Axtell

« first day (76 days earlier) ← previous day next day → last day (558 days later) »

Downgoat

12:30 AM

@Mego is the server still down?

Anonymous

1:18 AM

Downgoat

1:34 AM

Omy

Anonymous

2:16 AM

Server lives again

Downgoat

4:01 AM

new (slightly less temporary but probably still temporary) landing page

Anonymous

I like it, except the penguin is gone :(

Downgoat

5:06 AM

I will hide some penguins somewhere as an easter egg maybe

Yay, username update work

@Mego will SQLAlchemy do cascade delete? e.g. if we delete answer it will automaticalyl delete related vote items etc.

Anonymous

@Downgoat It can. It just needs to be set up right.

Anonymous

Actually that needs to be set on the actual DB server

Anonymous

But it's not hard

Downgoat

5:28 AM

@Mego am I doing this right:

@server.route("/user/<int:user_id>")
@server.route("/user/<int:user_id>/<name>")
def get_user(user_id):
    pass

so I can do things like /user/5/Downgoat for my profile page

Anonymous

@Downgoat The first one should be @server.route("/user/<int:user_id>", defaults={"name": None}). Also add a name param to get_user.

Anonymous

Also flask-login.readthedocs.io/en/latest

Anonymous

^ should make auth easier

Downgoat

5:56 AM

I ran a profiler and looks like rendering only takes around 15ms per block. Flask itself only takes .5ms to render so not sure why some pages (the ones with lots of answers) take 9s to load while others load in 1s

Anonymous

Might just be fluctuations in server load (aka gremlins)

Anonymous

Once we move to dedicated hosting, that shouldn't be a problem

Downgoat

looks like I profiled the wrong page :P

not sure why the render function is being called almost 300 times though

Anonymous

Where do you see 300 calls?

Downgoat

top left cell

Anonymous

6:00 AM

That's for socket.recv

Downgoat

I assume that's waiting time for the STDOUT stream of the child process?

hm there's also a {method 'recv' of '_socket.socket' objects}

Anonymous

Either that or from the client

Anonymous

Both the various subprocesses that are called (SASS via webassets, node via browserify, node via markdown rendering, Celery stuff, etc.) and the network connection (thanks TCP) use sockets for communication, so it's not surprising at all that there are a bunch of calls to socket.recv

Downgoat

webassets/browserify is only called once and then those files are cached though

should only be celery/node communication

Anonymous

Oh I also need to set up caching on Flask and Celery properly

Anonymous

6:06 AM

Oh also connections to Redis and memcached use sockets for IPC

Downgoat

what is memcached used for?

Anonymous

Results backend for Celery

Anonymous

For smaller tasks (like Markdown rendering of reasonably-sized text), it's more efficient to use memcached for results than Redis

Downgoat

ah, that should yeah, be minimizing the amount of calls to node too

Anonymous

Since this server has 8 delicious GB of RAM, we can afford the memory usage of memcached to get better speed :P

Downgoat

6:20 AM

huh mysql is bulk of the slow-down:

{method 'recv' of '_socket.socket' objects}                                                                                         <-     281    5.345    5.345  /usr/local/lib/python3.6/site-packages/mysql/connector/network.py:234(recv_plain)

5.345s and it does 281 calls

oh i see

for every request we're doing a query for user info (even CSS files, static images, etc.)

that can't be all thoguh

seems like some queries are being repeated?:

2018-01-19 22:28:15,717 INFO sqlalchemy.engine.base.Engine SELECT answer_votes.user_id AS answer_votes_user_id, answer_votes.answer_id AS answer_votes_answer_id, answer_votes.vote AS answer_votes_vote
FROM answer_votes
WHERE answer_votes.answer_id = %(answer_id_1)s
2018-01-19 22:28:15,717 INFO sqlalchemy.engine.base.Engine {'answer_id_1': 120}
2018-01-19 22:28:16,447 INFO sqlalchemy.engine.base.Engine SELECT answer_votes.user_id AS answer_votes_user_id, answer_votes.answer_id AS answer_votes_answer_id, answer_votes.vote AS answer_votes_vote

6 hours later…