doc: add basic troubleshooting guide for machines

Would be great to get some of these Tips & Tricks written down.

Refs: nodejs#775 (comment)

doc/fixing-machines.md

@@ -0,0 +1,109 @@
+# Fixing Machine Failures
+
+A basic guide to troubleshooting machine issues.
+
+## Jenkins job failures
+
+Usually the first sign of trouble is a failed Jenkins job. First thing to check
+is whether this is actually a Jenkins issue rather than a machine issue.
+
+1. Rebuild the job.
+2. Clean the workspace and rebuild the job.
+3. Check vital signs on the machine page
+   (https://ci.nodejs.org/computer/<machine>/)
+4. Take machine offline and rebuild, do other machines have the same problem?
+
+## On the machine itself
+
+Try to `su` to the Jenkins user if possible, this avoids accidentally doing
+things as root (which can destroy the machine, or break jobs).
+
+It's also good practice to raise/comment in a GitHub issue (or on IRC) to let
+others know what you did.
+
+### Out of space
+
+Most common problem is that a partition is full. You can quickly check with:
+
+```bash
+df -h # Or `df -m` on machines that don't support `-h`.
+```
+
+Look at the `%Used` column (not the `%Iused`, that's for inodes). If it's full
+it probably needs a cleanup.
+
+### Processes left behind
+
+```bash
+ps -ef | grep iojs | grep -v java | grep -v grep
+```
+
+If there are a lot of processes running on the machine, and nothing running on
+the machine according to Jenkins, then that's a warning flag. Paste the list of
+processes in an issue, and clean them up with:
+
+```bash
+# Show all iojs processes except the Jenkins process and grep, and kill them.
+ps -ef | grep iojs | grep -v java | grep -v grep | awk '{print $2}' | xargs kill
+```
+
+### Git or file permission issues
+
+Sometimes we get problems in job workspaces, either because someone left a file
+there as `root` that `iojs` can't remove, or because some git cleanup (like `git
+gc`) needs to happen.
+
+Jenkins job workspaces on machines can be wiped (as long as a job isn't running), so
+feel free to nuke a job's workspace, for example:
+
+```bash
+rm -rf /home/iojs/build/workspace/node-test-commit/
+```
+
+The next job will take longer as it re-clones the workspace.
+
+
+### Turn it off and on again
+
+Sometimes something weird happens, and it's easier to just reboot the machine.
+On Unix just do one of:
+
+```bash
+shutdown -r now
+# or:
+reboot
+```
+
+When the machine comes back the Jenkins slave should reconnect automatically
+(check on ci.nodejs.org). If it doesn't see the next step.
+
+### Restart the Jenkins agent
+
+Most machines have a service to restart the Jenkins agent. Try one of:
+
+```bash
+# Systemd init (some newer Linux distros):
+systemctl start jenkins.service
+# System V init (older Linux distros):
+/etc/init.d/jenkins start
+# AIX:
+/etc/rc.d/rc2.d/S20jenkins start
+# Things we don't have a service for yet:
+/home/iojs/start.sh
+```
+
+Service files should be stored [here][Jenkins Worker Template].
+
+## Problems with non-test machines
+
+Ask someone in [infra][Infra Admins] or [release][Release Admins] to take a look.
+
+## Machines needing reimaging or reconfiguring
+
+You need to run the ansible scripts on the machines again, see the [Ansible
+Readme][].
+
+[Infra Admins]: https://github.com/nodejs/build#infra-admins
+[Jenkins Worker Template]: https://github.com/nodejs/build/tree/master/ansible/roles/jenkins-worker/templates
+[Release Admins]: https://github.com/nodejs/build#release-admins
+[Ansible Readme]: https://github.com/nodejs/build/blob/master/ansible/README.md