CLOUD BACKUP & PHOTO LIBRARIES
rclone sync makes a destination match a source. That is useful for a controlled mirror, but it also means files present only at the destination can disappear.
Quick answer
Confirm direction and filters, run a dry run and inspect proposed deletions. Use copy when you want transfers without removing extra destination files.
Steps to follow
-
Before you start
Identify the authoritative source and the exact destination account and prefix. Preserve destination-only material separately. Prepare nested test folders and a destination-only sample so the preview reveals both layout and deletion behavior.
-
Set up the right scope
Configure the remote and identify a small source and an isolated destination prefix. List both locations first. Reversing source and destination changes which files are authoritative.
-
Apply the relevant settings
Preview with
rclone sync /path/source remote:test --dry-run --verbose. Read the proposed changes, including excluded files and deletions. Replace example paths only after identifying the actual locations. -
Complete the first run
Run the reviewed command without
--dry-run, then inspect logs and compare the destination. Add automation only once the same task behaves correctly on an unchanged second run.
Decisions and checks
The limitation that matters
Sync is one-way matching, not versioned backup or two-way conflict resolution. A mistaken deletion at the source can propagate. Use a separate versioned backup or a deliberately configured backup directory if recovery is required.
How to verify the result
Create a disposable extra file in the test destination and observe its treatment in a dry run. Confirm your filters and recovery policy before using the command on important data.
Keep the result reliable
Recheck mounts, filters and credentials before unattended mirrors, especially after storage changes. Inspect failures and missed jobs, and periodically recover an older file from the separate retention arrangement. Avoid concurrent writers that invalidate the reviewed source state.
Make a one-way mirror predictable before running it unattended
State which side is allowed to change the other
Write the intended direction in ordinary language before constructing the command. For example, the local published-assets folder is authoritative and the remote delivery folder should reflect it. That statement determines the source and destination arguments. If both locations receive independent edits, a simple one-way mirror is not automatically appropriate. Two reversed sync commands do not provide a considered conflict-resolution system. Preserve unique destination material before the first trial and use a dedicated disposable prefix. A label such as backup does not establish the intended semantics: it might be an archive that should retain old files or a mirror that should remove them, and those are different jobs.
Use a sample that reveals the directory layout
Create a source folder containing a small document and a named subdirectory. Add another harmless file only at the test destination. Preview the exact paths you intend to use and inspect where each file would land. Rclone transfers a directory's contents rather than automatically adding an extra level with the source directory's name. Quote paths appropriately and confirm the account behind the remote. A valid command pointed at a parent folder can affect a much larger scope than intended. List both locations before removing the dry-run option. The destination-only sample is particularly useful because it makes the proposed deletion behavior visible instead of leaving that crucial part of the mirror untested.
Read the proposed changes and filter rules together
Review updates, missing files and deletions in the preview, not just a total byte count. Filters determine which objects participate, and special options can alter how excluded destination content is treated. Keep the filter file with the job description and document why each rule exists. Do not copy a deletion option from a performance example without understanding its effect. Repeat the preview whenever the paths or filters change. It describes the state observed at that time, not a permanent authorization for future operations. If the next run unexpectedly proposes widespread deletion, stop and inspect the source and exclusions before trying to make the output look cleaner.
Verify that a mounted source is the intended source
An external disk or network share can disappear while leaving a valid empty directory behind. A scheduler that checks only whether that directory exists may proceed against the wrong state. Confirm the actual mount and a known identification file or expected content before allowing a mirror. Design the job to stop visibly when the source is uncertain. Rclone's handling of errors is not a substitute for this check because a wrongly empty directory may be a valid readable source from the operating system's perspective. Test the missing-source condition with disposable data. Also consider the scheduled user's permissions and mount visibility, which can differ from the interactive account used during initial setup.
Do not confuse copy with historical protection
Rclone copy preserves additional destination files that are absent from the source, unlike a mirror that deletes them. It can still update matching paths, so switching from sync to copy does not automatically preserve every earlier version. Define retention separately if you need to recover an accidental edit. A supported backup directory, provider versioning or a dedicated backup application can form part of that design, but verify its exact behavior and limits. Keep the historical recovery location outside the scope of the mirror that might remove it. Practice retrieving a deliberately changed sample from the earlier state before relying on the job for important working files.
Automate only the tested environment
Run the trial through the same user and scheduling mechanism that will handle the production task. Use explicit paths and protected access to the required credentials. Record exit results and useful logs without exposing secrets. Prevent overlapping operations against the same destination unless the chosen arrangement deliberately coordinates them. After the first completed mirror, run again without source changes and investigate unexpected work rather than assuming it is normal. Check the contents of representative destination files. An empty queue or successful exit cannot prove that the source selection included every required folder. Expand the job in reviewed stages so each new path has a clear expected effect.
Example: old client deliveries must remain available
Suppose a studio wants to upload the current project while retaining completed deliveries from previous months. Mirroring the current project into the root of that archive would make those older destination-only files candidates for deletion. Use separate dated delivery locations or a properly planned copy operation instead, while considering collisions at matching filenames. Test the chosen structure with a sample containing an older delivery. This shows why the retention requirement must come before the command choice. The phrase upload everything says nothing about whether existing material should remain, be updated or disappear after the transfer.
Compare the options
| Situation / component | Purpose / action | What to check |
|---|---|---|
| Destination must match current source | Review sync with a dry run | Extra destination files can be removed |
| Old deliveries must remain | Use separate delivery paths or review copy | Matching names can still be updated |
| Source appears unexpectedly empty | Verify mount and known content | Directory existence alone is insufficient |
Frequently asked questions
Is sync bidirectional?
No. It modifies the destination to match the source within the selected scope. If both sides legitimately change, use a supported two-way workflow with an understood conflict policy.
Does copy retain every old version?
No. It avoids removing extra destination files but can update files at matching paths. Historical recovery needs a separate configured and tested retention method.
Official documentation
This guide combines current official documentation with practical recovery checks. It does not claim hands-on testing of your installation. Menu labels and supported procedures can vary by release; follow the documentation for your installed version.