NODE SYNC on a single node instance ?

mwellbro
Xpert

Hi all,

this is an "offshoot" from a different thread ( https://community.exasol.com/t5/discussion-forum/scd-type-2-in-exasol/m-p/6428#M1417 ) which I didn´t want to "bog down" with a peripheral topic - so, @exa-Gabi  , since you asked:

mwellbro_0-1632338984074.png

The system this was run on is an EXAOne , version: 7.0.9 (+ OS patchlevel 202104111020) , running inside a VirtualBox 6.1
with 4 GB RAM allocated and ( at the time ) 1 CPU.

The very concept, like you said, seems strange since NODE SYNC as far as I previously understood them are used to show
"inter-node lag" ( which is kind of hard to produces with only one node ).
The only thing I could think of would be that this is also used to track "inter-thread-lag", which given a single CPU is more or less
probable...

I also re-ran the test set after gearing up the VM with 4 CPUs and didn´t see NODE SYNCs on the MERGE steps where I previously
encountered them ( just a bit on the INSERT part ).
Given what else is running on my "host" which is a Intel i5 2320 quadcore, could be supporting my assumption - then again, I´m just guessing here 😉 .

As for your question regarding which SQL could be used to reproduce this, check back with the "scenario" posted in the original thread: https://community.exasol.com/t5/discussion-forum/scd-type-2-in-exasol/m-p/6990/highlight/true#M1557
The NODE SYNCs were visible in 2 MERGE steps.

Looking forward to anything you dig up 🙂

Cheers,
Malte

1 ACCEPTED SOLUTION

Accepted Solutions

exa-Gabi
Team Exasol
Team Exasol

This is actually pretty easy to reproduce and is certainly an anomaly/bug occurring on single node systems only - the time and resources are charged on NODE SYNC which for single node systems makes no sense. It is now included in our internal bug tracker and will be fixed sometime in a future release.

EXA-Dude, database developer, history buff. Member of Team Exasol.

View solution in original post

2 REPLIES 2

mwellbro
Xpert

Hi again,

just to add an additional observation: the occurence of NODE SYNCs does seem to correlate with the presence or lack of CPU-resources in this scenario.

Taking the previous test case we can "push" the NODE SYNC stats higher by constricting CPU resources ( which in my case was done by limiting CPU-time from the HOST available to the EXAOne-VM) :

mwellbro_0-1632565704889.png

While having the obvious effect of increasing general runtime it also creates higher NODE SYNCs on the INSERT ( I do have a feeling that the CONNECT BY is the culprit here due to it´s CPU demand ) - and add to that: the NODE SYNCs on the MERGE steps return:

mwellbro_1-1632565845330.png

On a side note: We can limit/scale up CPU resources at runtime in this manner ? Haven´t thought of this before, but this
does allow for quite an elastic setup ( reminds me that I once posted an IDEA to make exasol "CPU hot plug capable" , https://community.exasol.com/t5/ideas/making-exasolution-cpu-hotplug-capable/idi-p/4006 , but if I can scale like this and without downtime ? Groovy 🙂 - will of course need further testing).

Cheers,
Malte

exa-Gabi
Team Exasol
Team Exasol

This is actually pretty easy to reproduce and is certainly an anomaly/bug occurring on single node systems only - the time and resources are charged on NODE SYNC which for single node systems makes no sense. It is now included in our internal bug tracker and will be fixed sometime in a future release.

EXA-Dude, database developer, history buff. Member of Team Exasol.

View solution in original post