Fix truncated stdin backups by closing pipe before signaling done

[arska]

Summary

Fix a race condition causing application-aware backups (stdin backups from pod exec) to be truncated. Multiple users reported database dumps losing the final few MB of data.

Root cause

In restic/kubernetes/pod_exec.go, the pipe writer was closed after signaling done:

// BEFORE (buggy ordering)
defer stdoutWriter.Close()  // deferred - runs after done signal
done <- true                // signals backup to call cmd.Wait()

This created a race: triggerBackup() receives the done signal and calls cmd.Wait() on restic before the pipe writer is closed. Restic's stdin may not have received all data and EOF yet.

Fix

Close the pipe writer before signaling done:

// AFTER (fixed ordering)
stdoutWriter.Close()  // ensures all data + EOF reaches restic
done <- true          // now safe to wait for restic to finish

On stream errors, CloseWithError() propagates the error through the pipe, then os.Exit(1) ensures the backup pod fails hard (preserving existing behavior for error cases).

Who is affected

Anyone using application-aware backups (k8up.io/backupcommand annotation) with large outputs (100MB+). Confirmed by multiple users with PostgreSQL and MariaDB dumps.

Fixes #1109

Checklist

• PR contains the label area:operator
• Commits are signed off
• I have not made any changes in the charts/ directory

Test plan

• Build passes
• Existing unit tests pass
• e2e test Improved K8up operator monitoring #11 (annotated failure) passes — validates error handling still works
• Test with a large (100MB+) database dump via backupcommand annotation

🤖 Generated with Claude Code

[tobru]

@Kidswiss I better leave this up to you for a review

[bastjan]

Does not fix the issue as the wait on done was unneeded. The code waits for the restic command completion which should only happen when the pipe is closed and read to completion.

Superseeded by #1183 which hopefully fixes the issue.