#musicbrainz-devel

that's harder because it's closed edits too

yeah; I'd looked into that but I couldn't figure out how to fix that one

:(

we may need to kick up the statistics target for some columns of the edit table, I dunno

it's really weird though, because plenty of people with plenty more edits work fine

and plenty of people with plenty less edits also work fine

it's choosing a query plan that's sort of silly

so

basically the thing is

it's stupid? :P

It's out to get me

above about 20% of the table being returned a sequential scan is best

ANd jeroenbot

below that an index scan usually is

however right around that boundary it's often sort of a toss-up

whihc is where navap and jeroenbot are

it's choosing a sequential scan but probably should be doing an index in those cases

but I'm not sure how to convince it of that :)

I don't understand what you mean by 20% of the table is being returned with a sequential scan

Can you elaborate?

20% of 20 million is 4 million edits...

looking at the ticket it's a frequency of about .0025 or higher that it's choosing a sequential scan instead of using edit_idx_editor

Isn't that 10,000? I've got 40,000+ edits

this applies to people with more edits, e.g. nikki/jesus as well, but for them it's the correct choice

it loads in about 5 seconds for me

yeah, you're near the upper end of the people having trouble

jeroenbot has 70,000

basically the issue is, sequential reads are easier than random reads

indexes are random reads

so it'll chose a sequential scan if it thinks it'll be touching enough of the table that the benefits of sequential over random reads will be better

it's wrong about where that boundary is though

and I don't know how to make it smarter about that

it may have something to do with the recency of edits, too -- the plan it's choosing is a backwards sequential scan

databases are hard :P

or perhaps more simply: I got lucky yesterday figuring out how to fix the /edit/open thing :)

heh

I notice that rika doesn't have a problem

hm, true

(I think I'd looked at that before and decided why, but I forget, so)

jeroenbot takes quite long on rika though

ALthough it does eventually load

the query is SELECT edit.id, edit.editor, edit.open_time, edit.expire_time, edit.close_time, edit.data, edit.language, edit.type, edit.yes_votes, edit.no_votes, edit.autoedit, edit.status, edit.quality FROM edit WHERE (editor = ?) ORDER BY id DESC OFFSET ? LIMIT 500 btw

and yeah, it's fast on my system, but it's using the index scan

hm

Thankfully (unfortunately) I don't edit much these days so I don't have to refer to my edits as much

what's in common here is that rika and my thing both have the default default_statistics_target (of 100), while totoro has 300

so I guess that's probably why

But today I introduced someone to MB and to give them an example of how people edit I tried to show them my edit queue. They weren't impressed :p

haha

I also showed them the IA today and tried to relate it by talking about the wayback machine.

They tried to view an old google page and kept getting 50x ater 50x

Not a good demo

heh

It all started from the 10PB sticker on my laptop :)

Anyone know if there's a way to attach a branch on GitHub to an issue?

it's annoying, as I recall

but possible

I don't remember specifics other than you have to use curl to their API, you can't do it through their web interface

I'll DDG it.

SO to the rescue!

ah, yeah, that

heh

navap, you haven't edited much lately

it's having to scan the whole way back to edit 15496502 evidently

Oh, fancy linking

(locally, that's 4,918,231 edits from current)

uh, what? I've edited much more recently

hm

that's supposedly your 500th-most-recent edit

That's possible

My editing numbers went down drastically around the time NGS released

(by contrast: nikki's is from the third of january :P)

:P

It hasn't edited since 2010

ah

yeah; the 'bad' plan it's choosing is doing an index scan backwards on edit_pkey

i.e., it's looking at every single edit in descending order by ID

of course I still have no idea why it's choosing that plan :/

There must be several users who have the same problem as mine, I wonder why no one else has said anything

I guess it seems like this is most likely to affect people who used to edit a lot but don't edit as much now -- which is a perfect category for not noticing it's broken, I guess, if people are on a downswing in participation

ianmcorvidae: If it's any help, CatCat's edit page always 50x's on me.

(To give you another data point to compare with, if needed, etc.)

well, it certainly gets the same query plan

could you comment on the ticket to add that note though? easy to forget the list :)

Quick link to the ticket? :)

MBS-3793

heh, once again it uses the index locally

But, hey, even if you don't know *how* to fix it, it seems you've found *what* needs fixing.

That's a big step.

well, not really

I know the way to solve it for right now given the data we have

(other than actually how to make postgres do it, but you know what I mean)

but I don't really know what needs fixing to make it not just break again when we get another million edits, or something

and yeah, catcat's edits are getting the same problematic query plan, which results in postgres having to scan through more than a million edits before it gets all of the most recent 500 :/

anyone knows who is "mwiencek" (on github) here ? i'm still trying to match MB/github/irc nicks ;)

zas: that's bitmap

ty ;)

'lalinsky' on github is 'luks' here, and I forget what Mineo's github name is

i'm zas, everywhere.

:)

ah, mineo is mineo on github

that's good

I've likewise chosen the completely different name 'ianmcorvidae' for github ;)

caller#6 isn't "caller#6" anywhere, because most places don't allow hash in the username. :|

yeah, he's usually callerNo6 or something

but that's understandable enough

ianmcorvidae: You should do something to really mindfuck with people and call yourself ianncorvidae or iamcorvidae etc.

hah

ianmcorvidae: People would recognise it being you, but they'd never be able to spell it right. >:D

haha

hrglgrgl comes to mind

hrglgrmpf, yes :P

btw, i'm trying to work on Picard, but it seems it takes ages to have anything get in, or even few reviews on PRs. It just doesn't work. Not that my PRs are good or anything, but it is very hard to have feedback or help to improve them.

:)

ianmcørvıdæ clearly

navap: Only people already cannot spell his name.

yeah, we don't have many people working on picard :/

navap: Ian's name is easy enough to spell, so add or remove or change a single character, and it'll mess with people.

ianmcovridae

that one would be fun :P

heh

about it (Picard), it doesn't support https, but it seems MB will require it soon

zas: Poke Mineo and bitmap and luks on IRC until they give sound. Wait a day or two. Then repeat.

ianmcorvidae: :D

nah, requiring HTTPS will be a ways off

Freso: then i need a bot ;)

ianmcorvidae: What about user authentication?

it's required for oauth, but considering that luks is the one who wrote MB's (pending) oauth support in the first place...

navap: for the /login page, sure, but that's all

for the webservice digest auth is still totally fine

zas: I doubt it'll be hard to add HTTPS support though.

picard certainly doesn't use mb.org/login

And OAuth still isn't merged into beta, let alone test or anywhere.

So no immediate plans to move to digest+ssl? Or is that not really required?

Freso: i agree, but since no one seems to work on this software.... it may be takes longer ;)

But Python supports HTTPS easily enough, and I'm sure the Qt stack that I've read several people praise wouldn't make the built-in support worse.

zas: Like ianmcorvidae said, luks is the principal maintainer of Picard, and he's also the one who wrote the OAuth patch - partially becuase he wanted OAuth for Picard.

well, what i meant is that Picard wasn't designed with https in mind, looking at hardcoded "http://..." string concat in it

zas: I don't think a Picard patch for OAuth support is far off once the patch actually moves closer to beta/master of musicbrainz-server.