← Back to team overview

maria-developers team mailing list archive

Re: Review of MDEV-162: Enhanced semisync replication

 

Jonas Oreland <jonaso@xxxxxxxxxx> writes:

> I think it might be easier you just took and applied it to your favorite
> branch (hence i'll won't send
> a new version of the patch).

Ok.

> however there is one think that I discovered that needs to be fixed with
> the interaction with
> the Dump Thread Enhancement...I'll add that patch to this or the other JIRA
> entry.

Ok.

>> > ---replace_result 305 304
>> >  show status like 'Rpl_semi_sync_master_yes_tx';
>>
>> I'm curious why you decided to remove this --replace_result?
>> I see in bzr history that Monty added this --replace_result without any
>> explanation why... :-/

> i removed it cause the result didn't contain 304 or 305
> my guess is that it hasn't for several years, and hence it was pure
> obfuscation.

Hehe, nice... thanks for the explanation.

>> Any reason not to use --source include/have_log_bin.inc instead?
>>
>
> hmm...damn it, don't remember...
> i guess it would work equally well (or even better from mtr.pl point of
> view)

Right, I was just curious, not that important.

>> > +  /** First(or alone) in a group commit */
>> > +  BINLOG_GROUP_COMMIT_LEADER = 2,
>> > +
>> > +  /** Last(or alone) in a group commit */
>> > +  BINLOG_GROUP_COMMIT_TRAILER = 4
>> >  };
>>
>> What is the reason for introducing these flags?
>> As far as I can see from the patch, they are set, but never read?

> this is for a (yet) unpublished optimization, that is
> to not sem-sync all individual transactions in a group-commit
> but only the last one (trailer) (and i added leader for completeness).

Ok, that sounds cool!

Thanks,

 - Kristian.


Follow ups

References