What the Grok!? - A Python script to convert grok epxressions to regex

Posted on Wed 17 August 2016 in Regex • 3 min read

What is Grok?

If you’ve never used Grok you’re missing out. It’s fantastic for parsing (semi-?)structured data using regular expressions. The basic premise is that you construct a complex expression by peicing together smaller epxressions. Each expression could actually be a raw regex, a collection of other expressions, or a mix of both.

A simple example

Say you had a simple log message with an ISO8601 timestamp:

2016-08-17 20:07:22 - Hello there

That’s easy right? Right!?

Well, what if you need to parse that into the full date and time, their individual components, and the message? You could write a parser but maybe you can’t hook that into an existing toolset. Maybe you’re not a developer. Maybe your cat ate your ‘Parsers for Dummies’ book. BTW, your cat is a monster.

Enter Regex - You’re winning already!

Oh. Wait. Nope, it’s not quite that easy. Unless you really like regex you’re probably going to be lazy and skip the part of the spec that said you needed each component of the date. You’ll end up with something like this:

^(?<date>[^\s]+) (?<time>[^\s]+) - (?<message>.*)$

Let’s take a look at that on regex101.com an online regex testing tool.

That’s not horrible and it works. But parsing the date and time components is extra work and I’m pretty lazy sometimes.

Enter Grok - Are we done yet?

Keep in mind this example is fairly trivial. Using Grok we have access to a library of prebuilt patterns for things like dates, numbers, text, etc.

Here’s our Grok pattern:


That’s a minimal pattern that will match and provide most of what we’re looking for. We can be more explicit though and meet all of our requirements.

Let’s break down the TIMESTAMP_ISO8601 pattern. You’ll find it defined in the grok-patterns file:


Let’s just enhance that a bit.

%{YEAR:year}-%{MONTHNUM:month}-%{MONTHDAY:day}[T ]%{HOUR:hour}:?%{MINUTE:minute}(?::?%{SECOND:second})?%{ISO8601_TIMEZONE:timezone}? - %{GREEDYDATA:message}

That’s a much more complete pattern and it provides names for all the items (note the %{PATTER:name} syntax). And, it was still pretty easy since we modified an existing pattern.

A non-trivial example

HAProxy logs

HAProxy has some pretty nice logs containing tons of information. They’re also well structured and should be easy to parse. However, when you’re limited to using regex it quickly becomes a nightmare. Luckily, thanks to the pattern library we have a prebuilt pattern to parse HAProxy logs.

The log format is also very well documented.

Let’s see what a log line looks like (line breaks added):

Feb  6 12:14:14 localhost \
      haproxy[14389]: [06/Feb/2009:12:14:14.655] http-in \
            static/srv1 10/0/30/69/109 200 2750 - - ---- 1/1/1/1/0 0/0 {1wt.eu} \
                  {} "GET /index.html HTTP/1.1"

Lot’s of work went into this behind the scenes so that all we need is this Grok pattern:


That’s it. That’s everything. It provides proper matching and named groups for all the log elements.

That’s great, but what about my regex?

Finally, we’re at the end. We have a Grok pattern that properly handles our logs but we need a regex to put into some other tool. Unfortunately, I was unable to find anything that would provide the final compiled regex of a Grok pattern.

So, I wrote one.

Before we see the code, let’s see what it can do!

How about an IP address?

Let’s see what the regex looks like for an IP address:

# First, clone the patterns repo
git clone git@github.com:logstash-plugins/logstash-patterns-core.git

# And run the tool
./grok-to-regex.py -d logstash-patterns-core/patterns/ '%{IP:client_ip}'

Wow, that’s a lot of regex! But, it’ll match both IPV4 and IPV6 addresses and I didn’t do any real work. : )

Back to HAProxy logs for a moment

Since this all started with HAProxy logs lets see what they look like.

./grok-to-regex.py -d logstash-patterns-core/patterns/ '%{HAPROXYHTTP}'

For the sake of your eyes I’m not going to insert the output here. Instead, let’s take a look at it on regex101.com again.

And now, finally, the code

It’s pure python and doesn’t depend on any extra bits and peices. I suspect, though YMMV, that it will work on Windows too.