Documentation
¶
Overview ¶
Package agent is the client side of the agent-hub contract. The same code runs inside the hub (the built-in agent, over the in-process transport) and, later, on remote servers.
Index ¶
Constants ¶
This section is empty.
Variables ¶
var ErrNotEnrolled = errors.New("the agent is not enrolled")
ErrNotEnrolled is returned by LoadCredentials when the agent has no credentials yet.
Functions ¶
Types ¶
type Collector ¶
type Collector interface {
Collect(ctx context.Context) (*agentv1.MetricsReport, error)
}
Collector gathers the metrics sent to the hub.
type Credentials ¶
type Credentials struct {
AgentID string
Key *ecdsa.PrivateKey
CertDER []byte
CADER []byte
}
Credentials identify an enrolled agent to the hub and tell it which hub to trust.
func Enroll ¶
func Enroll(ctx context.Context, hubAddr, fingerprint, token, version, dir string) (*Credentials, error)
Enroll registers the agent with the hub at hubAddr ("host:port") using a token from the dashboard, and saves the credentials in dir. fingerprint is the fingerprint of the hub's CA, which the agent checks before it sends the token: without it, anyone who can intercept the connection could take it.
func LoadCredentials ¶
func LoadCredentials(dir string) (*Credentials, error)
LoadCredentials reads the credentials saved in dir by Enroll.
func (*Credentials) Dial ¶
func (c *Credentials) Dial(hubAddr string) (*grpc.ClientConn, error)
Dial opens the connection to the hub at hubAddr with the credentials. The connection is made lazily and re-made by gRPC when it drops.
type LinuxCollector ¶
type LinuxCollector struct {
// contains filtered or unexported fields
}
LinuxCollector reads CPU, memory, disk, network and uptime metrics from /proc and statfs. CPU usage is computed between two consecutive calls, so the first report covers the time since the collector was created.
func NewLinuxCollector ¶
func NewLinuxCollector() *LinuxCollector
NewLinuxCollector returns a collector that reads /proc.
func (*LinuxCollector) Collect ¶
func (c *LinuxCollector) Collect(context.Context) (*agentv1.MetricsReport, error)
Collect gathers a report. A source that cannot be read leaves its part of the report empty; it fails only when neither the CPU nor the memory can be read (for example on a system without /proc).
type Options ¶
type Options struct {
// Version is the agent release version sent in Hello.
Version string
// ID is the agent id. Leave it empty to let the hub assign one; the
// assigned id is reused when the agent reconnects.
ID string
// Collector gathers metrics. Defaults to the Linux collector (CPU, memory,
// disks, network, uptime), or to one that reports only the time and the
// uptime where /proc is not available.
Collector Collector
// Log defaults to slog.Default().
Log *slog.Logger
// UpdateDir is the agent's data directory when this agent may update itself:
// the downloaded release and the request for the installer go there. Empty
// (the built-in agent, or an agent without the installer unit) refuses the
// ApplyUpdate command.
UpdateDir string
// Verifier checks the signature of an update before the installer is asked;
// the installer checks it again as root.
Verifier update.Verifier
// MinBackoff and MaxBackoff bound the wait between reconnection attempts
// (defaults 1s and 30s).
MinBackoff, MaxBackoff time.Duration
}
Options configure Run.
type RejectedError ¶
type RejectedError struct {
Reason agentv1.Reject_Reason
Message string
}
RejectedError is returned by Run when the hub refuses the agent for good: it is revoked or unknown. Retrying would not help, so Run does not. A refusal because of versions is different: updating the hub or the agent fixes it, so Run keeps trying (see versionRejection).
func (*RejectedError) Error ¶
func (e *RejectedError) Error() string
Source Files