On Sat, Dec 18, 2010 at 11:19 PM, Tom Keiser <[email protected]> wrote:
> On Dec 17, 2010 11:17 PM, "Matt W. Benjamin" <[email protected]> wrote:
>>
>> Hi Tom,
>>
>> I've read the tag-length-value draft specification multiple times quickly,
>> and once carefully.  Though I'm not expert in volser, the proposed protocol
>> seems very coherent and reasonable.
>>
>> Though it's a nit, I find some of the long token constants (e.g.,
>> "VOLSERTAGUNSUPPORTEDENCODING") hard to read.  It might be nice to break up
>> the token namespace with understores (e.g.,
>> "VOLSER_TAG_UNSUPPORTED_ENCODING"), if the token length exceeds some minumum
>> (14 characters seems plausible)?  This is already done for most token
>> namespaces in the draft.
>>
>
> Sounds good.  I'll make the necessary changes for draft -04.
>
>> I wondered also if the tag cache coherence ambiguity could be helped
>> through the use of some type of data version or serialization token in
>> relevant calls?
>>
>
> Hmm.  That's an interesting idea.  Proposal: If we were to add an OUT
> parameter to each Get...TLV RPC, which communicated the generation number
> for the tag namespace, then--combined with rx epoch checks--the caller would
> know when to re-fetch the list of available tags.
>
> Although I doubt this issue is likely to come up very often in the real
> world, an additional four byte payload hardly seems objectionable...
>
> Does that sound reasonable and sufficient?  If so, I can incorporate the
> idea into draft -04...
>
> Thanks very much for the feedback.

That sounds like an excellent plan to me.
_______________________________________________
AFS3-standardization mailing list
[email protected]
http://lists.openafs.org/mailman/listinfo/afs3-standardization

Reply via email to