: pin.ClientConfig(a.cfg.Fingerprint)}}
// The dial context only bounds the handshake; coder/websocket does not
// tie the established connection to it, so cancelling afterwards is safe.
dialCtx, cancelDial := context.WithTimeout(parent, dialTimeout)
c, _, err := websocket.Dial(dialCtx, a.cfg.URL, &websocket.DialOptions{
HTTPClient: httpClient,
HTTPHeader: http.Header{"Authorization": []string{"Bearer " + a.cfg.Token}},
})
cancelDial()
if err != nil {
return false, fmt.Errorf("dial master: %w", err)
}
defer c.CloseNow()
c.SetReadLimit(readLimit)
ctx, cancel := context.WithCancel(parent)
defer cancel()
// The master sends nothing yet in this phase, but reading is how a
// dropped connection is noticed, so a reader runs for the whole session.
go func() {
defer cancel()
for {
if _, _, err := c.Read(ctx); err != nil {
return
}
}
}()
hello, err := a.src.Hello(ctx)
if err != nil {
return false, err
}
hello.AgentVersion = a.cfg.Version
if err := send(ctx, c, protocol.TypeHello, hello); err != nil {
return false, err
}
last, err := a.src.Snapshot(ctx)
if err != nil {
return true, err
}
if err := send(ctx, c, protocol.TypeSnapshot, protocol.Snapshot{Services: last}); err != nil {
return true, err
}
a.log.Info("master_connected", "services", len(last))
for {
select {
case <-ctx.Done():
_ = c.Close(websocket.StatusNormalClosure, "")
return true, ctx.Err()
case <-a.changes:
next, err := a.src.Snapshot(ctx)
if err != nil {
return true, err
}
upserts, removes := inventory.Diff(last, next)
for _, s := range upserts {
if err := send(ctx, c, protocol.TypeUpsert, protocol.Upsert{Service: s}); err != nil {
return true, err
}
}
for _, k := range removes {
if err := send(ctx, c, protocol.TypeRemove, protocol.Remove{Key: k}); err != nil {
return true, err
}
}
last = next
}
}
}
func send(ctx context.Context, c *websocket.Conn, typ string, payload any) error {
env, err := protocol.Encode(typ, payload)
if err != nil {
return err
}
if err := wsjson.Write(ctx, c, env); err != nil {
return fmt.Errorf("send %s: %w", typ, err)
}
return nil
}