[PATCH proxmox-backup] fix #7305: client: restore: filter out last snapshot if not finished

[PATCH proxmox-backup] fix #7305: client: restore: filter out last snapshot if not finished

[Christian Ebner]

When invoking the client CLI for restore via:
```
proxmox-backup-client restore <snapshot> <archive> <target>
```
the <archive> provide a backup group only, in which case the client
falls back to restore the last snapshot of that group as convenience.

The snapshot listing used to identify the last snapshot returns
however all snapshots, including unfinished ones, a restore on an
unfinished one will therefore fail.

Fix this by filtering out snapshots which do not contain a manifest
file yet, and therefore are not considered as finished. Also adapt
the helper function name to reflect this change. Ideally this would
happen on the server side, guarded by some flag, but that will not
work with older server versions.

Fixes: https://bugzilla.proxmox.com/show_bug.cgi?id=7305
Signed-off-by: Christian Ebner <c.ebner@proxmox.com>
---
Note:
The behaviour can be easily tested by simply temporarily renaming the
index.json.blob in the last snapshot of a backup group to simulate an
ongoing backup for that snapshot.

 proxmox-backup-client/src/main.rs | 19 +++++++++++++++++--
 1 file changed, 17 insertions(+), 2 deletions(-)

diff --git a/proxmox-backup-client/src/main.rs b/proxmox-backup-client/src/main.rs
index 999e50205..7e20b9cc8 100644
--- a/proxmox-backup-client/src/main.rs
+++ b/proxmox-backup-client/src/main.rs
@@ -160,7 +160,7 @@ async fn api_datastore_list_snapshots(
     Ok(result["data"].take())
 }
 
-pub async fn api_datastore_latest_snapshot(
+pub async fn api_datastore_latest_finished_snapshot(
     client: &HttpClient,
     store: &str,
     ns: &BackupNamespace,
@@ -175,6 +175,21 @@ pub async fn api_datastore_latest_snapshot(
 
     list.sort_unstable_by(|a, b| b.backup.time.cmp(&a.backup.time));
 
+    // only once a snapshots contains the manifest it is considered finished
+    let last_is_finished = list[0]
+        .files
+        .iter()
+        .any(|content| content.filename == MANIFEST_BLOB_NAME.as_ref());
+
+    if !last_is_finished {
+        if list.len() > 1 {
+            log::info!("last snapshot of group {group} not finished, use previous instead");
+            return Ok((group, list[1].backup.time).into());
+        } else {
+            bail!("backup group {group} does not contain finished snapshots.");
+        }
+    }
+
     Ok((group, list[0].backup.time).into())
 }
 
@@ -187,7 +202,7 @@ pub async fn dir_or_last_from_group(
     match path.parse::<BackupPart>()? {
         BackupPart::Dir(dir) => Ok(dir),
         BackupPart::Group(group) => {
-            api_datastore_latest_snapshot(client, repo.store(), ns, group).await
+            api_datastore_latest_finished_snapshot(client, repo.store(), ns, group).await
         }
     }
 }
-- 
2.47.3

Re: [PATCH proxmox-backup] fix #7305: client: restore: filter out last snapshot if not finished

[Fabian Grünbichler]

On February 11, 2026 4:24 pm, Christian Ebner wrote:
> When invoking the client CLI for restore via:
> ```
> proxmox-backup-client restore <snapshot> <archive> <target>
> ```
> the <archive> provide a backup group only, in which case the client
> falls back to restore the last snapshot of that group as convenience.
> 
> The snapshot listing used to identify the last snapshot returns
> however all snapshots, including unfinished ones, a restore on an
> unfinished one will therefore fail.
> 
> Fix this by filtering out snapshots which do not contain a manifest
> file yet, and therefore are not considered as finished. Also adapt
> the helper function name to reflect this change. Ideally this would
> happen on the server side, guarded by some flag, but that will not
> work with older server versions.
> 
> Fixes: https://bugzilla.proxmox.com/show_bug.cgi?id=7305
> Signed-off-by: Christian Ebner <c.ebner@proxmox.com>
> ---
> Note:
> The behaviour can be easily tested by simply temporarily renaming the
> index.json.blob in the last snapshot of a backup group to simulate an
> ongoing backup for that snapshot.
> 
>  proxmox-backup-client/src/main.rs | 19 +++++++++++++++++--
>  1 file changed, 17 insertions(+), 2 deletions(-)
> 
> diff --git a/proxmox-backup-client/src/main.rs b/proxmox-backup-client/src/main.rs
> index 999e50205..7e20b9cc8 100644
> --- a/proxmox-backup-client/src/main.rs
> +++ b/proxmox-backup-client/src/main.rs
> @@ -160,7 +160,7 @@ async fn api_datastore_list_snapshots(
>      Ok(result["data"].take())
>  }
>  
> -pub async fn api_datastore_latest_snapshot(
> +pub async fn api_datastore_latest_finished_snapshot(
>      client: &HttpClient,
>      store: &str,
>      ns: &BackupNamespace,
> @@ -175,6 +175,21 @@ pub async fn api_datastore_latest_snapshot(
>  
>      list.sort_unstable_by(|a, b| b.backup.time.cmp(&a.backup.time));
>  
> +    // only once a snapshots contains the manifest it is considered finished
> +    let last_is_finished = list[0]
> +        .files
> +        .iter()
> +        .any(|content| content.filename == MANIFEST_BLOB_NAME.as_ref());
> +
> +    if !last_is_finished {
> +        if list.len() > 1 {
> +            log::info!("last snapshot of group {group} not finished, use previous instead");
> +            return Ok((group, list[1].backup.time).into());
> +        } else {
> +            bail!("backup group {group} does not contain finished snapshots.");
> +        }
> +    }

couldn't this be a combinator instead?

i.e., something like

    let last_finished_time = list
        .iter()
        .find_map(|snap| {
            snap.files.iter().find_map(|content| {
                if content.filename == MANIFEST_BLOB_NAME.as_ref() {
                    Some(snap.backup.time)
                } else {
                    None
                }
            })
        })
        .ok_or_else(|| {
            format_err!("backup group {group} does not contain any finished snapshots.")
        })?;

to account for multiple snapshots with missing manifests? those should
not occur "naturally", but might exist cause of accidents/..

> +
>      Ok((group, list[0].backup.time).into())
>  }
>  
> @@ -187,7 +202,7 @@ pub async fn dir_or_last_from_group(
>      match path.parse::<BackupPart>()? {
>          BackupPart::Dir(dir) => Ok(dir),
>          BackupPart::Group(group) => {
> -            api_datastore_latest_snapshot(client, repo.store(), ns, group).await
> +            api_datastore_latest_finished_snapshot(client, repo.store(), ns, group).await
>          }
>      }
>  }
> -- 
> 2.47.3
> 
> 
> 
> 
> 
> 

Re: [PATCH proxmox-backup] fix #7305: client: restore: filter out last snapshot if not finished

[Christian Ebner]

On 2/12/26 3:54 PM, Fabian Grünbichler wrote:
> On February 11, 2026 4:24 pm, Christian Ebner wrote:
>> +    // only once a snapshots contains the manifest it is considered finished
>> +    let last_is_finished = list[0]
>> +        .files
>> +        .iter()
>> +        .any(|content| content.filename == MANIFEST_BLOB_NAME.as_ref());
>> +
>> +    if !last_is_finished {
>> +        if list.len() > 1 {
>> +            log::info!("last snapshot of group {group} not finished, use previous instead");
>> +            return Ok((group, list[1].backup.time).into());
>> +        } else {
>> +            bail!("backup group {group} does not contain finished snapshots.");
>> +        }
>> +    }
> 
> couldn't this be a combinator instead?
> 
> i.e., something like
> 
>      let last_finished_time = list
>          .iter()
>          .find_map(|snap| {
>              snap.files.iter().find_map(|content| {
>                  if content.filename == MANIFEST_BLOB_NAME.as_ref() {
>                      Some(snap.backup.time)
>                  } else {
>                      None
>                  }
>              })
>          })
>          .ok_or_else(|| {
>              format_err!("backup group {group} does not contain any finished snapshots.")
>          })?;
> 
> to account for multiple snapshots with missing manifests? those should
> not occur "naturally", but might exist cause of accidents/..

Yes, good point! Would however opt for logging each skipped snapshot 
instead of silently progressing to the next one. This might help debug 
in cases where such issues come up.

Will send version 2 shortly, thanks!

superseded: [PATCH proxmox-backup] fix #7305: client: restore: filter out last snapshot if not finished

[Christian Ebner]

superseded-by version 2:
https://lore.proxmox.com/pbs-devel/20260213104304.375430-1-c.ebner@proxmox.com/T/