Advanced Sync Workflows

Last updated on 2026-08-27 | Edit this page

Estimated time: 25 minutes

Overview

Questions

  • How do I download large datasets from cloud to HPC?
  • How do I implement bidirectional workflows safely?
  • How can I archive deleted files instead of removing them?
  • How do I sync selectively by file type or directory?

Objectives

  • Download datasets from cloud storage to HPC
  • Implement safe bidirectional syncing patterns using separate folders
  • Use --backup-dir to archive old files during sync
  • Apply filters with --include and --exclude for selective syncing

Downloading Datasets from Cloud to HPC

A collaborator has shared a large dataset on OneDrive. Download it to Sagehen HPC for processing:

This example uses a second remote named collaborator-onedrive – configured with the same rclone config process as Episode 3, choosing onedrive instead of box. If you haven’t set one up, substitute a remote you do have (such as pomona-box:) to follow along; otherwise rclone will report didn't find section in config file.

BASH 

rclone sync collaborator-onedrive:/large-dataset /scratch/raw-data --dry-run

Once satisfied, execute:

BASH 

rclone sync collaborator-onedrive:/large-dataset /scratch/raw-data

If the collaborator updates the dataset, re-run sync to get only the changes.

Important: Watch /scratch Space

Large datasets can fill /scratch quickly. Monitor your usage:

BASH 

rclone size /scratch/raw-data

Remember that /scratch is non-persistent SSD storage that is deleted when your job completes.

Bidirectional Syncing (Separate Folders)

The Challenge

Sometimes you want to sync in both directions without conflicts or accidental deletions.

The Safe Pattern: Separate Sync Directories

Instead of syncing the same folder both ways, use separate folders:

HPC:
  /work/myproject/source/      <- You edit files here
  /work/myproject/from-cloud/  <- Cloud files sync here

Cloud (Box):
  /myproject/hpc-updates/      <- Your HPC updates go here
  /myproject/shared/           <- Shared files everyone edits here

Setup Commands

From HPC, copy new HPC-generated files to Box:

BASH 

rclone copy /work/myproject/source pomona-box:/myproject/hpc-updates

Download shared files from Box to HPC:

BASH 

rclone sync pomona-box:/myproject/shared /work/myproject/from-cloud

Why This Works

  • Each direction has its own folder
  • No conflicts between local and remote edits
  • You explicitly decide what goes where
  • No accidental deletions from automatic syncing

Backing Up Deleted Files with –backup-dir

When syncing, --backup-dir moves deleted files to an archive folder instead of permanently removing them:

BASH 

rclone sync /scratch/results pomona-box:/results \
  --backup-dir pomona-box:/results-archive

Before:

HPC /scratch/results:  file1.txt, file2.txt
Box /results:          file1.txt, file2.txt, file3.txt (old)

After sync:

Box /results:          file1.txt, file2.txt
Box /results-archive:  file3.txt  <- Moved here instead of deleted!

To recover an archived file:

BASH 

rclone copy pomona-box:/results-archive/file3.txt pomona-box:/results/

Selective Syncing with Filters

Filter by Extension

Sync only certain file types:

BASH 

rclone sync pomona-box:/large-dataset /scratch/data \
  --include "*.csv" \
  --exclude "*"

Filter by Directory

Sync only a specific subdirectory:

BASH 

rclone sync pomona-box:/large-dataset/month-2026-03 /scratch/march-data

Bandwidth Limiting

To avoid overwhelming the network during large syncs:

BASH 

rclone sync pomona-box:/data /scratch/data \
  --bwlimit 500M  # Limit to 500 MB/s

Summary

Task Command Destructive?
Backup HPC results rclone copy /scratch pomona-box:/backup No
Mirror HPC to cloud rclone sync /scratch pomona-box:/mirror Yes
Download dataset rclone sync pomona-box:/data /scratch/data Yes
Archive deleted files Add --backup-dir to sync No (archived)
One-way bidirectional Use separate folders + copy No
Challenge

Challenge 8.1: Cross-Cloud Transfer

Copy files from Box to OneDrive (using copy, not sync, to be safe):

BASH 

echo "Box data" > ~/test.txt
rclone copy ~/test.txt pomona-box:/copy-test
rclone copy pomona-box:/copy-test pomona-onedrive:/copy-test
rclone cat pomona-onedrive:/copy-test/test.txt

Record: Did the file successfully transfer from Box to OneDrive?

After running the cross-cloud copy:

BASH 

rclone cat pomona-onedrive:/copy-test/test.txt

Expected output: Box data

Cross-cloud transfers work seamlessly with rclone copy – rclone downloads from one provider and uploads to the other without saving to local disk.

Challenge

Challenge 8.2: Backup with Archive

Set up a sync with --backup-dir to archive deleted files:

BASH 

mkdir -p ~/backup-test
echo "File 1" > ~/backup-test/file1.txt
echo "File 2" > ~/backup-test/file2.txt

# Sync to Box
rclone sync ~/backup-test pomona-box:/backup-test \
  --backup-dir pomona-box:/backup-test-archive

# Delete locally and sync again
rm ~/backup-test/file2.txt
rclone sync ~/backup-test pomona-box:/backup-test \
  --backup-dir pomona-box:/backup-test-archive

# Check archive
rclone ls pomona-box:/backup-test-archive

Record: Is file2.txt now in the archive folder instead of deleted?

After the second sync, check the archive:

BASH 

rclone ls pomona-box:/backup-test-archive

Expected output:

        7 file2.txt

The --backup-dir flag moved file2.txt to the archive instead of permanently deleting it. This provides a safety net for recovering accidentally removed files.

Key Points
  • One-way mirroring (sync from HPC to cloud) is appropriate for daily results; use separate folders for bidirectional workflows
  • The –backup-dir flag preserves deleted files in an archive instead of permanently removing them during sync
  • Selective syncing with –include and –exclude filters allows you to sync only specific file types or directories
  • Bandwidth limiting with –bwlimit prevents large transfers from overwhelming the shared network