stop the merry-go-round - 2018-03-22 (page 1 of 2)

if a certain one gets picked you might get a slide

I might spell your name wrong on purpose though

can't be too nice

@JoeObbish I'd expect nothing less

Thanks to Sean, who loves Microsoft so much that he legally changed his last name

AHAHAHAHAHA

that's a good one liner opener

I need you in the crowd

you laugh at a lot of my jokes

2:35 AM

I don't get to go unless I am a speaker

except in the rare chance Sourabh asks me to man the clinic again

we've been over this

we'll do a gofundme

post a shirtless pic or something

get those donations rolling in

I don't think the crowd of men will want to see me shirtless

that's just for the gofundme

Though, having said that, I've been hit on by a bunch of men - I take it as a compliment

SON OF A @($UR(@(*#$*@#$*(@#%

FFS

keep it family friendly

2:41 AM

I can't see the history of the file so... I can't say which version it may have changed in

I'm beat, good night

good night

11:23 AM

@JoeObbish no worries, I'll poke at it today

@SeanGallardy-Microsoft hah no

I tore down my ha/dr labs rather than upgrade them from 2016

Felt tedious

1:27 PM

@JoeObbish a parallel deadlock involving columnstore doesn't seem to trigger the new columns for batch mode

1:39 PM

@sp_BlitzErik color me surprised

is that how that expression works?

color me shocked is nice as well

so last night I had a dream that I was the scheduler

assigning work to different numa nodes

am I not gonna make it?

bringing in @SeanGallardy-Microsoft for his thoughts as well

wew lad

1:58 PM

Have you tried installing the latest CU?

if you think about it

let's say you buy into the right and left brain stuff

could you say that our brains have two numa nodes?

I cannot - I don't know enough about the brain

can you prove that memory access isn't uniform?

you also have a frontal lobe -- is that l1 cache?

@sp_BlitzErik I don't know what an L1 cache is, but sure

> MFW

2:00 PM

hippity hoppity boy howdy

how many developers do you know that are aware of the L1 cache?

For SQL Server? Many.

For random Java "application" none

that spend as much time as you do talking to people like us?

well

perhaps a poor case study

you guys never talk about the L1 cache either

Sean briefly mentioned it when complaining about VMs

So I did talk about L1 cache

;)

2:06 PM

for a guy who spends as much time monkeying about with columnstore you should know about it too

how batch sizes and instructions are kinda decided so they fit better into it

unlike rowgroups nonsense

also fuck this plan spilling on the sorts and not on the exchanges

what a TURD

@sp_BlitzErik Sorry we're trying to give you good performance

but how is it practical?

if you wanted me to have good performance you'd show me the trace flags

I'll give you one digit

is it a middle finger?

2:17 PM

I see you got my double entendre

you act like i've never got rejected by a girl over 6' tall

smh

really though

would I write better code if I understood the L1 cache?

2:32 PM

perhaps how it relates to other parts of the hardware you're using, yes

it's not uncommon for developers to need to understand this stuff

blog.codinghorror.com/the-infinite-space-between-words

@sp_BlitzErik Legit though, I like average to short women - but recently there have been some tall women that... just wow... no words.

@JoeObbish You'd probably understand CS a little bit better... but now you have me CURIOUS... I do know that one of the big things was using SIMD(XMM) instructions for vectored or packed data processing which AFAIK is one of the ways Columnstore is fast

I wonder if a core with a larger L1/L2 cache at the same frequency can push the throughput higher in columnstore

maybe, but

i think primarily it would allow larger batch sizes?

like if you had stringy data in there

2:47 PM

That was my thought, but then I wonder if we'd just pack more batches in

like a packed bcd

maybe, not sure how much hardware detection the code does

A fair amount

D:

Don't look so shocked, Hekaton required specific cpu commands so we need to chek fo rit

are the checks purely for compatibility or do they change perf characteristics?

2:54 PM

@SeanGallardy-Microsoft I have zero CS training

sometimes it hurts

@SeanGallardy-Microsoft that is an interesting question

but I got no way to test that

and very little influence over hardware purchasing decisions

I mean I'm on a VM right now

CS = columnstore

oh

well that's true as well, technically

lol

hahaha

oh goddammit

columnstore deadlocks don't track database id

now you finally understand the despair of columnstore

you thought it was all fun and games, eh?

sunshine and rainbows?

3:05 PM

That's all rowgroup

it's a deadlock on a table with only a clustered columnstore index

YERRAROWGROUP

3:25 PM

we got hundreds of thousands of queries

not really sure where they are coming from

run profiler

profiler is deprecated

I want to use XE with the fancy bucketizer

wonder how many attempts that'll take to get right

deprecated but still appreciated yo

@JoeObbish So is sysprocesses and SP_WHO BUT EVERYONE FRIGGING USES IT LIKE IT WAS BRAND NEW AND GAVE THEM SEXUAL FAVORS

but I'm not like... upset or anything

3:47 PM

i use activity monitor and sp_who3 exclusively

> associatedObjectId="72057594059948032"

> SELECT OBJECT_NAME(72057594059948032)

> Arithmetic overflow error converting expression to data type int.

this is some gag

APRIL FOOLS

I really need to stop writing novella length emails

just throw a couple gifs in and call it a day

Now, the question is... "JIFS" or "GIFS", how do you say it?

hard g

Like SQL - S,Q,L, or SEQUEL like they did in deliverence

3:55 PM

sequel

i'm not european

What about meme?

MAYMAY

lulz

> mfw Erik says he's not European then says "zed" and "haych" for the lulz

yeah i love zedbras

Need Joe and Paul to weigh in on this

Since isn't Paul technically APAC?

4:00 PM

they won't let nz or tasmania in

too unruly

LOL

I needed some good jokes this AM

all writing and no gym makes Sean a dull boy with anger issues

yeah i can't wait to go back

all day i dream about squats

you want a good joke?

ADIDAS

> waitresource="ROWGROUP: 8:72057594059948032:0"

oh wait...

4:04 PM

if only there were a columnstore guru about

a man who needs 2.5 hours at pass to talk about it

2.5? holy balls.

@sp_BlitzErik it's not an object

I've run into that before

I think it's an HOBT

it is

but why is the deadlock telling me about it instead of the object id

yo

that's how stuff is tracked

I see it in other places too

lame af

4:11 PM

welcome to columnstore

Right? Who uses heaps?

@SeanGallardy-Microsoft I LOVE HEAPS

@JoeObbish YOU SHUT YOUR MOUTH WHEN YOURE TALKING TO ME

Great movie

the 2.5 hour session is about the ETL project

there's so much good stuff to talk about

if you want great shitty dad lines, watch this boy's life

4:14 PM

Soooooo how about you send me a copy of the recording?

since I won't be at pass

he said he'd do it via webex for everyone in this chat room

What a mench

@SeanGallardy-Microsoft I don't know what that means

but yes, of course you are welcome as well Sean

@JoeObbish As per usual on this keyboard I misspelled the word

Mensch

Mensch (Yiddish: מענטש‎ mentsh, cognate with the German word Mensch meaning a "human being") means "a person of integrity and honor". The opposite of a "mensch" is an "unmensch", meaning an utterly unlikeable or unfriendly person. According to Leo Rosten, the Yiddish maven and author of The Joys of Yiddish, a "mensch" is "someone to admire and emulate, someone of noble character. The key to being 'a real mensch' is nothing less than character, rectitude, dignity, a sense of what is right, responsible, decorous." The term is used as a high compliment, implying the rarity and value of that individual...

4:29 PM

too bad he's not a mensch on the bench iikwim

4:41 PM

@SeanGallardy-Microsoft Is this XE private? •XEvent: XeSqlPkg::calculate_dop

I can't seem to find anything for dop downgrades...

nevermind, apparently that's a call stack

still hoping that an XE exists for downgrades

ok

does it?

XE exists for DOP downgrades?

adjust_dop_for_joes_queries

it's in the debug items

Pretty sure once DOP is calculated that's what used, it won't/can't be downgraded

5:00 PM

i think what joe is looking for is reasons why his queries aren't reading their dop potential

Maybe they were bullied at optimizer school?

feedback.azure.com/forums/908035-sql-server/suggestions/…

5:15 PM

I figured it out

this is such a stupid issue

do tell!

if the query runs in the default workload group

it doesn't get downgraded

if it runs in a different workload group

it gets downgranded

definitions are the same

I can't see how this isn't a bug

5:32 PM

so what anime are you watching

how can you ignore my moment of triumph???

finding a bug in resource governor is like finding a dead body in antarctica

you don't even use it

how do you know that?

5:52 PM

exactly. hardly anyone uses it and whoever does shouldn't be surprised.

6:04 PM

you can't do a serious columnstore implementation without it

it appears you can't even do one with it ;)

I can on sql server 2016

perhaps sql server 2016 is the new 2008 R2

@sp_BlitzErik food wars season 4 next month

so hypu~

6:27 PM

i think if you were in better shape you'd like better anime