Re: question about transaction operation in ForCES protocol draft!

[Jamal Hadi Salim <[e-mail]>]

Dear Chuanhuang Li,

I think i understand you now - i apologize it took me 3 emails;->

You have been saying: TRCOMP matters only for the success case but not
for the failure case.
You are right in the case where we assume that the CE will be failing
the transaction immediately as the text currently reads - which is more
reasonable in the case of a single FE; however, it is feasible that a
failure may trigger the CE to do something else to make some resource
available.

In any case, i dont think the extra TRCOMP message is incorrect in the
case of failure. I also dont think we really should be making an update
to the protocol at this stage of the game for something that is an
optimization - I will let Avri who is the editor and shepherd of the
draft to make that decision.

cheers,
jamal

On Mon, 2007-24-12 at 19:26 +0800, Chuanhuang Li wrote:
> Dear jamal:
>     Thanks very much!
>     The following is my viewpoints:
> 1:As you said,yes,a single FE on the NE is just one case.But what I
> said is also fit for multiple FEs.
>
> 2:Only the FEs which executed the transaction messages successfully
> wait for the TRCOMP.The FEs which failed will clear the state they
> have kept immediately ,because the information which they have saved
> for possible rollback is useless,and thay needn't wait for the TRCOMP.
>
> 3:The CE needn't tell the FE "please purge state on failure without
> waiting for TRCOMP".
> For multiple FEs:
>    The whole transaction failed,if CE recevied one COMMIT-RESPONSE
> message which indicated the execution failed from any FE.in this
> case,CE only need send COMMIT message(abort) to the FEs which executed
> successfully,it needn't send the TRCOMP!
>    The whole transaction succeed,if CE recevied COMMIT-RESPONSE
> message which indicated the exec! ution succeed from all FEs!in this
> case ,CE need send TRCOMP to all FEs.
> For sigle FE:
>    It is the same with multiple FEs.If CE received COMMIT-RESPONSE
> message which indicated the execution failed,it needn't send the
> TRCOMP!