diff --git a/general/deploy/failures.md b/general/deploy/failures.md
index 62d31251150a8c164d5d52d5f21b58c7cb2d739a..3803278db70a31cef6dc257c4973af442c4db2bd 100644
--- a/general/deploy/failures.md
+++ b/general/deploy/failures.md
@@ -97,6 +97,18 @@ If you don't have access to the nodes, ask the SRE on call for GitLab Production
 /chatops run oncall production
 ```
 
+### Preparation job discovered different versions running
+
+Sometimes a preparation job may find that one or more hosts are running
+different versions, resulting in the job failing. When this happens, you'll need
+to first determine why this is happening, and if it's safe to ignore the
+failure.
+
+If it's safe to ignore, go to [the deployer
+settings](https://ops.gitlab.net/gitlab-com/gl-infra/deployer/-/settings/ci_cd)
+and set the CI variable "SKIP_OMNIBUS_ROLE_CHECK" to `true`, then retry the job.
+Make sure to remove the variable when the prepare job passes.
+
 ### Fleet Deploy
 
 * `E: Could not get lock /var/lib/apt/lists/lock - open (11: Resource