If bad data was committed into a Pachyderm input repository, you might need to delete a commit or surgically delete files from your history. Depending on whether or not the bad data is in the HEAD commit of the branch, you can perform one of the following actions:
- Delete the HEAD of a Branch. If the incorrect data was added in the latest commit and provided that the commit does not have children: Follow the steps in this section to fix the HEAD of the corrupted branch.
- If your changes are relatively recent (see conditions below), you can delete a particular file, then erase it from your history.
Additionally, although this is a separate use-case, you have the option to squash non-HEAD commits to rewrite your commit history.
Delete the HEAD of a Branch #
To fix a broken HEAD, run the following command:
pachctl delete commit <commit-ID>
When you delete a HEAD commit, Pachyderm performs the following actions:
- Changes HEADs of all the branches that had the bad commit as their HEAD to their bad commit’s parent and deletes the commit. The data in the deleted commit is lost. If the bad commit does not have a parent, Pachyderm sets the branch’s HEAD to a new empty commit.
- Interrupts all running jobs, including not only the jobs that use the bad commit as a direct input but also the ones farther downstream in your DAG.
- Deletes the output commits from the deleted jobs. All the actions listed above are applied to those commits as well.
This command will only succeed if the HEAD commit has no children on any branch.
pachctl delete commit will error when attempting to delete a HEAD commit with children.
Are you wondering how a HEAD commit can have children?
A commit can be the head of a branch and still have children.
For instance, given a
master branch in a repository named
repo, if you branch
master by running
pachctl create branch repo@staging --head repo@master, the
master’s HEAD will have an alias child on
Squash non-HEAD Commits #
If your commit has children, you have the option to use the
squash commit command.
Squashing is a way to rewrite your commit history; this helps clean up and simplify your commit history before sharing your work with team members.
Squashing a commit in Pachyderm means that you are combining all the file changes in the commits of a global commit
into their children and then removing the global commit.
This behavior is inspired by the squash option in git rebase.
No data stored in PFS is removed since they remain in the child commits.
pachctl squash commit <commit-ID>
- Squashing a global commit on the head of a branch (no children) will fail. Use
pachctl delete commitinstead.
- Squash commit only applies to user repositories. For example, you cannot squash a commit that updated a pipeline (Commit that lives in a spec repository).
- Similarly to
pachctl delete commit,
pachctl squash commitstops (but does not delete) associated jobs.
In the simple example below, we create three successive commits on the master branch of a repo
- In commit ID1, we added files A and B.
- In commit ID2, we added file C.
- In commit ID3, the latest commit, we altered the content of files A and C.
We then run
pachctl squash commit ID1, then
pachctl squash commit ID2, and look at our branch and remaining commit(s).
- A’ and C’ are altered versions of files A and C.
At any moment,
pachctl list file repo@master invariably returns the same files A’, B, C’.
pachctl list commit however, differs in each case, since, by squashing commits, we have deleted them from the branch.
Delete Files from History #
It is important to note that this use case is limited to simple cases where the “bad” changes were made relatively recently, as any pipeline update since then will make it impossible.
In rare cases, you might need to delete a particular file from a given commit and further choose to delete its complete history. In such a case, you will need to:
Create a new commit in which you surgically remove the problematic file.
Start a new commit:
pachctl start commit <repo>@<branch>
Delete all corrupted files from the newly opened commit:
pachctl delete file <repo>@<branch or commitID>:/path/to/files
Finish the commit:
pachctl finish commit <repo>@<branch>
Optionally, wipe this file from your history by squashing the initial bad commit and all its children up to the newly finished commit.
Unless the subsequent commits overwrote or deleted the bad data, the data might still be present in the children commits. Squashing those commits cleans up your commit history and ensures that the errant data is not available when non-HEAD versions of the data are read.
In the simple example below, we want to delete file C in commit 2. How would we do that?
pachctl list file repo@master returns the files A’, B, C’, E, F.
- A’ and C’ are altered versions of files A and C.
We create a new commit in which we surgically remove file C:
pachctl start commit repo@master pachctl delete file repo@master:path/to/C pachctl finish commit repo@master
At this point,
pachctl list file repo@masterreturns the files A’, B, E, F. We removed file C. However, it still exists in the commit history.
To remove C from the commit history, we squash the commits in which C appears, all the way down to the last commit.
pachctl squash commitID2 pachctl squash commitID3
It is as if C never existed.