Peterbe.com

Suppose you have a complex Django QuerySet query that is somewhat costly (in other words slow). And suppose you want to return:

1. The first N results
2. A count of the total possible results

So your implementation might be something like this:

def get_results(queryset, fields, size):
    count = queryset.count()
    results = []
    for record in queryset.values(*fields)[:size]
        results.append(record)
    return {"count": count, "results": results}

That'll work. If there are 1,234 rows in your database table that match those specific filters, what you might get back from this is:

>>> results = get_results(my_queryset, ("name", "age"), 5)
>>> results["count"]
1234
>>> len(results["results"])
5

Or, if the filters would only match 3 rows in your database table:

>>> results = get_results(my_queryset, ("name", "age"), 5)
>>> results["count"]
3
>>> len(results["results"])
3

Between your Python application and your database you'll see:

query 1: SELECT COUNT(*) FROM my_database WHERE ...
query 2: SELECT name, age FROM my_database WHERE ... LIMIT 5

The problem with this is that, in the latter case, you had to send two database queries when all you needed was one.
If you knew it would only match a tiny amount of records, you could do this:

def get_results(queryset, fields, size):
-   count = queryset.count()
    results = []
    for record in queryset.values(*fields)[:size]:
        results.append(record)
+   count = len(results)
    return {"count": count, "results": results}

But that is wrong. The count would max out at whatever the size is.

The solution is to try to avoid the potentially unnecessary .count() query.

def get_results(queryset, fields, size):
    count = 0
    results = []
    for i, record in enumerate(queryset.values(*fields)[: size + 1]):
        if i == size:
            # Alas, there are more records than the pagination
            count = queryset.count()
            break
        count = i + 1
        results.append(record)
    return {"count": count, "results": results}

This way, you only incur one database query when there wasn't that much to find, but if there was more than what the pagination called for, you have to incur that extra database query.

Imagine you have something like this in Django:

class MyModel(models.Models):
    last_name = models.CharField(max_length=255, blank=True)
    ...

The most basic sorting is either: queryset.order_by('last_name') or queryset.order_by('-last_name'). But what if you want entries with a blank string last? And, you want it to be case insensitive. Here's how you do it:

from django.db.models.functions import Lower, NullIf
from django.db.models import Value


if reverse:
    order_by = Lower("last_name").desc()
else:
    order_by = Lower(NullIf("last_name", Value("")), nulls_last=True)


ALL = list(queryset.values_list("last_name", flat=True))
print("FIRST 5:", ALL[:5])
# Will print either...
#   FIRST 5: ['Zuniga', 'Zukauskas', 'Zuccala', 'Zoller', 'ZM']
# or 
#   FIRST 5: ['A', 'aaa', 'Abrams', 'Abro', 'Absher']
print("LAST 5:", ALL[-5:])
# Will print...
#   LAST 5: ['', '', '', '', '']

This is only tested with PostgreSQL but it works nicely.
If you're curious about what the SQL becomes, it's:

SELECT "main_contact"."last_name" FROM "main_contact" 
ORDER BY LOWER(NULLIF("main_contact"."last_name", '')) ASC

or

SELECT "main_contact"."last_name" FROM "main_contact" 
ORDER BY LOWER("main_contact"."last_name") DESC

Note that if your table columns is either a string, an empty string, or null, the reverse needs to be: Lower("last_name", nulls_last=True).desc().

My personal blog was a regular Django website with jQuery (later switched to Cash) for dynamic bits. In December 2021 I rewrote it in NextJS. It was a fun journey and NextJS is great but it's really not without some regrets.

Some flashpoints for note and comparison:

React SSR is awesome

The way infinitely nested comments are rendered is isomorphic now. Before I had to code it once as a Jinja2 template thing and once as a Cash (a fork of jQuery) thing. That's the nice and the promise of JavaScript React and server-side rendering.

JS bloat

The total JS payload is now ~111KB in 16 files. It used to be ~36KB in 7 files. :(

Before

After

Data still comes from Django

Like any website, the web pages are made up from A) getting the raw data from a database, B) rendering that data in HTML.
I didn't want to rewrite all the database queries in Node (inside getServerSideProps).

What I did was I moved all the data gathering Django code and put them under a /api/v1/ prefix publishing simple JSON blobs. Then this is exposed on 127.0.0.1:3000 which the Node server fetches. And I wired up that that API endpoint so I can debug it via the web too. E.g. /api/v1/plog/sort-a-javascript-array-by-some-boolean-operation

Now, all I have to do is write some TypeScript interfaces that hopefully match the JSON that comes from Django. For example, here's the getServerSideProps code for getting the data to this page:

const url = `${API_BASE}/api/v1/plog/`;
const response = await fetch(url);
if (!response.ok) {
  throw new Error(`${response.status} on ${url}`);
}
const data: ServerData = await response.json();
const { groups } = data;

return {
  props: {
    groups,
  },
};

I like this pattern! Yes, there are overheads and Node could talk directly to PostgreSQL but the upside is decoupling. And with good outside caching, performance never matters.

Server + CDN > static site generation

I considered full-blown static generation, but it's not an option. My little blog only has about 1,400 blog posts but you can also filter by tags and combinations of tags and pagination of combinations of tags. E.g. /oc-JavaScript/oc-Python/p3 So the total number of pages is probably in the tens of thousands.

So, server-side rendering it is. To accomplish that I set up a very simple Express server. It proxies some stuff over to the Django server (e.g. /rss.xml) and then lets NextJS handle the rest.

import next from "next";
import express from "express";

const app = next();
const handle = app.getRequestHandler();

app
  .prepare()
  .then(() => {
    const server = express();

    server.use(handle);

    server.listen(port, (err) => {
      if (err) throw err;
      console.log(`> Ready on http://localhost:${port}`);
    });
  })

Now, my site is behind a CDN. And technically, it's behind Nginx too where I do some proxy_pass in-memory caching as a second line of defense.
Requests come in like this:

1. from user to CDN
2. from CDN to Nginx
3. from Nginx to Express (proxy_pass)
4. from Express to next().getRequestHandler()

And I set Cache-Control in res.setHeader("Cache-Control", "public,max-age=86400") from within the getServerSideProps functions in the src/pages/**/*.tsx files. And once that's set, the response will be cached both in Nginx and in the CDN.

Any caching is tricky when you need to do revalidation. Especially when you roll out a new central feature in the core bundle. But I quite like this pattern of a slow-rolling upgrade as individual pages eventually expire throughout the day.

This is a nasty bug with this and I don't yet know how to solve it. Client-side navigation is dependent of hashing. So loading this page, when done with client-side navigation, becomes /_next/data/2ps5rE-K6E39AoF4G6G-0/en/plog.json (no, I don't know how that hashed URL is determined). But if a new deployment happens, the new URL becomes /_next/data/UhK9ANa6t5p5oFg3LZ5dy/en/plog.json so you end up with a 404 because you started on a page based on an old JavaScript bundle, that is now invalid.

Thankfully, NextJS handles it quite gracefully by throwing an error on the 404 so it proceeds with a regular link redirect which takes you away from the old page.

Client-side navigation still sucks. Kinda.

Next has a built-in <Link> component that you use like this:

import Link from "next/link";

...

<Link href={"/plog/" + post.oid}>
  {post.title}
</Link>

Now, clicking any of those links will automatically enable client-side routing. Thankfully, it takes care of preloading the necessary JavaScript (and CSS) simply by hovering over the link, so that when you eventually click it just needs to do an XHR request to get the JSON necessary to be able to render the page within the loaded app (and then do the pushState stuff to change the URL accordingly).

It sounds good in theory but it kinda sucks because unless you have a really good Internet connection (or could be you hit upon a CDN-cold URL), nothing happens when you click. This isn't NextJS's fault, but I wonder if it's actually horribly for users.

Yes, it sucks that a user clicks something but nothing happens. (I think it would be better if it was a button-press and not a link because buttons feel more like an app whereas links have deeply ingrained UX expectations). But most of the time, it's honestly very fast and when it works it's a nice experience. It's a great piece of functionality for more app'y sites, but less good for websites whose most of the traffic comes from direct links or Google searches.

NextJS has built-in critical CSS optimization

Critical inline CSS is critical (pun intended) for web performance. Especially on my poor site where I depend on a bloated (and now ancient) CSS framework called Semantic-UI. Without inline CSS, the minified CSS file would become over 200KB.

In NextJS, to enable inline critical CSS loading you just need to add this to your next.config.js:

    experimental: { optimizeCss: true },

and you have to add critters to your package.json. I've found some bugs with it but nothing I can't work around.

Conclusion and what's next

I'm very familiar and experienced with React but NextJS is new to me. I've managed to miss it all these years. Until now. So there's still a lot to learn. With other frameworks, I've always been comfortable that I don't actually understand how Webpack and Babel work (internally) but at least I understood when and how I was calling/depending on it. Now, with NextJS there's a lot of abstracted magic that I don't quite understand. It's hard to let go of that. It's hard to get powerful tools that are complex and created by large groups of people and understand it all too. If you're desperate to understand exactly how something works, you inevitably have to scale back the amount of stuff you're leveraging. (Note, it might be different if it's absolute core to what you do for work and hack on for 8 hours a day)

The JavaScript bundles in NextJS lazy-load quite decently but it's definitely more bloat than it needs to be. It's up to me to fix it, partially, because much of the JS code on my site is for things that technically can wait such as the interactive commenting form and the auto-complete search.

But here's the rub; my site is not an app. Most traffic comes from people doing a Google search, clicking on my page, and then bugger off. It's quite static that way and who am I to assume that they'll stay and click around and reuse all that loaded JavaScript code.

With that said; I'm going to start an experiment to rewrite the site again in Remix.

tl;dr; Periodically, the whole of MDN is built, by our Node code, in a GitHub Action. A Python script bulk-publishes this to Elasticsearch. Our Django server queries the same Elasticsearch via /api/v1/search. The site-search page is a static single-page app that sends XHR requests to the /api/v1/search endpoint. Search results' sort-order is determined by match and "popularity".

Jamstack'ing

The challenge with "Jamstack" websites is with data that is too vast and dynamic that it doesn't make sense to build statically. Search is one of those. For the record, as of Feb 2021, MDN consists of 11,619 documents (aka. articles) in English. Roughly another 40,000 translated documents. In English alone, there are 5.3 million words. So to build a good search experience we need to, as a static site build side-effect, index all of this in a full-text search database. And Elasticsearch is one such database and it's good. In particular, Elasticsearch is something MDN is already quite familiar with because it's what was used from within the Django app when MDN was a wiki.

Note: MDN gets about 20k site-searches per day from within the site.

Build

When we build the whole site, it's a script that basically loops over all the raw content, applies macros and fixes, dumps one index.html (via React server-side rendering) and one index.json. The index.json contains all the fully rendered text (as HTML!) in blocks of "prose". It looks something like this:

{
  "doc": {
    "title": "DOCUMENT TITLE",
    "summary": "DOCUMENT SUMMARY",
    "body": [
      {
        "type": "prose", 
        "value": {
          "id": "introduction", 
          "title": "INTRODUCTION",
          "content": "<p>FIRST BLOCK OF TEXTS</p>"
       }
     },
     ...
   ],
   "popularity": 0.12345,
   ...
}

You can see one here: /en-US/docs/Web/index.json

Indexing

Next, after all the index.json files have been produced, a Python script takes over and it traverses all the index.json files and based on that structure it figures out the, title, summary, and the whole body (as HTML).

Next up, before sending this into the bulk-publisher in Elasticsearch it strips the HTML. It's a bit more than just turning <p>Some <em>cool</em> text.</p> to Some cool text. because it also cleans up things like <div class="hidden"> and certain <div class="notecard warning"> blocks.

One thing worth noting is that this whole thing runs roughly every 24 hours and then it builds everything. But what if, between two runs, a certain page has been removed (or moved), how do you remove what was previously added to Elasticsearch? The solution is simple: it deletes and re-creates the index from scratch every day. The whole bulk-publish takes a while so right after the index has been deleted, the searches won't be that great. Someone could be unlucky in that they're searching MDN a couple of seconds after the index was deleted and now waiting for it to build up again.
It's an unfortunate reality but it's a risk worth taking for the sake of simplicity. Also, most people are searching for things in English and specifically the Web/ tree so the bulk-publishing is done in a way the most popular content is bulk-published first and the rest was done after. Here's what the build output logs:

Found 50,461 (potential) documents to index
Deleting any possible existing index and creating a new one called mdn_docs
Took 3m 35s to index 50,362 documents. Approximately 234.1 docs/second
Counts per priority prefixes:
    en-us/docs/web                 9,056
    *rest*                         41,306

So, yes, for 3m 35s there's stuff missing from the index and some unlucky few will get fewer search results than they should. But we can optimize this in the future.

Searching

The way you connect to Elasticsearch is simply by a URL it looks something like this:

https://USER:PASSWD@HASH.us-west-2.aws.found.io:9243

It's an Elasticsearch cluster managed by Elastic running inside AWS. Our job is to make sure that we put the exact same URL in our GitHub Action ("the writer") as we put it into our Django server ("the reader").
In fact, we have 3 Elastic clusters: Prod, Stage, Dev.
And we have 2 Django servers: Prod, Stage.
So we just need to carefully make sure the secrets are set correctly to match the right environment.

Now, in the Django server, we just need to convert a request like GET /api/v1/search?q=foo&locale=fr (for example) to a query to send to Elasticsearch. We have a simple Django view function that validates the query string parameters, does some rate-limiting, creates a query (using elasticsearch-dsl) and packages the Elasticsearch results back to JSON.

How we make that query is important. In here lies the most important feature of the search; how it sorts results.

In one simple explanation, the sort order is a combination of popularity and "matchness". The assumption is that most people want the popular content. I.e. they search for foreach and mean to go to /en-US/docs/Web/JavaScript/Reference/Global_Objects/Array/forEach not /en-US/docs/Web/API/NodeList/forEach both of which contains forEach in the title. The "popularity" is based on Google Analytics pageviews which we download periodically, normalize into a floating-point number between 1 and 0. At the of writing the scoring function does something like this:

rank = doc.popularity * 10 + search.score

This seems to produce pretty reasonable results.

But there's more to the "matchness" too. Elasticsearch has its own API for defining boosting and the way we apply is:

• match phrase in the title: Boost = 10.0
• match phrase in the body: Boost = 5.0
• match in title: Boost = 2.0
• match in body: Boost = 1.0

This is then applied on top of whatever else Elasticsearch does such as "Term Frequency" and "Inverse Document Frequency" (tf and if). This article is a helpful introduction.

We're most likely not done with this. There's probably a lot more we can do to tune this myriad of knobs and sliders to get the best possible ranking of documents that match.

Web UI

The last piece of the puzzle is how we display all of this to the user. The way it works is that developer.mozilla.org/$locale/search returns a static page that is blank. As soon as the page has loaded, it lazy-loads JavaScript that can actually issue the XHR request to get and display search results. The code looks something like this:

function SearchResults() {
  const [searchParams] = useSearchParams();
  const sp = createSearchParams(searchParams);
  // add defaults and stuff here
  const fetchURL = `/api/v1/search?${sp.toString()}`;

  const { data, error } = useSWR(
    fetchURL,
    async (url) => {
      const response = await fetch(URL);
      // various checks on the response.statusCode here
      return await response.json();
    }
  );

  // render 'data' or 'error' accordingly here

A lot of interesting details are omitted from this code snippet. You have to check it out for yourself to get a more up-to-date insight into how it actually works. But basically, the window.location (and pushState) query string drives the fetch() call and then all the component has to do is display the search results with some highlighting.

The /api/v1/search endpoint also runs a suggestion query as part of the main search query. This extracts out interest alternative search queries. These are filtered and scored and we issue "sub-queries" just to get a count for each. Now we can do one of those "Did you mean...". For example: search for intersections.

In conclusion

There are a lot of interesting, important, and careful details that are glossed over here in this blog post. It's a constantly evolving system and we're constantly trying to improve and perfect the system in a way that it fits what users expect.

A lot of people reach MDN via a Google search (e.g. mdn array foreach) but despite that, nearly 5% of all traffic on MDN is the site-search functionality. The /$locale/search?... endpoint is the most frequently viewed page of all of MDN. And having a good search engine that's reliable is nevertheless important. By owning and controlling the whole pipeline allows us to do specific things that are unique to MDN that other websites don't need. For example, we index a lot of raw HTML (e.g. <video>) and we have code snippets that needs to be searchable.

Hopefully, the MDN site-search will elevate from being known to be very limited to something now that can genuinely help people get to the exact page better than Google can. Yes, it's worth aiming high!

tl;dr; <img src="/avatar.random.png" alt="Random avataaar"> generates this image:

(try reloading to get a random new one. funny aren't they?)

When you use Gravatar you can convert people's email addresses to their mugshot.
It works like this:

<img src="https://www.gravatar.com/avatar/$(md5(user.email))">

But most people don't have their mugshot on Gravatar.com unfortunately. But you still want to display an avatar that is distinct per user. Your best option is to generate one and just use the user's name or email as a seed (so it's always random but always deterministic for the same user). And you can also supply a fallback image to Gravatar that they use if the email doesn't match any email they have. That's where this blog post comes in.

I needed that so I shopped around and found avataaars generator which is available as a React component. But I need it to be server-side and in Python. And thankfully there's a great port called: py-avataaars.

It depends on CairoSVG to convert an SVG to a PNG but it's easy to install. Anyway, here's my hack to generate random "avataaars" from Django:

import io
import random

import py_avataaars
from django import http
from django.utils.cache import add_never_cache_headers, patch_cache_control


def avatar_image(request, seed=None):
    if not seed:
        seed = request.GET.get("seed") or "random"

    if seed != "random":
        random.seed(seed)

    bytes = io.BytesIO()

    def r(enum_):
        return random.choice(list(enum_))

    avatar = py_avataaars.PyAvataaar(
        style=py_avataaars.AvatarStyle.CIRCLE,
        # style=py_avataaars.AvatarStyle.TRANSPARENT,
        skin_color=r(py_avataaars.SkinColor),
        hair_color=r(py_avataaars.HairColor),
        facial_hair_type=r(py_avataaars.FacialHairType),
        facial_hair_color=r(py_avataaars.FacialHairColor),
        top_type=r(py_avataaars.TopType),
        hat_color=r(py_avataaars.ClotheColor),
        mouth_type=r(py_avataaars.MouthType),
        eye_type=r(py_avataaars.EyesType),
        eyebrow_type=r(py_avataaars.EyebrowType),
        nose_type=r(py_avataaars.NoseType),
        accessories_type=r(py_avataaars.AccessoriesType),
        clothe_type=r(py_avataaars.ClotheType),
        clothe_color=r(py_avataaars.ClotheColor),
        clothe_graphic_type=r(py_avataaars.ClotheGraphicType),
    )
    avatar.render_png_file(bytes)

    response = http.HttpResponse(bytes.getvalue())
    response["content-type"] = "image/png"
    if seed == "random":
        add_never_cache_headers(response)
    else:
        patch_cache_control(response, max_age=60, public=True)

    return response

It's not perfect but it works. The URL to this endpoint is /avatar.<seed>.png and if you make the seed parameter random the response is always different.

To make the image not random, you replace the <seed> with any string. For example (use your imagination):

{% for comment in comments %}
  <img src="/avatar.{{ comment.user.id }}.png" alt="{{ comment.user.name }}">
  <blockquote>{{ comment.text }}</blockquote>
  <i>{{ comment.date }}</i>
{% endfor %}

Django's Form framework is excellent. It's intuitive and versatile and, best of all, easy to use. However, one little thing that is not so intuitive is how do you render a bound form with default/initial values when the form is never rendered unbound.

If you do this in Django:

class MyForm(forms.Form):
    name = forms.CharField(required=False)

def view(request):
    form = MyForm(initial={'name': 'Peter'})
    return render(request, 'page.html', form=form)

# Imagine, in 'page.html' that it does this:
#  <label>Name:</label>
#  {{ form.name }}

...it will render out this:

<label>Name:</label>
<input type="text" name="name" value="Peter">

The whole initial trick is something you can set on the whole form or individual fields. But it's only used in UN-bound forms when rendered.

If you change your view function to this:

def view(request):
    form = MyForm(request.GET, initial={'name': 'Peter'}) # data passed!
    if form.is_valid():  # makes it bound!
        print(form.cleaned_data['name'])
    return render(request, 'page.html', form=form)

Now, the form is bound and the initial stuff is essentially ignored.
Because name is not present in request.GET. And if it was present, but an empty string, it wouldn't be able to benefit for the default value.

My solution

I tried many suggestions and tricks (based on rapid Stackoverflow searching) and nothing worked.

I knew one thing: Only the view should know the actual initial values.

Here's what works:

import copy


class MyForm(forms.Form):
    name = forms.CharField(required=False)

    def __init__(self, data, **kwargs):
        initial = kwargs.get('initial', {})
        data = {**initial, **data}
        super().__init__(data, **kwargs)

Now, suppose you don't have ?name=something in request.GET the line print(form.cleaned_data['name']) will print Peter and the rendered form will look like this:

<label>Name:</label>
<input type="text" name="name" value="Peter">

And, as expected, if you have ?name=Ashley in request.GET it will print Ashley and produce this rendered HTML too:

<label>Name:</label>
<input type="text" name="name" value="Ashley">

UPDATE June 2020

If data is a QueryDict object (e.g. <QueryDict: {'days': ['90']}>), and initial is a plain dict (e.g. {'days': 30}),
then you can merge these with {**data, **initial} because it produces a plain dict of value {'days': [90]} which Django's form stuff doesn't know is supposed to be "flattened".

The solution is to use:

from django.utils.datastructures import MultiValueDict

...

    def __init__(self, data, **kwargs):
        initial = kwargs.get("initial", {})
        data = MultiValueDict({**{k: [v] for k, v in initial.items()}, **data})
        super().__init__(data, **kwargs)

(To be honest; this might work in the app I'm currently working on but I don't feel confident that this is covering all cases)

Heroku is great but it's sometimes painful when your app isn't just in one single language. What I have is a project where the backend is Python (Django) and the frontend is JavaScript (Preact). The folder structure looks like this:

/
  - README.md
  - manage.py
  - requirements.txt
  - my_django_app/
     - settings.py
     - asgi.py
     - api/
        - urls.py
        - views.py
  - frontend/
     - package.json
     - yarn.lock
     - preact.config.js
     - build/
        ...
     - src/
        ...

A bunch of things omitted for brevity but people familiar with Django and preact-cli/create-create-app should be familiar.
The point is that the root is a Python app and the front-end is exclusively inside a sub folder.

When you do local development, you start two servers:

• ./manage.py runserver - starts http://localhost:8000
• cd frontend && yarn start - starts http://localhost:3000

The latter is what you open in your browser. That preact app will do things like:

const response = await fetch('/api/search');

and, in preact.config.js I have this:

export default (config, env, helpers) => {

  if (config.devServer) {
    config.devServer.proxy = [
      {
        path: "/api/**",
        target: "http://localhost:8000"
      }
    ];
  }

};

...which is hopefully self-explanatory. So, calls like GET http://localhost:3000/api/search actually goes to http://localhost:8000/api/search.

That's when doing development. The interesting thing is going into production.

Before we get into Heroku, let's first "merge" the two systems into one and the trick used is Whitenoise. Basically, Django's web server will be responsibly not only for things like /api/search but also static assets such as / --> frontend/build/index.html and /bundle.17ae4.js --> frontend/build/bundle.17ae4.js.

This is basically all you need in settings.py to make that happen:

MIDDLEWARE = [
    "django.middleware.security.SecurityMiddleware",
    "whitenoise.middleware.WhiteNoiseMiddleware",
    ...
]

WHITENOISE_INDEX_FILE = True

STATIC_URL = "/"
STATIC_ROOT = BASE_DIR / "frontend" / "build"

However, this isn't quite enough because the preact app uses preact-router which uses pushState() and other code-splitting magic so you might have a URL, that users see, like this: https://myapp.example.com/that/thing/special and there's nothing about that in any of the Django urls.py files. Nor is there any file called frontend/build/that/thing/special/index.html or something like that.
So for URLs like that, we have to take a gamble on the Django side and basically hope that the preact-router config knows how to deal with it. So, to make that happen with Whitenoise we need to write a custom middleware that looks like this:

from whitenoise.middleware import WhiteNoiseMiddleware


class CustomWhiteNoiseMiddleware(WhiteNoiseMiddleware):
    def process_request(self, request):
        if self.autorefresh:
            static_file = self.find_file(request.path_info)
        else:
            static_file = self.files.get(request.path_info)

            # These two lines is the magic.
            # Basically, the URL didn't lead to a file (e.g. `/manifest.json`)
            # it's either a API path or it's a custom browser path that only
            # makes sense within preact-router. If that's the case, we just don't
            # know but we'll give the client-side preact-router code the benefit
            # of the doubt and let it through.
            if not static_file and not request.path_info.startswith("/api"):
                static_file = self.files.get("/")

        if static_file is not None:
            return self.serve(static_file, request)

And in settings.py this change:

MIDDLEWARE = [
    "django.middleware.security.SecurityMiddleware",
-   "whitenoise.middleware.WhiteNoiseMiddleware",
+   "my_django_app.middleware.CustomWhiteNoiseMiddleware",
    ...
]

Now, all traffic goes through Django. Regular Django view functions, static assets, and everything else fall back to frontend/build/index.html.

Heroku

Heroku tries to make everything so simple for you. You basically, create the app (via the cli or the Heroku web app) and when you're ready you just do git push heroku master. However that won't be enough because there's more to this than Python.

Unfortunately, I didn't take notes of my hair-pulling excruciating journey of trying to add buildpacks and hacks and Procfiles and custom buildpacks. Nothing seemed to work. Perhaps the answer was somewhere in this issue: "Support running an app from a subdirectory" but I just couldn't figure it out. I still find buildpacks confusing when it's beyond Hello World. Also, I didn't want to run Node as a service, I just wanted it as part of the "build process".

Docker to the rescue

Finally I get a chance to try "Deploying with Docker" in Heroku which is a relatively new feature. And the only thing that scared me was that now I need to write a heroku.yml file which was confusing because all I had was a Dockerfile. We'll get back to that in a minute!

So here's how I made a Dockerfile that mixes Python and Node:

FROM node:12 as frontend

COPY . /app
WORKDIR /app
RUN cd frontend && yarn install && yarn build


FROM python:3.8-slim

WORKDIR /app

RUN groupadd --gid 10001 app && useradd -g app --uid 10001 --shell /usr/sbin/nologin app
RUN chown app:app /tmp

RUN apt-get update && \
    apt-get upgrade -y && \
    apt-get install -y --no-install-recommends \
    gcc apt-transport-https python-dev

# Gotta try moving this to poetry instead!
COPY ./requirements.txt /app/requirements.txt
RUN pip install --upgrade --no-cache-dir -r requirements.txt

COPY . /app
COPY --from=frontend /app/frontend/build /app/frontend/build

USER app

ENV PORT=8000
EXPOSE $PORT

CMD uvicorn gitbusy.asgi:application --host 0.0.0.0 --port $PORT

If you're not familiar with it, the critical trick is on the first line where it builds some Node with as frontend. That gives me a thing I can then copy from into the Python image with COPY --from=frontend /app/frontend/build /app/frontend/build.

Now, at the very end, it starts a uvicorn server with all the static .js, index.html, and favicon.ico etc. available to uvicorn which ultimately runs whitenoise.

To run and build:

docker build . -t my_app
docker run -t -i --rm --env-file .env -p 8000:8000 my_app

Now, opening http://localhost:8000/ is a production grade app that mixes Python (runtime) and JavaScript (static).

Heroku + Docker

Heroku says to create a heroku.yml file and that makes sense but what didn't make sense is why I would add cmd line in there when it's already in the Dockerfile. The solution is simple: omit it. Here's what my final heroku.yml file looks like:

build:
  docker:
    web: Dockerfile

Check in the heroku.yml file and git push heroku master and voila, it works!

To see a complete demo of all of this check out https://github.com/peterbe/gitbusy and https://gitbusy.herokuapp.com/

Last week, I blogged about "How much faster is Redis at storing a blob of JSON compared to PostgreSQL?". Judging from a lot of comments, people misinterpreted this. (By the way, Redis is persistent). It's no surprise that Redis is faster.

However, it's a fact that I have do have a lot of blobs stored and need to present them via the web API as fast as possible. It's rare that I want to do relational or batch operations on the data. But Redis isn't a slam dunk for simple retrieval because I don't know if I trust its integrity with the 3GB worth of data that I both don't want to lose and don't want to load all into RAM.

But is it entirely wrong to look at WHICH database to get the best speed?

Reviewing this corner of Song Search helped me rethink this. PostgreSQL is, in my view, a better database for storing stuff. Redis is faster for individual lookups. But you know what's even faster? Nginx

Nginx??

The way the application works is that a React web app is requesting the Amazon product data for the sake of presenting an appropriate affiliate link. This is done by the browser essentially doing:

const response = await fetch('https://songsear.ch/api/song/5246889/amazon');

Internally, in the app, what it does is that it looks this up, by ID, on the AmazonAffiliateLookup ORM model. Suppose it wasn't there in the PostgreSQL, it uses the Amazon Affiliate Product Details API, to look it up and when the results come in it stores a copy of this in PostgreSQL so we can re-use this URL without hitting rate limits on the Product Details API. Lastly, in a piece of Django view code, it carefully scrubs and repackages this result so that only the fields used by the React rendering code is shipped between the server and the browser. That "scrubbed" piece of data is actually much smaller. Partly because it limits the results to the first/best match and it deletes a bunch of things that are never needed such as ProductTypeName, Studio, TrackSequence etc. The proportion is roughly 23x. I.e. of the 3GB of JSON blobs stored in PostgreSQL only 130MB is ever transported from the server to the users.

Again, Nginx?

Nginx has a built in reverse HTTP proxy cache which is easy to set up but a bit hard to do purges on. The biggest flaw, in my view, is that it's hard to get a handle of how much RAM this it's eating up. Well, if the total possible amount of data within the server is 130MB, then that is something I'm perfectly comfortable to let Nginx handle cache in RAM.

Good HTTP performance benchmarking is hard to do but here's a teaser from my local laptop version of Nginx:

▶ hey -n 10000 -c 10 https://songsearch.local/api/song/1810960/affiliate/amazon-itunes

Summary:
  Total:    0.9882 secs
  Slowest:  0.0279 secs
  Fastest:  0.0001 secs
  Average:  0.0010 secs
  Requests/sec: 10119.8265


Response time histogram:
  0.000 [1] |
  0.003 [9752]  |■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.006 [108]   |
  0.008 [70]    |
  0.011 [32]    |
  0.014 [8] |
  0.017 [12]    |
  0.020 [11]    |
  0.022 [1] |
  0.025 [4] |
  0.028 [1] |


Latency distribution:
  10% in 0.0003 secs
  25% in 0.0006 secs
  50% in 0.0008 secs
  75% in 0.0010 secs
  90% in 0.0013 secs
  95% in 0.0016 secs
  99% in 0.0068 secs

Details (average, fastest, slowest):
  DNS+dialup:   0.0000 secs, 0.0001 secs, 0.0279 secs
  DNS-lookup:   0.0000 secs, 0.0000 secs, 0.0026 secs
  req write:    0.0000 secs, 0.0000 secs, 0.0011 secs
  resp wait:    0.0008 secs, 0.0001 secs, 0.0206 secs
  resp read:    0.0001 secs, 0.0000 secs, 0.0013 secs

Status code distribution:
  [200] 10000 responses

10,000 requests across 10 clients at rougly 10,000 requests per second. That includes doing all the HTTP parsing, WSGI stuff, forming of a SQL or Redis query, the deserialization, the Django JSON HTTP response serialization etc. The cache TTL is controlled by simply setting a Cache-Control HTTP header with something like max-age=86400.

Now, repeated fetches for this are cached at the Nginx level and it means it doesn't even matter how slow/fast the database is. As long as it's not taking seconds, with a long Cache-Control, Nginx can hold on to this in RAM for days or until the whole server is restarted (which is rare).

Conclusion

If you the total amount of data that can and will be cached is controlled, putting it in a HTTP reverse proxy cache is probably order of magnitude faster than messing with chosing which database to use.

By analyzing my Nginx logs, I've concluded that SongSearch's autocomplete JSON API now gets about 2.2 requests per second. I.e. these are XHR requests to /api/search/autocomplete?q=....

Roughly, 1.8 requests per second goes back to the Django/Elasticsearch backend. That's a hit ratio of 16%. These Django/Elasticsearch requests take roughly 200ms on average. I suspect about 150-180ms of that time is spent querying Elasticsearch, the rest being Python request/response and JSON "paperwork".

Caching strategy

Caching is hard because the queries are so vastly different over time. Had I put a Redis cache decorator on the autocomplete Django view function I'd quickly bloat Redis memory and cause lots of evictions.

What I used to do was something like this:

def search_autocomplete(request):
   q = request.GET.get('q') 

   cache_key = None
   if len(q) < 10:
      cache_key = 'autocomplete:' + q
      results = cache.get(cache_key)
      if results is not None:
          return http.JsonResponse(results)

   results = _do_elastisearch_query(q)
   if cache_key:
       cache.set(cache_key, results, 60 * 60)

   return http.JsonResponse(results)   

However, after some simple benchmarking it was clear that using Nginx' uwsgi_cache it was much faster to let the cacheable queries terminate already at Nginx. So I changed the code to something like this:

def search_autocomplete(request):
   q = request.GET.get('q') 
   results = _do_elastisearch_query(q)
   response = http.JsonResponse(results)   

   if len(q) < 10:
       patch_cache_control(response, public=True, max_age=60 * 60)

   return response

The only annoying thing about Nginx caching is that purging is hard unless you go for that Nginx Plus (or whatever their enterprise version is called). But more annoying, to me, is that fact that I can't really see what this means for my server. When I was caching with Redis I could just use redis-cli and...

> INFO
...
# Memory
used_memory:123904288
used_memory_human:118.16M
...

Nginx Amplify

My current best tool for keeping an eye on Nginx is Nginx Amplify. It gives me some basic insights about the state of things. Here are some recent screenshots:

Thoughts and conclusion

Caching is hard. But it's also fun because it ties directly into performance work.

In my business logic, I chose that autocomplete queries that are between 1 and 9 characters are cacheable. And I picked a TTL of 60 minutes. At this point, I'm not sure exactly why I chose that logic but I remember doing some back-of-envelope calculations about what the hit ratio would be and roughly what that would mean in bytes in RAM. I definitely remember picking 60 minutes because I was nervous about bloating Nginx's memory usage. But as of today, I'm switching that up to 24 hours and let's see what that does to my current 16% Nginx cache hit ratio. At the moment, /var/cache/nginx-cache/ is only 34MB which isn't much.

Another crux with using uwsgi_cache (or proxy_cache) is that you can't control the cache key very well. When it was all in Python I was able to decide about the cache key myself. A plausible implementation is cache_key = q.lower().strip() for example. That means you can protect your Elasticsearch backend from having to do {"q": "A"} and {"q": "a"}. Who knows, perhaps there is a way to hack this in Nginx without compiling in some Lua engine.

The ideal would be some user-friendly diagnostics tool that I can point somewhere, towards Nginx, that says how much my uwsgi_cache is hurting or saving me. Autocomplete is just one of many things going on on this single DigitalOcean server. There's also a big PostgreSQL server, a node-express cluster, a bunch of uwsgi workers, Redis, lots of cron job scripts, and of course a big honking Elasticsearch 6.

UPDATE (July 12 2019)

Currently, and as mentioned above, I only set Cache-Control headers (which means Nginx snaps it up) for queries that at max 9 characters long. I wanted to appreciate and understand how ratio of all queries are longer than 9 characters so I wrote a report and its output is this:

POINT: 7
Sum show 75646 32.2%
Sum rest 159321 67.8%

POINT: 8
Sum show 83702 35.6%
Sum rest 151265 64.4%

POINT: 9
Sum show 90870 38.7%
Sum rest 144097 61.3%

POINT: 10
Sum show 98384 41.9%
Sum rest 136583 58.1%

POINT: 11
Sum show 106093 45.2%
Sum rest 128874 54.8%

POINT: 12
Sum show 113905 48.5%
Sum rest 121062 51.5%

It means that (independent of time expiry) 38.7% of queries are 9 characters or less.

Suppose that you have so many thousands of pages that you can't just create a single /sitemap.xml file that has all the URLs (aka <loc>) listed. Then you need to make a /sitemaps.xml that points to the other sitemap files. And if you're in the thousands, you'll need to gzip these files.

The blog post demonstrates how Song Search generates a sitemap file that points to 63 sitemap-{M}-{N}.xml.gz files which spans about 1,000,000 URLs. The context here is Python and the getting of the data is from Django. Python is pretty key here but if you have something other than Django, you can squint and mentally replace that with your own data mapper.

Generate the sitemap .xml.gz file(s)

Here's the core of the work. A generator function that takes a Django QuerySet instance (that is ordered and filtered!) and then starts generating etree trees and dumps them to disk with gzip.

import gzip

from lxml import etree


outfile = "sitemap-{start}-{end}.xml"
batchsize = 40_000


def generate(self, qs, base_url, outfile, batchsize):
    # Use `.values` to make the query much faster
    qs = qs.values("name", "id", "artist_id", "language")

    def start():
        return etree.Element(
            "urlset", xmlns="http://www.sitemaps.org/schemas/sitemap/0.9"
        )

    def close(root, filename):
        with gzip.open(filename, "wb") as f:
            f.write(b'<?xml version="1.0" encoding="utf-8"?>\n')
            f.write(etree.tostring(root, pretty_print=True))

    root = filename = None

    count = 0
    for song in qs.iterator():
        if not count % batchsize:
            if filename:  # not the very first loop
                close(root, filename)
                yield filename
            filename = outfile.format(start=count, end=count + batchsize)
            root = start()
        loc = "{}{}".format(base_url, make_song_url(song))
        etree.SubElement(etree.SubElement(root, "url"), "loc").text = loc
        count += 1
    close(root, filename)
    yield filename

The most important lines in terms of lxml.etree and sitemaps are:

root = etree.Element("urlset", xmlns="http://www.sitemaps.org/schemas/sitemap/0.9")
...         
etree.SubElement(etree.SubElement(root, "url"), "loc").text = loc

Another important thing is the note about using .values(). If you don't do that Django will create a model instance for every single row it returns of the iterator. That's expensive. See this blog post.

Another important thing is to use a Django ORM iterator as that's much more efficient than messing around with limits and offsets.

Generate the map of sitemaps

Making the map of maps doesn't need to be gzipped since it's going to be tiny.

def generate_map_of_maps(base_url, outfile):
    root = etree.Element(
        "sitemapindex", xmlns="http://www.sitemaps.org/schemas/sitemap/0.9"
    )

    with open(outfile, "wb") as f:
        f.write(b'<?xml version="1.0" encoding="UTF-8"?>\n')
        files_created = sorted(glob("sitemap-*.xml.gz"))
        for file_created in files_created:
            sitemap = etree.SubElement(root, "sitemap")
            uri = "{}/{}".format(base_url, os.path.basename(file_created))
            etree.SubElement(sitemap, "loc").text = uri
            lastmod = datetime.datetime.fromtimestamp(
                os.stat(file_created).st_mtime
            ).strftime("%Y-%m-%d")
            etree.SubElement(sitemap, "lastmod").text = lastmod
        f.write(etree.tostring(root, pretty_print=True))

And that sums it up. On my laptop, it takes about 60 seconds to generate 39 of these files (e.g. sitemap-1560000-1600000.xml.gz) and that's good enough.

Bonus and Thoughts

The bad news is that this is about as good as it gets in terms of performance. The good news is that there are no low-hanging fruit fixes. I know, because I tried. I experimented with not using pretty_print=True and I experimented with not writing with gzip.open and instead gzipping the files on later. Nothing made any significant difference. The lxml.etree part of this, in terms of performance, is order of maginitude marginal in comparison to the cost of actually getting the data out of the database plus later writing to disk. I also experimenting with generating the gzip content with zopfli and it didn't make much of a difference.

I originally wrote this code years ago and when I did, I think I knew more about sitemaps. In my implementation I use a batch size of 40,000 so each file is called something like sitemap-40000-80000.xml.gz and weighs about 800KB. Not sure why I chose 40,000 but perhaps not important.