Make iPXE script kernel, initrd retries configurable:

This allows for transient network issues to not cause the iPXE script to fail downloading kernel and initramfs. Signed-off-by: Jacob Weinstock <[email protected]>
tinkerbell · May 13, 2024 · cf4aad5 · cf4aad5
1 parent 20323fa
commit cf4aad5
Show file tree

Hide file tree

Showing 5 changed files with 7 additions and 0 deletions.
diff --git a/cmd/smee/flag.go b/cmd/smee/flag.go
@@ -106,6 +106,7 @@ func ipxeHTTPScriptFlags(c *config, fs *flag.FlagSet) {
 	fs.StringVar(&c.ipxeHTTPScript.hookURL, "osie-url", "", "[http] URL where OSIE (HookOS) images are located")
 	fs.StringVar(&c.ipxeHTTPScript.tinkServer, "tink-server", "", "[http] IP:Port for the Tink server")
 	fs.BoolVar(&c.ipxeHTTPScript.tinkServerUseTLS, "tink-server-tls", false, "[http] use TLS for Tink server")
+	fs.IntVar(&c.ipxeHTTPScript.retries, "ipxe-script-retries", 0, "[http] number of retries to attempt when fetching kernel and initrd files in the iPXE script")
 }
 
 func dhcpFlags(c *config, fs *flag.FlagSet) {

diff --git a/cmd/smee/flag_test.go b/cmd/smee/flag_test.go
@@ -111,6 +111,7 @@ FLAGS
   -http-addr                          [http] local IP:Port to listen on for iPXE HTTP script requests (default "%[1]v:80")
   -http-ipxe-binary-enabled           [http] enable iPXE HTTP binary server (default "true")
   -http-ipxe-script-enabled           [http] enable iPXE HTTP script server (default "true")
+  -ipxe-script-retries                [http] number of retries to attempt when fetching kernel and initrd files in the iPXE script (default "0")
   -osie-url                           [http] URL where OSIE (HookOS) images are located
   -tink-server                        [http] IP:Port for the Tink server
   -tink-server-tls                    [http] use TLS for Tink server (default "false")

diff --git a/cmd/smee/main.go b/cmd/smee/main.go
@@ -83,6 +83,7 @@ type ipxeHTTPScript struct {
 	tinkServerUseTLS              bool
 	trustedProxies                string
 	disableDiscoverTrustedProxies bool
+	retries                       int
 }
 
 type dhcpConfig struct {
@@ -219,6 +220,7 @@ func main() {
 			PublicSyslogFQDN:   cfg.dhcp.syslogIP,
 			TinkServerTLS:      cfg.ipxeHTTPScript.tinkServerUseTLS,
 			TinkServerGRPCAddr: cfg.ipxeHTTPScript.tinkServer,
+			IPXEScriptRetries:  cfg.ipxeHTTPScript.retries,
 		}
 		// serve ipxe script from the "/" URI.
 		handlers["/"] = jh.HandlerFunc()

diff --git a/internal/ipxe/script/hook.go b/internal/ipxe/script/hook.go
@@ -52,4 +52,5 @@ type Hook struct {
 	TraceID           string
 	VLANID            string // string number between 1-4095
 	WorkerID          string // example 3c:ec:ef:4c:4f:54 or worker1
+	Retries           int // number of retries to attempt when fetching kernel and initrd files
 }
diff --git a/internal/ipxe/script/ipxe.go b/internal/ipxe/script/ipxe.go
@@ -26,6 +26,7 @@ type Handler struct {
 	PublicSyslogFQDN   string
 	TinkServerTLS      bool
 	TinkServerGRPCAddr string
+	IPXEScriptRetries  int
 }
 
 type data struct {
@@ -227,6 +228,7 @@ func (h *Handler) defaultScript(span trace.Span, hw data) (string, error) {
 		TinkGRPCAuthority: h.TinkServerGRPCAddr,
 		VLANID:            hw.VLANID,
 		WorkerID:          wID,
+		Retries:           h.IPXEScriptRetries,
 	}
 	if sc := span.SpanContext(); sc.IsSampled() {
 		auto.TraceID = sc.TraceID().String()