Sun Fire 4810 Server
 
Sun Fire 3800 Server
 
Sun Netra 1290 Server
 
Sun Fire E6900 Server
 
Sun Fire 6800 Server
 
Sun Fire 4800 Server
 
Sun Fire E2900 Server
 
Sun Fire V1280 Server
 
Sun Fire E4900 Server
 
Sun Netra 1280 Server
 

PLA-Support>Sun Systems>SPARC>Enterprise>SN-SPARC: SF-x8x0/Ex900
 
_Old GCS Categories>Sun Microsystems>Servers>Midrange V and Netra Servers
 
_KM>Content>Video
 
_Old GCS Categories>Sun Microsystems>Servers>Midrange Servers
 

Applies to:

Goal

Instructions on how to gather data from a hung Sun Fire[TM] SF3800/SF4800/SF4810/SF6800/E4900/E6900/E2900/V1280.

Available for this topic, a brief how-to video tutorial that provides step-by-step instructions answering Sun's most frequently asked questions. View the video and/or follow the detailed instructions below.

Video - Troubleshooting a hung domain (5:00)

Solution

Please make sure to follow each step in the order in which it is presented.
The instructions for a Sun Fire[TM] E2900, V1280 and Netra[TM] 1280, 1290 are highlighted separately as they employ Lights Out Management(LOM) instead of the System Controller(SC) employed by the Serengeti/Amazon class of servers.

Instructions for Platforms employing the System controller(SC)

1. Ensure that the domain is actually hung:

       - Can you ping the domain?

      - Can you telnet to the domain?

2. Ensure that the SC (System Controller) is not hung, If you can access the System Controller, proceed to login to the SC and obtain a platform shell.

       A.If you get to the platform shell run the following commands:

               SCname:SC> showlogs

              SCname:SC> showplatform

       B. If the SC is hung See Document 1002033.1 for details on how to recover from a hung system controller. Then go back to step 2A. 

3. Once in the platform shell attempt to get a domain shell:

               SCname:SC> console -d 

- If the command appears to hang, then we need to send a break signal to the domain.

       - if you are using telnet: Press CTRL ]

        at the telnet prompt type: send break

       - if you are connected to the SC via tip: use ~#

At this point you should have a domain shell prompt, continue with the following commands, otherwise continue to step 4.

- If you get the domain shell run the following commands:

              SCname:A> showdomain -p status

              SCname:A> showlogs



Then type break to get to the OBP. if this takes you to the ok prompt then type sync to force a core file.

4. If you were not able to get to the ok prompt, then the system is really hung and we will need to send an XIR (externally initiated reset) to the domain.

From the domain shell type: reset

This command will give different behavior depending on what the OBP variable error-reset-recovery is set to.

If this variable is set to sync, a core file will attempt to be taken. If it is set to boot, then the system will just reboot as if the boot command was issued at the ok prompt.

If it is set to none it should drop you to the ok prompt, where you can run the following commands, the '#' sign represents the cpu that we took the XIR on, 

use that number in the cbuf command if possible run this command on each of the cpus (some depend on firmware level of the SC):

{#} ok dump-sigblock

{#} ok # cbuf

{#} ok .xir-state-all

- If you were not able to return to the ok prompt, but have a domain prompt type the following command:

SCname:A> showresetstate

5. If none of these tactics work you may be forced in to just powering off the domain.

If this is the case then do a setkeyswitch off for the domain.

Instructions for Platforms employing Lights Out Management(LOM)

1. Ensure that the domain is actually hung:

• Can you ping the domain?

• Can you telnet to the domain?

2. Login to the LOM prompt via telnet/ssh or tip.

    A. once you get the lom prompt, run the following commands:

        lom>showsc -v

       lom>showlogs -v

3. Try to connect to the domain and see what state it is in:

   A. use the console commands to connect to domain

       lom> console

   B. If there's no response from console, use escape sequence to break out. The default escape sequence is "#."

       lom>console

      #.

      lom>

   C. Once the domain is confirmed to be un-reachable, go to next step.

4. Using the 'break' or 'reset' command to recover.

   A. Try to break into the OBP by 'break' and if you get to OBP, do a sync to collect a corefile.

lom>break

This will suspend Solaris.

Do you want to continue? [no] yes

Type 'go' to resume

debugger entered.

{3} ok sync

  B. If 'break' does not work, a 'reset' has to be used and 'showresetstate' collected as well. The behaviour of reset also depends on the settings used in OBP 

     for error-reset-recovery which should preferably be set to 'sync'.

      lom>reset

      This will abruptly terminate Solaris.

Do you want to continue? [no] yes

      lom>showresetstate

5. If none of the procedures above work, a poweroff/poweron needs to be issued.

power off the platform

lom> poweroff

power on the platform, but do not start the domain

lom> poweron all

power on the platform and start the domain

lom> poweron

References

Following Manuals contain procedures and details about setting up platform loghost (where available) and configuring automatic recovery of hung domains:

• Sun Fire Midrange Systems Platform Administration Manual (Automatic Diagnosis and Recovery section)
• Sun Fire Midrange System Controller Command Reference Manual; it will include details about setupdomain parameters for hang automatic recovery (hang-policy, log-reset-data, verbose-reset-data, reset-data-ftp-url)

Also check:

• Doc 1008702.1: Console Logging Options to capture Fatal Reset output for Sun systems
• Doc 1018813.1: Sun Fire [TM] SF3800/SF4800/SF4810/SF6800 - E4900/E6900 Server: Domains running firmware 5.15.x or later with hang-policy set to "notify" may lose critical troubleshooting data

Product
Sun Fire 6800 Server
Sun Fire 4810 Server
Sun Fire 4800 Server
Sun Fire 3800 Server
Sun Fire E6900 Server
Sun Fire E4900 Server
Sun Fire E2900 Server
Sun Fire V1280 Server

Internal Section

NOTE: Procedures given in this document are dependent on OBP and SC versions

Keywords: System Controller, SC, Sun Fire, Serengeti, kernel, XIR

Previously Published As 46780

References

This solution has no attachment