Update retry and reboot logic in epoxy template and 1to2 scripts (#154)

* Reboot after max retry
* Reboot after max delay is reached
* Reduce max retry

actions/stage2/stage1to2.ipxe

@@ -11,20 +11,17 @@ echo -- Downloading stage2 image from ${vmlinuz_url}
 
 # Initialize retry counters.
 set retry_count:int32 1
-set max_retry_count 20
+set max_retry_count 15
 
 goto firstfetch
 
 :loop
-  inc retry_count 1
-  echo -- Retries ${retry_count} ${max_retry_count}
-  echo Failed ${retry_count} times... Retrying after ${retry_count} seconds
-  # This doesn't work?
-  # iseq ${retry_count} ${max_retry_count} && goto fetch_timeout
+  echo Failed ${retry_count} of ${max_retry_count} times... Retrying after ${retry_count} seconds
   sleep ${retry_count}
+  inc retry_count 1
+  iseq ${retry_count} ${max_retry_count} && goto fetch_timeout ||
 
 :firstfetch
-  echo -- Retries ${retry_count} ${max_retry_count}
   kernel --name vmlinuz ${vmlinuz_url} || goto loop
 
 # Initialize retry counters.
@@ -34,13 +31,12 @@ set max_retry_count 20
 goto firstfetch_initrd
 
 :loop_initrd
-  inc retry_count 1
-  echo -- Retries ${retry_count} ${max_retry_count}
-  echo Failed ${retry_count} times... Retrying after ${retry_count} seconds
+  echo Failed ${retry_count} of ${max_retry_count} times... Retrying after ${retry_count} seconds
   sleep ${retry_count}
+  inc retry_count 1
+  iseq ${retry_count} ${max_retry_count} && goto fetch_timeout ||
 
 :firstfetch_initrd
-  echo -- Retries ${retry_count} ${max_retry_count}
   initrd --name initrd ${initram_url} || goto loop_initrd
 
 imgstat
@@ -81,7 +77,6 @@ set kargs ${kargs} epoxy.project=${project}
 boot vmlinuz ${kargs} || shell
 
 :fetch_timeout
-  echo -- Retries ${retry_count} ${max_retry_count}
-  echo Failed too many times..
-  # Reboot?
-  shell
+  echo Rebooting after ${retry_count} of ${max_retry_count} attempts.
+  sleep 10
+  reboot || shell

configs/stage1_mlxrom/stage1-template.ipxe

@@ -6,8 +6,8 @@ set epoxyaddress epoxy-boot-api.{{project}}.measurementlab.net:4430
 set menu_timeout_ms:int32 5000
 set fetch_timeout_ms 10000
 
-set retry_delay_s:int32 30
-set max_retry_delay_s 480
+set retry_delay_s:int32 15
+set max_retry_delay_s 240
 
 # IPv6 network configuration.
 # ipv6_enabled will always be defined. When ipv6_enabled is "false", then other
@@ -50,7 +50,6 @@ set stage1_url https://${epoxyaddress}/v1/boot/${hostname}/stage1.ipxe
   menu M-Lab iPXE boot menu: ${epoxyaddress}
     item --gap --   Production options:
     item stage1     -- Boot stage1 script
-    item localboot  -- Boot local disk
     item --gap
     item --gap --   Diagnostic options:
     item shell      -- iPXE Shell
@@ -84,7 +83,7 @@ set stage1_url https://${epoxyaddress}/v1/boot/${hostname}/stage1.ipxe
   echo Fetching stage1 script.
   goto firstfetch
 
-:retry_loop iseq ${retry_delay_s} ${max_retry_delay_s} && goto fetch_timeout_local_boot ||
+:retry_loop iseq ${retry_delay_s} ${max_retry_delay_s} && goto fetch_timeout_reboot ||
   inc retry_delay_s ${retry_delay_s}
   echo Sleeping ${retry_delay_s} seconds...
   sleep ${retry_delay_s}
@@ -121,7 +120,7 @@ set stage1_url https://${epoxyaddress}/v1/boot/${hostname}/stage1.ipxe
 
 
 :localboot
-:fetch_timeout_local_boot
-  echo Sleeping 10 seconds and booting from local hard drive.
+:fetch_timeout_reboot
+  echo Sleeping 10 seconds and rebooting system to retry.
   sleep 10
-  sanboot --no-describe --drive 0x80 || shell
+  reboot || shell