Re: [pbs-devel] [PATCH proxmox-backup 2/2] backup/datastore: move manifest locking to /run

[Dominik Csapak <d.csapak@proxmox.com>]

On 12/2/20 2:50 PM, Wolfgang Bumiller wrote:
> On Wed, Dec 02, 2020 at 02:19:57PM +0100, Dominik Csapak wrote:
>> this fixes the issue that on some filesystems, you cannot recursively
>> remove a directory when you hold a lock on a file inside (e.g. nfs/cifs)
>>
>> it is not really backwards compatible (so during an upgrade, there
>> could be two daemons have the lock), but since the locking was
>> broken before (see previous patch) it should not really matter
>> (also it seems very unlikely that someone will trigger this)
>>
>> Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
>> ---
>>   src/backup/datastore.rs | 26 +++++++++++++++++++++++---
>>   1 file changed, 23 insertions(+), 3 deletions(-)
>>
>> diff --git a/src/backup/datastore.rs b/src/backup/datastore.rs
>> index 0f74ac3c..9cc88906 100644
>> --- a/src/backup/datastore.rs
>> +++ b/src/backup/datastore.rs
>> @@ -257,6 +257,12 @@ impl DataStore {
>>                   )
>>               })?;
>>   
>> +        // the manifest does not exists anymore, we do not need to keep the lock
>> +        if let Ok(path) = self.manifest_lock_path(backup_dir) {
>> +            // ignore errors
>> +            let _ = std::fs::remove_file(path);
>> +        }
>> +
>>           Ok(())
>>       }
>>   
>> @@ -698,13 +704,27 @@ impl DataStore {
>>           ))
>>       }
>>
> 
> please describe the path in a doc comment here

ok, but even in a private api?

> 
>> +    fn manifest_lock_path(
>> +        &self,
>> +        backup_dir: &BackupDir,
>> +    ) -> Result<PathBuf, Error> {
>> +
>> +        let mut path = PathBuf::from("/run/proxmox-backup/.locks/");
> 
> why `.locks` and not just `locks`? I don't see the benefit in "hidden"
> files in `/run`?

yeah you're right, no sense in making this hidden

> 
>> +        path.push(self.name());
>> +        path.push(backup_dir.group().backup_type());
>> +        path.push(backup_dir.group().backup_id());
>> +        std::fs::create_dir_all(&path)?;
> 
> Is there a particular reason you use a `PathBuf` here this way? Looks
> like you could just `format!()` it all the same? Since none of these
> types are `Path`s to begin with anyway.
> 
> Since those components are all strings, IMO you could work with a
> `String` from the start and only convert to PathBuf at the end.
> 
> Would save you the extra String allocation below.

ok will do

> 
> So if I see this right, the file will then be
> /run/proxmox-backup/.locks/$store/${type}/${id}/${timestamp}.index.json.lck
> 
> seems reasonable apart from the dot in `.locks` ;-)
> 
> However, do we really need the directory structure here?
> Shouldn't a flat `.../locks/${type}.${id}.${timestamp}.index.json` be
> fine as well? (I don't really mind, it would just be less code ;-) )

for now, ids do not really have a length limit besides the fs filename 
limit of 255 bytes
and since i had to factor that out, i did for datastore/type as well
(would look even weirder to use something like:
.../locks/${datastore}.${type}/${id}/${timestamp}.index.json.lck
)

though we probably should limit the id length anyway...

> 
>> +
>> +        path.push(format!( "{}{}", backup_dir.backup_time_string(), &MANIFEST_LOCK_NAME));
>> +
>> +        Ok(path)
>> +    }
>> +
>>       fn lock_manifest(
>>           &self,
>>           backup_dir: &BackupDir,
>>       ) -> Result<File, Error> {
>> -        let mut path = self.base_path();
>> -        path.push(backup_dir.relative_path());
>> -        path.push(&MANIFEST_LOCK_NAME);
>> +        let path = self.manifest_lock_path(backup_dir)?;
>>   
>>           // update_manifest should never take a long time, so if someone else has
>>           // the lock we can simply block a bit and should get it soon
>> -- 
>> 2.20.1