[Tarantool-patches] [PATCH] limbo: introduce request processing hooks

[Serge Petrenko]

12.07.2021 18:48, Cyrill Gorcunov пишет:
> Guys, there are some moments even in current code structure which
> looks somehow strange so I would like to discuss them. Lets consider
> the following case: one replic sends us a promote request (assume
> we're sitting in term 2 and max term is 2).
>
> applier 1
> ---------
> applier_apply_tx
>    (promote term = 3
>     current max term = 2)
>    applier_synchro_filter_tx
>    apply_synchro_row
>      journal_write
>        (sleeping)
>
> at this moment another applier comes in with obsolete
> data and term 2
>                                applier 2
>                                ---------
>                                applier_apply_tx
>                                  (term 2)
>                                  applier_synchro_filter_tx
>                                    txn_limbo_is_replica_outdated -> false
>                                  journal_write (sleep)
>
> applier 1
> ---------
> journal wakes up
>    apply_synchro_row_cb
>      set max term to 3
> return to applier read and
> applier 2 could finish its write
> and wake up too
>
> at this moment the data from applier 2 is actually queued
> for write as valid but we just wrote the term 3, so if we would
> had been updating terms map earlier (before jornal write) the data
> from applier 2 should be NOPified. I think there is some problem
> because due to journal write lag the data is not as it could be
> if terms map updated early. Serge, Vlad, am I right? Which consequences
> can be here? IOW, should not we track terms earlier even without
> my filter series?
Looks like a bug, indeed.

We may either introduce a limbo latch or start tracking terms before the
WAL write.

I'm starting to like the idea with limbo latch more.

It's come up a couple of times already for various occasions, so maybe 
it's time to
finally implement it.

-- 
Serge Petrenko