[Tarantool-patches] [PATCH 1/1] wal: simplify rollback

Konstantin Osipov kostja.osipov at gmail.com
Sun May 3 13:46:29 MSK 2020


* Vladislav Shpilevoy <v.shpilevoy at tarantool.org> [20/05/01 22:16]:

This isn't a particularly bad patch. It changes one fairly trivial
implementation to another. You claim it simplifies the rollback, I
believe it makes it more convoluted with the rest of the code. 

In any case it's a matter of taste - so perhaps your taste is
better than mine.

But the reason you had to do this patch *only* because you want to manage
rollback in tx - as you claim yourself here:

> 
> Also there is now a single route for commit and rollback cbus
> messages, called tx_complete_batch(). This change will come in
> hand in scope of synchronous replication, when WAL write won't be
> enough for commit. And therefore 'commit' as a concept should be
> washed away from WAL's code gradually. Migrate to solely txn
> module.

The reason it has to be managed in TX is that you don't want to
build synchronous replication on top of in-memory relay, which
would allow you to manage RAFT rollback in WAL.

When I called you out on the fact that you can't simply build your
implementation on top of existing code without an in-memory relay,
and it will require a bunch of hacks you got mad at me.

Now what? You will ask Nikita to review this and Kirill will
"LGTM" it and you will proceed with your own design - and even may
release it as a "stable" version over some time.

This "stable version may be even useful to some people. 

Will it be worth a dime on a worldwide scale? No.

-- 
Konstantin Osipov, Moscow, Russia


More information about the Tarantool-patches mailing list