Re: XLogInsert scaling, revisited

On 2013-07-17 15:46:00 +0530, Amit Kapila wrote:
> On Monday, July 08, 2013 2:47 PM Heikki Linnakangas wrote:
> > Ok, I've committed this patch now. Finally, phew!
>
> Few doubts while reading the code:
>
> 1. Why in function WALInsertSlotAcquireOne(int slotno), it does
> START_CRIT_SECTION() to
> Lock out cancel/die interrupts, whereas other places call
> HOLD_INTERRUPTS()

A crit section does more than just stopping interrupts. They also ensure
that errors that occur while inside one get converted to a PANIC. That
seems apt for SlotAcquire/Release. Although the comments could possibly
improved a bit.

> 2. In function GetXLogBuffer(), why the logic to wakeup waiters is
> different when expectedEndPtr != endptr;
> When the wakeupwaiters is done in case expectedEndPtr == endptr?

I am not sure what you're asking here. We wakeup waiters if
expectedEndPtr != endptr because that means the wal buffer page the
'ptr' fits on currently has different content. Which in turn means we've
finished with the last page and progressed to a new one. So we wake up
everyone waiting for us.
WakeupWaiters() doesn't get passed expectedEndPtr but expectedEndPtr -
XLOG_BLCKSZ (up to there we are guaranteed to have inserted
successfully). And we're comparing with the xlogInsertingAt value which
basically measures up to where we've successfully inserted.

> 3.
> static bool
> ReserveXLogSwitch(..)
>
> In above function header, why EndPos_p/StartPos_p is used when
> function arguments are EndPos/StartPos?

I guess that's bitrot...

Greetings,

Andres Freund

--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services