Re:Django project optimization guide – Part 1
by git-pull on Tuesday June 20, 2017 @07:55AM
(#14593807)
This is my first time hearing about Silk. Thanks for mentioning that, looks like a dream for profiling. GitHub for clicking: https://github.com/django-silk/silk
Re:Django project optimization guide – Part 1
by avelis on Tuesday June 20, 2017 @10:40AM
(#14594981)
I help maintain django-silk. I wasn't the original creator but that person has virtually no time to maintain it.
I wanted to take a moment here and tell the HN community thanks for all the contributions/support the library has received to continue along. It can definitely use all the support it can get. Silk has so much potential but works as a free tool for many.
Re:Django project optimization guide – Part 1
by Alex3917 on Tuesday June 20, 2017 @11:12AM
(#14595203)
Silk is great for making sure queries aren't run multiple times unnecessarily, but definitely don't run it is production. There are a lot of subtle threading issues that, as far as I know, haven't quite gotten resolved.
Re:Django project optimization guide – Part 1
by danpalmer on Tuesday June 20, 2017 @10:34AM
(#14594929)
For the profiling, I'd also recommend `run_profile_server` from django-extensions, and Snakeviz as a visualiser for that data. It records cprofile reports per-request so can be useful for seeing what takes the time in a single request.
Common hot-spots I've noticed (other than n+1 queries) is deserialisation of database rows into Django ORM objects - it's relatively expensive, and template rendering. I'd definitely choose Jinja2 on a new project for templating as it's more generally more capable and approximately 2x faster, the downside being the goal of "logic-less templates" is slightly harder to achieve.
Re:Django project optimization guide – Part 1
by sharmi on Tuesday June 20, 2017 @11:55AM
(#14595597)
Jinja2 has some crucial features missing in django templates, but we can add this to the django template as filters and functions.
On the other hand, many django plugins are written to accommodate only django templates. By choosing jinja2, you will be severely limiting what plugins you can use.
Jinja2 is more like a superset of django templates with some minor syntax modifications.
So I would recommend you go with django templates and bring in jinja2 only if needed (which would be 0.001% of the time)
This is from experience where I went with Jinja2 for one project and learned my lesson. Now for the next one I just went with the default. Some of django template's choices are bit of an hindrance but an affordable price for the interoperability it provides.
Re:Django project optimization guide – Part 1
by elevensies on Tuesday June 20, 2017 @03:10PM
(#14597538)
I'm not who you asked, but the first thing that comes to my mind is contrib.auth.views, which look for templates at the default locations or the locations you provide (as described here: https://docs.djangoproject.com/en/1.11/topics/auth/default/#... )
So using non-django templates would rule out this kind of thing right?
Re:Django project optimization guide – Part 1
by danpalmer on Tuesday June 20, 2017 @10:31AM
(#14594899)
I was quite excited to see this, but I feel like almost everything in this post was "table stakes" for production Django, and nothing particularly groundbreaking. Having said that, I work on a large Django codebase, maybe these things aren't as common in smaller projects.
To add, if using Django Debug Toolbar, there is a line_profiler [1] extension for it: https://github.com/dmclain/django-debug-toolbar-line-profile...
[1] https://github.com/rkern/line_profiler
I wanted to take a moment here and tell the HN community thanks for all the contributions/support the library has received to continue along. It can definitely use all the support it can get. Silk has so much potential but works as a free tool for many.
Previously I've hacked in printing out SQL statements, but the logging configuration is a lot nicer.
Common hot-spots I've noticed (other than n+1 queries) is deserialisation of database rows into Django ORM objects - it's relatively expensive, and template rendering. I'd definitely choose Jinja2 on a new project for templating as it's more generally more capable and approximately 2x faster, the downside being the goal of "logic-less templates" is slightly harder to achieve.
On the other hand, many django plugins are written to accommodate only django templates. By choosing jinja2, you will be severely limiting what plugins you can use.
Jinja2 is more like a superset of django templates with some minor syntax modifications.
So I would recommend you go with django templates and bring in jinja2 only if needed (which would be 0.001% of the time)
This is from experience where I went with Jinja2 for one project and learned my lesson. Now for the next one I just went with the default. Some of django template's choices are bit of an hindrance but an affordable price for the interoperability it provides.
We've been using Jinja2 at my startup since the beginning; in the four years I'be been working here I can't recall any issues with plug-ins using it.
So using non-django templates would rule out this kind of thing right?
Looking forward to Part 2!