Re: [pbs-devel] [PATCH v3 proxmox-backup 45/58] client: pxar: add method for metadata comparison

[Christian Ebner <c.ebner@proxmox.com>]

On 4/5/24 10:14, Christian Ebner wrote:
> On 4/5/24 10:08, Fabian Grünbichler wrote:
>> Quoting Christian Ebner (2024-03-28 13:36:54)
>>> Adds a method to compare the metadata of the current file entry
>>> against the metadata of the entry looked up in the previous backup
>>> snapshot.
>>>
>>> If the metadata matched, the start offset for the payload stream is
>>> returned.
>>>
>>> This is in preparation for reusing payload chunks for unchanged files.
>>>
>>> Signed-off-by: Christian Ebner <c.ebner@proxmox.com>
>>> ---
>>> changes since version 2:
>>> - refactored to new padding based threshold
>>>
>>>   pbs-client/src/pxar/create.rs | 31 ++++++++++++++++++++++++++++++-
>>>   1 file changed, 30 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/pbs-client/src/pxar/create.rs 
>>> b/pbs-client/src/pxar/create.rs
>>> index 79925bba2..c64084a74 100644
>>> --- a/pbs-client/src/pxar/create.rs
>>> +++ b/pbs-client/src/pxar/create.rs
>>> @@ -21,7 +21,7 @@ use pbs_datastore::index::IndexFile;
>>>   use proxmox_sys::error::SysError;
>>>   use pxar::accessor::aio::{Accessor, Directory};
>>>   use pxar::encoder::{LinkOffset, PayloadOffset, SeqWrite};
>>> -use pxar::Metadata;
>>> +use pxar::{EntryKind, Metadata};
>>>   use proxmox_io::vec;
>>>   use proxmox_lang::c_str;
>>> @@ -466,6 +466,35 @@ impl Archiver {
>>>           .boxed()
>>>       }
>>> +    async fn is_reusable_entry(
>>> +        &mut self,
>>> +        previous_metadata_accessor: &mut 
>>> Directory<LocalDynamicReadAt<RemoteChunkReader>>,
>>> +        file_name: &Path,
>>> +        stat: &FileStat,
>>> +        metadata: &Metadata,
>>> +    ) -> Result<Option<u64>, Error> {
>>> +        if stat.st_nlink > 1 {
>>> +            log::debug!("re-encode: {file_name:?} has hardlinks.");
>>> +            return Ok(None);
>>> +        }
>>
>> it would be nice if we had a way to handle those as well.. what's the 
>> current
>> blocker? shouldn't we be able to use the same scheme as for regular 
>> archives?
>>
>> first encounter adds (possibly re-uses) the payload and remembers the 
>> offset,
>> subsequent ones just add another reference/meta entry?
> 
> True, this is a leftover from the initial approach with the appendix 
> section instead of the split archive where it caused issues.
> 

Hardlinks will be encoded as such with the upcoming version of the 
patches, this however required additional changes to the pxar encoder, 
such that it returns the LinkOffset also for the `add_payload_ref` calls 
and requires an additional HashSet on the Archiver to remember cached 
regular files for which the payload chunks have already been looked up, 
so that the encountered, hard-linked file does not lead to re-injected 
chunks again.