Re: [HACKERS] make async slave to wait for lsn to be replayed

On 2020-03-27 04:15, Kartyshov Ivan wrote:
> Anna, feel free to work on this patch.

Ivan and I worked on this patch a bit more. We fixed the bugs that we
could find and cleaned up the code. For now, we've kept both options:
WAIT as a standalone statement and WAIT as a part of BEGIN. The new
patch is attached.

The syntax looks a bit different now:

- WAIT FOR [ANY | ALL] event [, ...]
- BEGIN [ WORK | TRANSACTION ] [ transaction_mode [, ...] ] [ WAIT FOR
[ANY | ALL] event [, ...]]
where event is one of:
LSN value
TIMEOUT number_of_milliseconds
timestamp

Now, one event cannot contain both an LSN and a TIMEOUT. With such
syntax, the behaviour seems to make sense. For the (default) WAIT FOR
ALL strategy, we pick the maximum LSN and maximum allowed timeout, and
wait for the LSN till the timeout is over. If no timeout is specified,
we wait forever. If no LSN is specified, we just wait for the time to
pass. For the WAIT FOR ANY strategy, it's the same but we pick minimum
LSN and timeout.

There are still some questions left:
1) Should we only keep the BEGIN option, or does the standalone command
have potential after all?
2) Should we change the grammar so that WAIT can be in any position of
the BEGIN statement, not necessarily in the end? Ivan and I haven't come
to a consensus about this, so more opinions would be helpful.
3) Since we added the "wait" attribute to TransactionStmt struct, do we
need to add something to _copyTransactionStmt() /
_equalTransactionStmt()?

--
Anna Akenteva
Postgres Professional:
The Russian Postgres Company
http://www.postgrespro.com