#django

jd|2: that is the same mechanism used by default to generate pkids on rows

omarek: did you put your alert inside a function that is never called, maybe?

Melamo: yes I am using postgres

doismellburning: yeah, I just felt the same

jd|2: you may have to drop down to raw sql to take advantage of the sequence data type along with some built-in functions to perhaps do your id generation at the DB level atomically, but that would be the safest most robust way since it's done at the DB level instead of the app level

moldy: in the first function. I assume it's automatically called if it's in head ?

omarek: no

you will have to call it through some mechanism

Okay, back to school.

jd|2: What happens when a record gets deleted?

moldy: thank you.

Melamo: I havn't worked with postgres before this, so dropping down to raw sql I was just avoiding

mattmcc: There is no interface to delete that object other than the DB level

jd|2: the alternative solution is to learn about PG's transaction isolation levels, be willing to up the isolation level to serializable, and be ready to re-try a transaction on failure... not fun, and not a solution I would recommend: http://www.postgresql.org/docs/9.4/static/trans...

jd|2: I guess the biggest question I haven't asked is: why not just stick with the standard int pk ids for your objects?

okay, i need to ind an example of this

of django LDAP authentication.

like, a full example.

toothe: this looks to be in the ballpark of what I would do myself: https://github.com/etianen/django-python3-ldap

Melamo: One of my co-worker warned that in conflict cases there is skip of pk

I mean the entire series skips some 1000 ints

Melamo: Is that true ?

Why is gaps in the sequence a problem?

jd|2: I don't think that is the case, but I'm not positive. It would be a good question to ask the folks in #postgresql. The default int pkids are generated using sequences. According to the docs (http://www.postgresql.org/docs/9.4/static/sql-c...) it looks like values my be assigned out of order to different sessions if the cache value is set to greater than 1, but I don't see anything about it skipping a

number completely

mattmcc: Its a product requirement, as these id's are not only for internal purpose, but the ID is linked to a physical form and the squence will be the mapping

make a human enter the id then

auto-generate a suggestion that can be overridden

moldy: That could be done, but isn't there a reliable way to generate a continuous sequence

there probably is, but sooner or later, somebody will want to change one of these IDs

Got it working.

Melamo: I read a few cases on stackoverflow as well, and they said sequence gaps are normal

jd|2: this looks relevant to what you are trying to do: http://bytes.com/topic/postgresql/answers/42334...

jd|2: it looks like one possible case if a transaction rollback after a sequence has been polled will not decrement the sequence, so you might very well be right

Melamo: So what would you suggest ?

jd|2: like I said, I would talk to some of the overly smart folks in #postgresql as it sounds like it's a bit more complicated than I initally thought

http://www.postgresql.org/message-id/501B1494.9... suggests using a separate counter table.

jd|2: concurrency is hard :)

jd|2: are you 110% sure it's a product requirement to never have gaps? what is created first, the physical form or the db record?

really hard (to get right)

moldy: Physical form

moldy: yeah they dont want gap, thats why we dont have option to delete

jd|2: and how do you map db records to physical forms?

supbuds :D

Form number are in a sequence, from 0 to 100,000

so my objects when created should have same ending sequence

What if someone tears up one of the forms? :P

jd|2: so the users have to create the objects in the right order?!

Melamo: yeah, my life was pretty simple, before concurrency came in

jd|2: it really sounds like you should just have the user enter the id

jd|2: concurrency becomes easy, then

mattmcc: thats a likely case, but as I was told that the guy will end up creating a dummy object, they have a format for that

Eh, I'd go with having the user enter the form number.

If it's a paper form that's already been printed, that seems by far to be the most straightforward approach.

and we have a active flag which is set flase if that object has to be deleted

moldy: i have a very basic knowledge of html, css, and python

its not what i do for a living or anything

i just really wonder if django could work. I don't know.

im so surprised there is nothing quite like wordpress but in python

emma: there are a number of plogging engines in python. https://wiki.python.org/moin/PythonBlogSoftware

mattmcc: moldy: Currently I am generating ID's with post_save signals and getting the row_count and as signals are synchronous so, I was wondering wont they lock the table before getting the row count ?

jd|2: No, that'd be horrible.

mattmcc: I am using signals for the first time, so can you please elaborate the horrible part

Locking the table just to get a row count, that destroys concurrency performance.

Postgres works really hard to minimize locks.

Does anyone know how to get the URL for a generic view function?

mattmcc: moldy: Melamo: Thanks for everything, I guess manual approach is the only thing I can go with for now.

I tried naming it, but that doesn't seem to work since it is expecting arguments

jd|2: if combined with an auto-suggestion, it is the best solution, imho.

shawnadelic: Then you need to provide arguments.

moldy: yeah, will check that as well :)

E.g, {% url 'pattern-name' arg=foo %}

jd|2: ask the table for the max number, add 1, and set that in the form's initial. if save() fails because the id was not unique, display an error and have the user enter another one (or do whatever). concurrency solved.

mattmcc: Okay, but I'm not sure what it needs since it's a ListView and it's basically just doing the query automatically

shawnadelic: Well, what does the urlpattern look like?

mattmcc: https://gist.github.com/shawnadelic/238eda6bb5e...

You can have a look at the URL and Model there

What's the full error?

NoReverseMatch at /

Reverse for 'all_articles' with arguments '()' and keyword arguments '{}' not found. 0 pattern(s) tried: []

moldy: what if I do the same when saving ?

by overriding the save method

Is that urls.py your root urlconf, or is it being include()'d?

It's being included

With a namespace, I imagine.

oh, okay, yeah, I think so

So i need articles:all_articles

mattmcc: Awesome, that was it, thanks!

I'm pretty new to django

jd|2: there is a possible race condition.

jd|2: between the max() and the save(), another request could have used up the id.

yeah

jd|2: but you could catch the exception and redisplay the form, i guess, yep

moldy: yeah, if there is conflict I can ask user to save again

that way without anyone manually entering the id will be generated and concurrency will not be a probem

Hey, got this error when using celery + gevent : DatabaseError: DatabaseWrapper objects created in a thread can only be used in that same thread. The object with alias 'default' was created in thread id 140372966528768 and this is thread id 72687120.

any ideas where this come from ?

jd|2: yes, but how do you make sure id matches the one on the paper form?

moldy: by allowing to edit it

haha

jd|2: something else to consider is explicit locking

jd|2: http://www.postgresql.org/docs/9.4/static/expli...

jd|2: create a table, where each row is a sequence value. Doing a select for update on the row to take out a lock on the counter row. Now you can safely read and increment it.

jd|2: I *think* this will be transaction safe as well, as in if your transaction rolls back, your counter increment will also roll back