Anton,
> Since this fix made to speedup pme-free switch which prohibits the merges...
Sounds as a bit of knowledge that I missed. Good to understand it now.
> The "magic numbers" are always the "magic numbers" :)
> We must get rid of them to see problems covered by them.
Let's try =)
чт, 26
Ivan,
The "magic numbers" are always the "magic numbers" :)
We must get rid of them to see problems covered by them.
>> Was there any
>> performance measurements for such multiple left nodes cases?
Since this fix made to speedup pme-free switch which prohibits the merges,
the answer is "no".
Anton,
Thank you for your efforts! And sorry for a late reply.
I am a little bit familiar with tx recovery. I personally like the
idea of removing such "magic" logic from the code. I think a proper
way is either justify and sustain (tests, documentation) some behavior
or get rid of it.
Rechecked TC two more times.
Going to merge to master in case no objections here.
On Mon, Dec 23, 2019 at 1:44 PM Anton Vinogradov wrote:
> Igniters,
>
> One more PME optimization ready to be reviewed.
> I found a strange tx recovery delay caused by IGNITE_TX_SALVAGE_TIMEOUT.
> I've checked the
Igniters,
One more PME optimization ready to be reviewed.
I found a strange tx recovery delay caused by IGNITE_TX_SALVAGE_TIMEOUT.
I've checked the code and tests and found no reason to delay recovery.
So, the issue [1] is ready to be reviewed.
Improvement checked with benchmark [2] and fix,