Peterbe.com

tl;dr; minimalcss is much better than mincss to slew out the minimal CSS your page needs to render. More accurate and more powerful features. This site now uses minimalcss in inline the minimum CSS needed to render the page.

I started minimalcss back in August 2017 and its goal was ultimately to replace mincss.

The major difference between minimalcss and mincss isn't that one is Node and one is Python, but that minimalcss is based on a full headless browser to handle all the CSS downloading and the proper rendering of the DOM. The other major difference is that mincss was based in regular expressions to analyze the CSS and minimalcss is based on proper abstract syntax tree ("AST") implemented by csso.

Because minimalcss is AST based, it can do a lot more. Smarter. For example, it's able to analyze the CSS to correctly and confidently figure out if any/which keyframe animations and font-face at-rules are actually needed.
Also, because minimalcss is based on csso, when it minifies the CSS it's able to restructure the CSS in a safe and smart way. I.e. p { color: blue; } h2 { color: blue; } becomes p,h2{color:blue}.

So, now I use minimalcss here on this blog. The pages are rendered in Django and a piece of middleware sniffs all outgoing HTML responses and depending on the right conditions it dumps the HTML as a file on disk as path/in/url/index.html. Then, that newly created file is sent to a background worker in Celery which starts post-processing it. Every index.html file is accompanied with the full absolute URL that it belongs to and that's the URL that gets sent to minimalcss which returns the absolute minimal CSS the page needs to load and lastly, a piece of Python script basically does something like this:

From...

<!-- before -->
<link rel="stylesheet" href="/file.css"/>

To...

<!-- after -->
<noscript><link rel="stylesheet" href="/file.css"/></noscript>
<style> ... /* minimal CSS selectors for rendering from /file.css */ ... </style>

There is also a new JavaScript dependency which is the cssrelpreload.js from the loadCSS project. So all the full (original) CSS is still downloaded and inserted into the CSSOM but it happens much later which ultimately means the page can be rendered and useful much sooner than if we'd have to wait to download and parse all of the .css URLs.

I can go into more details if there's interest and others want to do this too. Because this site is all Python and minimalcss is all Node, the integration is done over HTTP on localhost with minimalcss-server.

The results

Unfortunately, this change was mixed in with other smaller optimizations that makes the comparison unfair. (Hey! my personal blog is just a side-project after all). But I downloaded a file before and after the upgrade and compared:

▶ ls -lh *.html
-rw-r--r--  1 peterbe  wheel    19K Mar  7 13:22 after.html
-rw-r--r--  1 peterbe  wheel    96K Mar  7 13:21 before.html

If I extract out the inline style block from both pages and compare it looks like this:
https://gist.github.com/peterbe/fc2fdddd5721fb35a99dc1a50c2b5311

So, downloading the initial HTML document is now 19KB instead of previous 96KB. And visually there's absolutely no difference.

Granted, in the after.html version, a piece of JavaScript kicks in and downloads /static/css/base.min.91f6fc577a60.css and /static/css/base-dynamic.min.e335b9bfa0b1.css from the CDN. So you have to download these too:

▶ ls -lh *.css.gz
-rw-r--r--  1 peterbe  wheel   5.0K Mar  7 10:24 base-dynamic.min.e335b9bfa0b1.css.gz
-rw-r--r--  1 peterbe  wheel    95K Mar  7 10:24 base.min.91f6fc577a60.css.gz

The reason the difference appears to be huge is because I changed a couple of other things around the same time. Sorry. For example, certain DOM nodes were rendered as HTML but made hidden until some jQuery script made it not hidden anymore. For example, the "dimmer" effect over a comment textarea after you hit the submit button. Now, I've changed the jQuery code to build up the DOM when it needs it rather than relying on it being there (hidden). This means that certain base64 embedded font-faces are no longer needed in the minimal CSS payload.

Why this approach is better

So the old approach was to run mincss on the HTML and inject that as an inline style block and throw away the original (relevant) <link rel="stylesheet" href="..."> tags.
That had the annoying drawback that there was CSS in the stylesheets that I knew was going to be needed by some XHR or JavaScript later. For example, if you post a comment some jQuery code changes the DOM and that new DOM needs these CSS selectors later. So I had to do things like this:

.project a.perm { /* no mincss */
    font-size: 0.7em;
    padding-left: 8px;
}
.project a.perm:link { /* no mincss */
    color: rgb(151,151,151);
}
.project a.perm:hover { /* no mincss */
    color: rgb(51,51,51);
}

This was to inform mincss to leave those untouched even though no DOM node uses them right now. With minimalcss this is no longer needed.

What's next?

Keep working on minimalcss and make it even better.

Also, the scripting I used to modify the HTML file is a hack and should probably be put into the minimalcss project.

Last but not least, every time I put in some effort to web performance optimize my blog pages my Google ranking goes up and I usually see an increase in Google referrals in my Google Analytics because it's pretty obvious that Google loves fast sites. So I'm optimistically waiting for that effect.

This is a quick-and-dirty how-to on how to use csso to handle the minification/compression of CSS in django-pipeline.

First create a file called compressors.py somewhere in your project. Make it something like this:

import subprocess
from pipeline.compressors import CompressorBase
from django.conf import settings


class CSSOCompressor(CompressorBase):

    def compress_css(self, css):
        proc = subprocess.Popen(
            [
                settings.PIPELINE['CSSO_BINARY'], 
                '--restructure-off'
            ],
            stdin=subprocess.PIPE,
            stdout=subprocess.PIPE,
        )
        css_out = proc.communicate(
            input=css.encode('utf-8')
        )[0].decode('utf-8')
        # was_size = len(css)
        # new_size = len(css_out)
        # print('FROM {} to {} Saved {}  ({!r})'.format(
        #     was_size,
        #     new_size,
        #     was_size - new_size,
        #     css_out[:50]
        # ))
        return css_out

In your settings.py where you configure django-pipeline make it something like this:

PIPELINE = {
    'STYLESHEETS': PIPELINE_CSS,
    'JAVASCRIPT': PIPELINE_JS,

    # These two important lines. 
    'CSSO_BINARY': path('node_modules/.bin/csso'),
    # Adjust the dotted path name to where you put your compressors.py
    'CSS_COMPRESSOR': 'peterbecom.compressors.CSSOCompressor',

    'JS_COMPRESSOR': ...

Next, install csso-cli in your project root (where you have the package.json). It's a bit confusing. The main package is called csso but to have a command line app you need to install csso-cli and when that's been installed you'll have a command line app called csso.

$ yarn add csso-cli

or

$ npm i --save csso-cli

Check that it installed:

$ ./node_modules/.bin/csso --version
3.5.0

And that's it!

--restructure-off

So csso has an advanced feature to restructure the CSS and not just remove whitespace and not needed semicolons. It costs a bit of time to do that so if you want to squeeze the extra milliseconds out, enable it. Trading time for space.
See this benchmark for a comparison with and without --restructure-off in csso.

Why csso you might ask

Check out the latest result from css-minification-benchmark. It's not super easy to read by it seems the best performing one in terms of space (bytes) is crass written by my friend and former colleague @mattbasta. However, by far the fastest is csso when using --restructre-off. Minifiying font-awesome.css with crass takes 326.52 ms versus 3.84 ms in csso.

But what's great about csso is Roman @lahmatiy Dvornov. I call him a friend too for all the help and work he's done on minimalcss (not a CSS minification tool by the way). Roman really understands CSS and csso is actively maintained by him and other smart people who actually get into the scary weeds of CSS browser hacks. That gives me more confidence to recommend csso. Also, squeezing a couple bytes extra out of your .min.css files isn't important when gzip comes into play. It's better that the minification tool is solid and stable.

Check out Roman's slides which, even if you don't read it all, goes to show that CSS minification is so much more than just regex replacing whitespace.
Also crass admits as one of its disadvantages: "Certain "CSS hacks" that use invalid syntax are unsupported".

So the context is this; a zip file is uploaded into a web service and Python then needs extract that and analyze and deal with each file within. In this particular application what it does is that it looks at the file's individual name and size, compares that to what has already been uploaded in AWS S3 and if the file is believed to be different or new, it gets uploaded to AWS S3.

The challenge is that these zip files that come in are huuuge. The average is 560MB but some are as much as 1GB. Within them, there are mostly plain text files but there are some binary files in there too that are huge. It's not unusual that each zip file contains 100 files and 1-3 of those make up 95% of the zip file size.

At first I tried unzipping the file, in memory, and deal with one file at a time. That failed spectacularly with various memory explosions and EC2 running out of memory. I guess it makes sense. First you have the 1GB file in RAM, then you unzip each file and now you have possibly 2-3GB all in memory. So, the solution, after much testing, was to dump the zip file to disk (in a temporary directory in /tmp) and then iterate over the files. This worked much better but I still noticed the whole unzipping was taking up a huge amount of time. Is there perhaps a way to optimize that?

Baseline function

First it's these common functions that simulate actually doing something with the files in the zip file:

def _count_file(fn):
    with open(fn, 'rb') as f:
        return _count_file_object(f)


def _count_file_object(f):
    # Note that this iterates on 'f'.
    # You *could* do 'return len(f.read())'
    # which would be faster but potentially memory 
    # inefficient and unrealistic in terms of this 
    # benchmark experiment. 
    total = 0
    for line in f:
        total += len(line)
    return total

Here's the simplest one possible:

def f1(fn, dest):
    with open(fn, 'rb') as f:
        zf = zipfile.ZipFile(f)
        zf.extractall(dest)

    total = 0
    for root, dirs, files in os.walk(dest):
        for file_ in files:
            fn = os.path.join(root, file_)
            total += _count_file(fn)
    return total

If I analyze it a bit more carefully, I find that it spends about 40% doing the extractall and 60% doing the looping over files and reading their full length.

First attempt

My first attempt was to try to use threads. You create an instance of zipfile.ZipFile, extract every file name within and start a thread for each name. Each thread is given a function that does the "meat of the work" (in this benchmark, iterating over the file and getting its total size). In reality that function does a bunch of complicated S3, Redis and PostgreSQL stuff but in my benchmark I just made it a function that figures out the total length of file. The thread pool function:

def f2(fn, dest):

    def unzip_member(zf, member, dest):
        zf.extract(member, dest)
        fn = os.path.join(dest, member.filename)
        return _count_file(fn)

    with open(fn, 'rb') as f:
        zf = zipfile.ZipFile(f)
        futures = []
        with concurrent.futures.ThreadPoolExecutor() as executor:
            for member in zf.infolist():
                futures.append(
                    executor.submit(
                        unzip_member,
                        zf,
                        member,
                        dest,
                    )
                )
            total = 0
            for future in concurrent.futures.as_completed(futures):
                total += future.result()
    return total

Result: ~10% faster

Second attempt

So perhaps the GIL is blocking me. The natural inclination is to try to use multiprocessing to spread the work across multiple available CPUs. But doing so has the disadvantage that you can't pass around a non-pickleable object so you have to send just the filename to each future function:

def unzip_member_f3(zip_filepath, filename, dest):
    with open(zip_filepath, 'rb') as f:
        zf = zipfile.ZipFile(f)
        zf.extract(filename, dest)
    fn = os.path.join(dest, filename)
    return _count_file(fn)



def f3(fn, dest):
    with open(fn, 'rb') as f:
        zf = zipfile.ZipFile(f)
        futures = []
        with concurrent.futures.ProcessPoolExecutor() as executor:
            for member in zf.infolist():
                futures.append(
                    executor.submit(
                        unzip_member_f3,
                        fn,
                        member.filename,
                        dest,
                    )
                )
            total = 0
            for future in concurrent.futures.as_completed(futures):
                total += future.result()
    return total

Result: ~300% faster

That's cheating!

The problem with using a pool of processors is that it requires that the original .zip file exists on disk. So in my web server, to use this solution, I'd first have to save the in-memory ZIP file to disk, then invoke this function. Not sure what the cost of that it's not likely to be cheap.

Well, it doesn't hurt to poke around. Perhaps, it could be worth it if the extraction was significantly faster.

But remember! This optimization depends on using up as many CPUs as it possibly can. What if some of those other CPUs are needed for something else going on in gunicorn? Those other processes would have to patiently wait till there's a CPU available. Since there's other things going on in this server, I'm not sure I'm willing to let on process take over all the other CPUs.

Conclusion

Doing it serially turns out to be quite nice. You're bound to one CPU but the performance is still pretty good. Also, just look at the difference in the code between f1 and f2! With concurrent.futures pool classes you can cap the number of CPUs it's allowed to use but that doesn't feel great either. What if you get the number wrong in a virtual environment? Of if the number is too low and don't benefit any from spreading the workload and now you're just paying for overheads to move the work around?

I'm going to stick with zipfile.ZipFile(file_buffer).extractall(temp_dir). It's good enough for this.

Want to try your hands on it?

I did my benchmarking using a c5.4xlarge EC2 server. The files can be downloaded from:

wget https://www.peterbe.com/unzip-in-parallel/hack.unzip-in-parallel.py
wget https://www.peterbe.com/unzip-in-parallel/symbols-2017-11-27T14_15_30.zip

The .zip file there is 34MB which is relatively small compared to what's happening on the server.

The hack.unzip-in-parallel.py is a hot mess. It contains a bunch of terrible hacks and ugly stuff but hopefully it's a start.

Django 2.0 came out a couple of weeks ago. It now supports "conditional aggregation" which is SQL standard I didn't even know about.

Before

So I have a Django app which has an endpoint that generates some human-friendly stats about the number of uploads (and their total size) in various different time intervals.

First of all, this is how it set up the time intervals:

today = timezone.now()
start_today = today.replace(hour=0, minute=0, second=0)
start_yesterday = start_today - datetime.timedelta(days=1)
start_this_month = today.replace(day=1)
start_this_year = start_this_month.replace(month=1)

And then, for each of these, there's a little function that returns a dict for each time interval:

def count_and_size(qs, start, end):
    sub_qs = qs.filter(created_at__gte=start, created_at__lt=end)
    return {
        'count': sub_qs.count(),
        'total_size': sub_qs.aggregate(size=Sum('size'))['size'],
}

numbers['uploads'] = {
    'today': count_and_size(upload_qs, start_today, today),
    'yesterday': count_and_size(upload_qs, start_yesterday, start_today),
    'this_month': count_and_size(upload_qs, start_this_month, today),
    'this_year': count_and_size(upload_qs, start_this_year, today),
}

What you get is exactly 2 x 4 = 8 queries. One COUNT and one SUM for each time interval. E.g.

SELECT SUM("upload_upload"."size") AS "size" 
FROM "upload_upload" 
WHERE ("upload_upload"."created_at" >= ...

SELECT COUNT(*) AS "__count" 
FROM "upload_upload" 
WHERE ("upload_upload"."created_at" >= ...

...6 more queries...

Middle

Oops. I think this code comes from a slightly rushed job. We can do the COUNT and the SUM at the same time for each query.

# New, improved count_and_size() function!
def count_and_size(qs, start, end):
    sub_qs = qs.filter(created_at__gte=start, created_at__lt=end)
    return sub_qs.aggregate(
        count=Count('id'),
        total_size=Sum('size'),
    )

numbers['uploads'] = {
    'today': count_and_size(upload_qs, start_today, today),
    'yesterday': count_and_size(upload_qs, start_yesterday, start_today),
    'this_month': count_and_size(upload_qs, start_this_month, today),
    'this_year': count_and_size(upload_qs, start_this_year, today),
}

Much better, now there's only one query per time bucket. So 4 queries in total. E.g.

SELECT COUNT("upload_upload"."id") AS "count", SUM("upload_upload"."size") AS "total_size" 
FROM "upload_upload" 
WHERE ("upload_upload"."created_at" >= ...

...3 more queries...

After

But we can do better than that! Instead, we use conditional aggregation. The syntax gets a bit hairy because there's so many keyword arguments, but I hope I've indented it nicely so it's easy to see how it works:

def make_q(start, end):
    return Q(created_at__gte=start, created_at__lt=end)

q_today = make_q(start_today, today)
q_yesterday = make_q(start_yesterday, start_today)
q_this_month = make_q(start_this_month, today)
q_this_year = make_q(start_this_year, today)

aggregates = upload_qs.aggregate(
    today_count=Count('pk', filter=q_today),
    today_total_size=Sum('size', filter=q_today),

    yesterday_count=Count('pk', filter=q_yesterday),
    yesterday_total_size=Sum('size', filter=q_yesterday),

    this_month_count=Count('pk', filter=q_this_month),
    this_month_total_size=Sum('size', filter=q_this_month),

    this_year_count=Count('pk', filter=q_this_year),
    this_year_total_size=Sum('size', filter=q_this_year),
)
numbers['uploads'] = {
    'today': {
        'count': aggregates['today_count'],
        'total_size': aggregates['today_total_size'],
    },
    'yesterday': {
        'count': aggregates['yesterday_count'],
        'total_size': aggregates['yesterday_total_size'],
    },
    'this_month': {
        'count': aggregates['this_month_count'],
        'total_size': aggregates['this_month_total_size'],
    },
    'this_year': {
        'count': aggregates['this_year_count'],
        'total_size': aggregates['this_year_total_size'],
    },
}

Voila! One single query to get all those pieces of data.
The SQL sent to PostgreSQL looks something like this:

SELECT 
  COUNT("upload_upload"."id") FILTER (WHERE ("upload_upload"."created_at" >= ...)) AS "today_count", 
  SUM("upload_upload"."size") FILTER (WHERE ("upload_upload"."created_at" >= ...)) AS "today_total_size", 

  COUNT("upload_upload"."id") FILTER (WHERE ("upload_upload"."created_at" >= ...)) AS "yesterday_count", 
  SUM("upload_upload"."size") FILTER (WHERE ("upload_upload"."created_at" >= ...)) AS "yesterday_total_size", 

  ...

FROM "upload_upload";

Is this the best thing to do? I'm starting to have my doubts.

Watch Out!

When I take this now 1 monster query for a spin with an EXPLAIN ANALYZE prefix I notice something worrying!

QUERY PLAN
-------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=74.33..74.34 rows=1 width=16) (actual time=0.587..0.587 rows=1 loops=1)
   ->  Seq Scan on upload_upload  (cost=0.00..62.13 rows=813 width=16) (actual time=0.012..0.210 rows=813 loops=1)
 Planning time: 0.427 ms
 Execution time: 0.674 ms
(4 rows)

A sequential scan! That's terrible. The created_at column is indexed in a BTREE so why can't it use the index.

The short answer is: I don't know!
I've uploaded a reduced, but still complete, example demonstrating this in a gist. It's very similar to the example in the stackoverflow question I asked.

So what did I do? I went back to the "middle" solution. One SELECT query per time bucket. So 4 queries in total, but at least all 4 is able to use an index.

This is an advanced topic for people who do serious stuff in Redis. I need to do serious stuff in Redis so I'm trying to learn about the best way to store lots of keys with hash maps.

It seems that this article by Salvatore Sanfilippo (creator of Redis) himself seems to be a much cited article for this topic. If you haven't read it, the gist is that Redis can employ some clever optimizations for storing hash maps in a very memory efficient way instead of storing each key-value separately.

"Hashes, Lists, Sets composed of just integers, and Sorted Sets, when smaller than a given number of elements, and up to a maximum element size, are encoded in a very memory efficient way that uses up to 10 times less memory (with 5 time less memory used being the average saving)"

This efficient storage optimization is called a ziplist.

tl;dr; Whatsdeployed.io is an impressively simple web app to help web developers and web ops people quickly see what GitHub commits have made it into your Dev, Stage or Prod environment. Today it got a facelift.

The code is now more than 5 years old and has served me well. It's weird to talk too positively about the app because I actually wrote it but because it's so simple in terms of design and effort it feels less personal to talk about it.

Here's what's in the facelift

• Upgraded to Bootstrap 4.
• Instead of relying on downloading a heavy Glyphicon web font, just to display a single checkmark, that's now a simple image.
• Ability to use a GitHub developer personal token to avoid rate limitations on GitHub's API.
• The first lookup to get all commits is now done via the Flask app to use my auth token to avoid the rate limit.
• Much better error handling if any of the underlying requests.get() that the Flask app does, fails. Also includes which URL it failed on.
• Basic validation to prevent submitting the main form without typing anything in.
• You can hack on it with Docker. Thanks @willkg.
• Improved the code that extracts Bugzilla bug numbers out of commit messages. Thanks @edmorely.
• Refreshed screenshots in the README.md
• A brand new introduction text on the home page for people who end up on the site not knowing what it is.
• If any XHR errors happen figuring out the "culprits", you now get a pretty error describing this instead of swallowing it all.

Please let me know if there's anything broken or missing.

This is an update to a old blog post from 2006 called Fastest way to uniquify a list in Python. But this, time for Python 3.6. Why, because Python 3.6 preserves the order when inserting keys to a dictionary. How, because the way dicts are implemented in 3.6, the way it does that is different and as an implementation detail the order gets preserved. Then, in Python 3.7, which isn't released at the time of writing, that order preserving is guaranteed.

Anyway, Raymond Hettinger just shared a neat little way to uniqify a list. I thought I'd update my old post from 2006 to add list(dict.fromkeys('abracadabra')).

Functions

Reminder, there are two ways to uniqify a list. Order preserving and not order preserving. For example, the unique letters in peter is p, e, t, r in their "original order". As opposed to t, e, p, r.

def f1(seq):  # Raymond Hettinger
    hash_ = {}
    [hash_.__setitem__(x, 1) for x in seq]
    return hash_.keys()

def f3(seq):
    # Not order preserving
    keys = {}
    for e in seq:
        keys[e] = 1
    return keys.keys()

def f5(seq, idfun=None):  # Alex Martelli ******* order preserving
    if idfun is None:
        def idfun(x): return x
    seen = {}
    result = []
    for item in seq:
        marker = idfun(item)
        # in old Python versions:
        # if seen.has_key(marker)
        # but in new ones:
        if marker in seen:
            continue
        seen[marker] = 1
        result.append(item)
    return result

def f5b(seq, idfun=None):  # Alex Martelli ******* order preserving
    if idfun is None:
        def idfun(x): return x
    seen = {}
    result = []
    for item in seq:
        marker = idfun(item)
        # in old Python versions:
        # if seen.has_key(marker)
        # but in new ones:
        if marker not in seen:
            seen[marker] = 1
            result.append(item)

    return result

def f7(seq):
    # Not order preserving
    return list(set(seq))

def f8(seq):  # Dave Kirby
    # Order preserving
    seen = set()
    return [x for x in seq if x not in seen and not seen.add(x)]

def f9(seq):
    # Not order preserving, even in Py >=3.6
    return {}.fromkeys(seq).keys()

def f10(seq, idfun=None):  # Andrew Dalke
    # Order preserving
    return list(_f10(seq, idfun))

def _f10(seq, idfun=None):
    seen = set()
    if idfun is None:
        for x in seq:
            if x in seen:
                continue
            seen.add(x)
            yield x
    else:
        for x in seq:
            x = idfun(x)
            if x in seen:
                continue
            seen.add(x)
            yield x

def f11(seq):  # f10 but simpler
    # Order preserving
    return list(_f10(seq))

def f12(seq):
    # Raymond Hettinger
    # https://twitter.com/raymondh/status/944125570534621185
    return list(dict.fromkeys(seq))

Results

FUNCTION        ORDER PRESERVING     MEAN       MEDIAN
f12             yes                  111.0      112.2
f8              yes                  266.3      266.4
f10             yes                  304.0      299.1
f11             yes                  314.3      312.9
f5              yes                  486.8      479.7
f5b             yes                  494.7      498.0
f7              no                   95.8       95.1
f9              no                   100.8      100.9
f3              no                   143.7      142.2
f1              no                   406.4      408.4

Conclusion

The fastest way to uniqify a list of hashable objects (basically immutable things) is:

list(set(seq))

And the fastest way, if the order is important is:

list(dict.fromkeys(seq))

Now we know.

tl;dr; I see no reason worth switching to Msgpack instead of good old JSON.

I was curious, how much more efficient is Msgpack at packing a bunch of data into a file I can emit from a web service.

In this experiment I take a massive JSON file that is used in a single-page-app I worked on. If I download the file locally as a .json file, the file is 2.1MB.

Converting it to Msgpack:

>>> import json, msgpack
>>> with open('events.json') as f:
...   events=json.load(f)
...
>>> len(events)
3
>>> events.keys()
dict_keys(['max_modified', 'events', 'urls'])
>>> with open('events.msgpack', 'wb') as f:
...   f.write(msgpack.packb(events))
...
1880266

Now, let's compared the two file formats, as seen on disk:

▶ ls -lh events*
-rw-r--r--  1 peterbe  wheel   2.1M Dec 19 10:16 events.json
-rw-r--r--  1 peterbe  wheel   1.8M Dec 19 10:19 events.msgpack

But! How well does it compress?

More common than not your web server can return content encoded in Gzip as content-encoding: gzip. So, let's compare that:

▶ gzip events.json ; gzip events.msgpack
▶ ls -l events*
-rw-r--r--  1 peterbe  wheel  304416 Dec 19 10:16 events.json.gz
-rw-r--r--  1 peterbe  wheel  305905 Dec 19 10:19 events.msgpack.gz

Oh my! When you gzip the files the .json file ultimately becomes smaller. By a whopping 0.5%!

What about speed?

First let's open the files a bunch of times and see how long it takes to unpack:

def f1():
    with open('events.json') as f:
        s = f.read()
    t0 = time.time()
    events = json.loads(s)
    t1 = time.time()
    assert len(events['events']) == 4365
    return t1 - t0


def f2():
    with open('events.msgpack', 'rb') as f:
        s = f.read()
    t0 = time.time()
    events = msgpack.unpackb(s, encoding='utf-8')
    t1 = time.time()
    assert len(events['events']) == 4365
    return t1 - t0


def f3():
    with open('events.json') as f:
        s = f.read()
    t0 = time.time()
    events = ujson.loads(s)
    t1 = time.time()
    assert len(events['events']) == 4365
    return t1 - t0

(Note that the timing is around the json.loads() etc without measuring how long it takes to get the files to strings)

Result (using Python 3.6.1): All about the same.

FUNCTION: f1 Used 56 times
    MEDIAN 30.509352684020996
    MEAN   31.09178798539298
    STDEV  3.5620914333233595
FUNCTION: f2 Used 68 times
    MEDIAN 27.882099151611328
    MEAN   28.704492484821994
    STDEV  3.353800228776872
FUNCTION: f3 Used 76 times
    MEDIAN 27.746915817260742
    MEAN   27.920340236864593
    STDEV  2.21554251130519

Same benchmark using PyPy 3.5.3, but skipping the f3() which uses ujson:

FUNCTION: f1 Used 99 times
    MEDIAN 20.905017852783203
    MEAN   22.13949386519615
    STDEV  5.142071370453135
FUNCTION: f2 Used 101 times
    MEDIAN 36.96393966674805
    MEAN   40.54664857316725
    STDEV  17.833577642246738

Dicussion and conclusion

One of the benefits of Msgpack is that it can used for streaming. "Streaming unpacking" as they call it. But, to be honest, I've never used it. That can useful when you have structured data trickling in and you don't want to wait for it all before using the data.

Another cool feature Msgpack has is ability to encode custom types. E.g. datetime.datetime. Like bson can do. With JSON you have to, for datetime objects do string conversions back and forth and the formats are never perfectly predictable so you kinda have to control both ends.

But beyond some feature differences, it seems that JSON compressed just as well as Msgpack when Gzipped. And unlike Msgpack JSON is not binary so it's easy to poke around with any tool. And decompressing JSON is just as fast. Almost. But if you need to squeeze out a couple of extra free milliseconds from your JSON files you can use ujson.

Conclusion; JSON is fine. It's bigger but if you're going to Gzip anyway, it's just as small as Msgpack.

Bonus! BSON

Another binary encoding format that supports custom types is BSON. This one is a pure Python implementation. BSON is used by MongoDB but this bson module is not what PyMongo uses.

Size comparison:

▶ ls -l events*son
-rw-r--r--  1 peterbe  wheel  2315798 Dec 19 11:07 events.bson
-rw-r--r--  1 peterbe  wheel  2171439 Dec 19 10:16 events.json

So it's 7% larger than JSON uncompressed.

▶ ls -l events*son.gz
-rw-r--r--  1 peterbe  wheel  341595 Dec 19 11:07 events.bson.gz
-rw-r--r--  1 peterbe  wheel  304416 Dec 19 10:16 events.json.gz

Meaning it's 12% fatter than JSON when Gzipped.

Doing a quick benchmark with this:

def f4():
    with open('events.bson', 'rb') as f:
        s = f.read()
    t0 = time.time()
    events = bson.loads(s)
    t1 = time.time()
    assert len(events['events']) == 4365
    return t1 - t0

Compared to the original f1() function:

FUNCTION: f1 Used 106 times
    MEDIAN 29.58393096923828
    MEAN   30.289863640407347
    STDEV  3.4766612593557173
FUNCTION: f4 Used 94 times
    MEDIAN 231.00042343139648
    MEAN   231.40889786659403
    STDEV  8.947746458066405

In other words, bson is about 600% slower than json.

This blog post was supposed to be about how well the individual formats size up against each other on disk but it certainly would be interesting to do a speed benchmark comparing Msgpack and JSON (and maybe BSON) where you have a bunch of datetimes or decimal.Decimal objects and see if the difference is favoring the binary formats.

I use this sometimes to get insight into how long some view functions take. Perhaps you find it useful too:

def view_function_timer(prefix='', writeto=print):

    def decorator(func):
        @functools.wraps(func)
        def inner(*args, **kwargs):
            try:
                t0 = time.time()
                return func(*args, **kwargs)
            finally:
                t1 = time.time()
                writeto(
                    'View Function',
                    '({})'.format(prefix) if prefix else '',
                    func.__name__,
                    args[1:],
                    'Took',
                    '{:.2f}ms'.format(1000 * (t1 - t0)),
                    args[0].build_absolute_uri(),
                )
        return inner

    return decorator

And to use it:

from wherever import view_function_timer


@view_function_timer()
def homepage(request, thing):
    ...
    return render(request, template, context)

And then it prints something like this:

View Function  homepage ('valueofthing',) Took 23.22ms http://localhost:8000/home/valueofthing

It's useful when you don't want a full-blown solution to measure all view functions with a middleware or something.
It can be useful also to see how a cache decorator might work:

from django.views.decorators.cache import cache_page
from wherever import view_function_timer


@view_function_timer('possibly cached')
@cache_page(60 * 60 * 2)  # two hours cache
@view_function_timer('not cached')
def homepage(request, thing):
    ...
    return render(request, template, context)

That way you can trace that, with tail -f or something, to see how/if the cacheing decorator works.

There are many better solutions that are more robust but might be a bigger investment. For example, I would recommend markus which, if you don't have a statsd server you can configure to logger.info call the timings.

The documentation about how to use synonyms in Elasticsearch is good but because it's such an advanced topic, even if you read the documentation carefully, you're still left with lots of questions. Let me show you some things I've learned about how to use synonyms in Python with elasticsearch-dsl.

What's the nature of your documents?

I'm originally from Sweden but moved to London, UK in 1999 and started blogging a few years after. So I wrote most of my English with British English spelling. E.g. "centre" instead of "center". Later I moved to California in the US and slowly started to change my own English over to American English. I kept blogging but now I would prefer to write "center" instead of "centre".

Another example... Certain technical words or namings are tricky. For example, is it "go" or is it "golang"? Is it "React" or is it "ReactJS"? Is it "PostgreSQL" or "Postgres". I never know. Not only is it sometimes hard to know which is right because people use them differently, but also sometimes "brands" like that change over time since inception, the creator might have preferred something but the masses of people call it something else.

So with all that in mind, not only has the nature of my documents (my blog post texts) changed in terminology over the years. My visitors are also coming both from British English and American English. Or, suppose that I knew the perfect way to phrase that relational database that starts with "Postg...". Even if my text is always spelled one particular way, perfectly, my visitors will most likely refer to it as "postgres" sometimes and "postgresql" sometimes.

The simple solution, match all!

Create a custom analyzer

Let's jump straight into the code. People who have used elasticsearch_dsl should be familiar with most of this:

from elasticsearch_dsl import (
    DocType,
    Text,
    Index,
    analyzer,
    Keyword,
    token_filter,
)
from django.conf import settings


index = Index(settings.ES_INDEX)
index.settings(**settings.ES_INDEX_SETTINGS)


synonym_tokenfilter = token_filter(
    'synonym_tokenfilter',
    'synonym',
    synonyms=[
        'reactjs, react',  # <-- important
    ],
)

text_analyzer = analyzer(
    'text_analyzer',
    tokenizer='standard',
    filter=[
        # The ORDER is important here.
        'standard',
        'lowercase',
        'stop',
        synonym_tokenfilter,
        # Note! 'snowball' comes after 'synonym_tokenfilter'
        'snowball',
    ],
    char_filter=['html_strip']
)

class BlogItemDoc(DocType):
    oid = Keyword(required=True)
    title = Text(
        required=True, 
        analyzer=text_analyzer
    )
    text = Text(analyzer=text_analyzer)

index.doc_type(BlogItemDoc)

This code above is copied from the "real code" but a lot of distracting things that aren't important to the point, have been removed.

The magic sauce here is that you create a token_filter and you can call it whatever you want. I called mine synonym_tokenfilter and that's also what the instance variable is called.

Notice the list of synonyms. It's a plain list of strings. Specifically, it's a list of 1 string reactjs, react.

Let's see how Elasticsearch analyzes this:
First with the text react.

$ curl -XGET 'http://127.0.0.1:9200/peterbecom/_analyze?analyzer=text_analyzer&text=react&pretty=1'
{
  "tokens" : [
    {
      "token" : "react",
      "start_offset" : 0,
      "end_offset" : 5,
      "type" : "<ALPHANUM>",
      "position" : 0
    },
    {
      "token" : "reactj",
      "start_offset" : 0,
      "end_offset" : 5,
      "type" : "SYNONYM",
      "position" : 0
    }
  ]
}

Note that the analyzer snowball, converted reactjs to reactj which is wrong in a sense, because there's not plural "reacts", but it ultimately doesn't matter much. At least not in this particular case.

Secondly, analyze it with the text reactjs:

$ curl -XGET 'http://127.0.0.1:9200/peterbecom/_analyze?analyzer=text_analyzer&text=reactjs&pretty=1'
{
  "tokens" : [
    {
      "token" : "reactj",
      "start_offset" : 0,
      "end_offset" : 7,
      "type" : "<ALPHANUM>",
      "position" : 0
    },
    {
      "token" : "react",
      "start_offset" : 0,
      "end_offset" : 7,
      "type" : "SYNONYM",
      "position" : 0
    }
  ]
}

Same tokens! Just different order.

Test it for reals

Now, the real proof is in actually doing a search on this. Look at these two screenshots:

It worked! Different ways of phrasing your search but ultimately found all the documents that matched independent of different people or different authors might prefer to spell it.

Try it for yourself:

• https://www.peterbe.com/search?q=reactjs&debug-search=1
• https://www.peterbe.com/search?q=react&debug-search=1
• https://www.peterbe.com/search?q=postgres&debug-search=1
• https://www.peterbe.com/search?q=postgresql&debug-search=1

What it looked like before

Check out these two screenshots of how it would look like before, when synonyms for postgres and postgresql had not been set up yet:

One immediate thought I have is what a mess I've been in blogging about that database. Clearly I struggled to pick one way to spell it consistently.

And here's what it would look like once that synonym has been set up:

"go" versus "golang"

Go is a programming language. That term, too, struggles with a name ambiguity. Granted, I rarely hear people say "golang", but it's definitely a written word that turns up a lot.

The problem with setting up a synonym for go == golang is that "go" is common English word. It's also the stem of the word "going" and such. So if you set up a synonym, like I did for react and reactjs above, this is what happens:

This is now the exact search results as if I had searched for go. But look what it matched! It matched "Go" (good) but also "Going real simple..." (bad) and "...I should go" (bad).

If someone searches for the simple term "go" they probably intend to search for the Go programming language. All that snowball stemming is critical for a bunch of other non-computer-term searches so we can't remove the stemming.

The solution is to use what's called "Simple Contraction". And it looks like this:

all_synonyms = [
    'go => golang',
    'react => reactjs',
    'postgres => postgresql',
]

That basically means that a search for go is a search for golang. And a document that uses the word go (alone) is indexed as golang.

What happens is that the word go gets converted to golang which doesn't get stemming converted down to any other forms.

However, this is no silver bullet. Any search for the term go is ultimately a search for the word golang and the regular English word go. So the benefit of all of this was that we got rid of search results matching on going and gone.

What you have to decide...

The case for go is similar to the case for react. Both of these words are nouns but they're also verbs.

Should people find "reacting to events" when they search for "react"? If so, use react, reactjs in the synonyms list.

Should people only find documents related to noun "React" when they search for "event handing in react"? If so, use react => reactjs in the synonyms list.

It's up to you and your documents and what your users tend to search for.

Bonus! For American vs British English

AVKO.org publishes a list of all British to American English synonyms. You can download the whole list here. Unfortunately I can't find a license for this file but the compiled synonyms file is part of this repo which is licensed under MIT.

I download this list and keep it in the repo. Then when setting up the analyzer and token filters I load it in like this:

synonyms_root = os.path.join(
    settings.BASE_DIR, 'peterbecom/es-synonyms'
)
american_british_syns_fn = os.path.join(
    synonyms_root, 'be-ae.synonyms'
)

with open(american_british_syns_fn) as f:
    for line in f:
        if (
            '=>' not in line or 
             line.strip().startswith('#')
         ):
            continue
        all_synonyms.append(line.strip())

Now I can finally enjoy not having to worry about the fact that sometimes I spell it "license" and sometimes I spell it "licence". It's all the same now. Brits and Americans, rejoice on common ground!

Bonus! For terrible spellers

Although I don't have a big problem with this on my techy blog but you can use the Simple Contraction technique to list unambiguously bad spelling. Add dont => don't to the list of synonyms and a search for dont is a search for don't.

Last but not least, the official Elasticsearch documentation is the place to go. This blog post hopefully phrases it in more approachable terms. Especially for Python peeps.