Julia Kreger 5f6b94db45 Update troubleshooting text for custom IPA images

Additional text to help users understand how to gain access to a
running IPA image for troubleshooting purposes.

Change-Id: I337f02f7fa496312893bff8bb2bfb67dadb67cd9

2015-10-23 16:44:37 -04:00

6.7 KiB

Raw Blame History

Troubleshooting

Firewalling

Due to the nature of firewall settings and customizations, bifrost does not change any local firewalling on the node. Users must ensure that their firewalling for the node running bifrost is such that the nodes that are being booted can connect to the following ports:

67/UDP for DHCP requests to be serviced
69/UDP for TFTP file transfers (Initial iPXE binary)
6301/TCP for the ironic API
8080/TCP for HTTP File Downloads (iPXE, Ironic-Python-Agent)

If you encounter any additional issues, use of tcpdump is highly recommended while attempting to deploy a single node in order to capture and review the traffic exchange between the two nodes.

NodeLocked Errors

This is due to node status checking thread in ironic, which is a locking action as it utilizes IPMI. The best course of action is to retry the operation. If this is occurring with a high frequency, tuning might be required.

Example error:

NodeLocked: Node 00000000-0000-0000-0000-046ebb96ec21 is locked by
host $HOSTNAME, please retry after the current operation is completed.

Unexpected/Unknown failure with the IPA Agent

New image appears not to be deploying

When deploying a new image with the same previous name, it is necessary to purge the contents of the TFTP master_images folder which caches the image file for deployments. The default location for this folder is /tftpboot/master_images.

Additionally, a playbook has been included that can be used prior to a re-installation to ensure fresh images are deployed. This playbook can be found at playbooks/cleanup-deployment-images.yaml.

Building an IPA image

Troubleshooting issues involving IPA can be time consuming. The IPA developers HIGHLY recommend that users build their own custom IPA images in order to inject things such as SSH keys, and turn on agent debugging which must be done in a custom image as there is no mechanism to enable debugging via the kernel command line at present.

IPA's instructions on building a custom image can be found at: http://git.openstack.org/cgit/openstack/ironic-python-agent/tree/imagebuild/coreos/README.rst

This essentially boils down to the following steps:

git clone https://git.openstack.org/openstack/ironic-python-agent
cd ironic-python-agent
pip install -r ./requirements.txt

#. If you don't already have docker installed, execute: sudo apt-get install docker docker.io
cd imagebuild/coreos
Edit oem/cloudconfig.yml and add --debug to the end of the ExecStart setting for the ironic-python-agent.service unit.
Execute make to complete the build process.

Once your build is completed, you will need to copy the images files written to the UPLOAD folder, into the /httpboot folder. If your utilizing the default file names, executing cp UPLOAD/* /httpboot/ should achieve this.

Since you have updated the image to be deployed, you will need to purge the contents of /tftpboot/master_images for the new image to be utilized for the deployment process.

Obtaining IPA logs via the console

By default, bifrost sets the agent journal to be logged to the system console. Due to the variation in hardware, you may need to tune the parameters passed to the deployment ramdisk. This can be done, as shown below in ironic.conf:
```
agent_pxe_append_params=nofb nomodeset vga=normal console=ttyS0 systemd.journald.forward_to_console=yes
```
Parameters will vary by your hardware type and configuration, however the systemd.journald.forward_to_console=yes setting is a default, and will only work for systemd based IPA images such as the default CoreOS image.

The example above, effectively disables all attempts by the kernel to set the video mode, defines the console as ttyS0 or the first serial port, and instructs systemd to direct logs to the console.
Once set, restart the ironic-conductor service, e.g. service ironic-conductor restart and attempt to redeploy the node. You will want to view the system console occurring. If possible, you may wish to use ipmitool and write the output to a log file.

Gaining access via SSH to the node running IPA on CoreOS based images

If you wish to SSH into the node in order to perform any sort of post-mortem, you will need to do the following:

Set an sshkey="ssh-rsa AAAA....." value on the agent_pxe_append_params setting in /etc/ironic/ironic.conf
You will need to short circuit the ironic conductor process. An ideal place to do so is in ironic/drivers/modules/agent.py in the reboot_to_instance method. Temporarily short circuiting this step will force you to edit the MySQL database if you wish to re-deploy the node, but the node should stay online after IPA has completed deployment.
ssh -l core <ip-address-of-node>

Gaining access via SSH to the node running IPA for custom images

Custom built images will require a user to be burned into the image. Typically a user would use the diskimage-builder devuser element to achieve this. More detail on this can be located at:

https://github.com/openstack/diskimage-builder/tree/master/elements/devuser

Example:

export DIB_DEV_USER_USERNAME=customuser
export DIB_DEV_USER_PWDLESS_SUDO=yes
export DIB_DEV_USER_AUTHORIZED_KEYS=$HOME/.ssh/id_rsa.pub
disk-image-create -o /path/to/custom-ipa debian ironic-agent devuser

`ssh_public_key_path is not valid`

Bifrost requires that the user who executes bifrost have an SSH key in their user home, or that the user defines a variable to tell bifrost where to identify this file. Once this variable is defined to a valid file, the deployment playbook can be re-run.

Generating a new ssh key

See the manual page for the ssh-keygen command.

Defining a specific public key file

A user can define a specific public key file by utilizing the ssh_public_key_path variable. This can be set in the group_vars/inventory/all file, or on the ansible-playbook command line utilizing the -e command line parameter.