Creating a Prometheus parser: Fennel

A quick tour of using FParsec to write a Prometheus parser

F# Prometheus Fennel FsAdvent

A year back I ran into the need for a library that provided a model for creating valid Prometheus log lines. The libraries I looked at sent these metrics for export rather than giving me access to the model or allowing me to create the corresponding log string. I had been wanting to play around with FParsec for a while so this seemed like a perfect opportunity to give it a try.

This post is part of FsAdvent 2020.

The result was a library called Fennel. It can parse Prometheus text to objects, and turn these metric objects into valid Prometheus text.

This was my first time using a library to do a custom parser. In the past when I had needed to parse text I had used a state machine and consumed a character at a time. The idea here is that depending on the state, you expect certain characters to follow. See here for an example. It turns out this is not too different to how you use a library like FParsec. Although there is a bit of a learning curve, and not many resources outside of the docs, using FParsec was fun. I am sure there are 100 ways I could improve the parser (feedback welcome...preferably be polite) but I am happy with the end result.


This post is not meant to be a tutorial on FParsec but in-case you have never used it, let's look at some of the things it allows you to do.

FParsec gives you a boatload of parsers that can be combined to make a new parser. Parser factory functions like satisfy will give you back a Parser<> that satisfies the given predicate. The library also gives you some operators. Below <|> means try the first parser, if that fails, try the second.

The example below also uses manyChar2 which uses the first parser for the first char and then the next for all following chars. In this case, because a Prometheus metric name must start with an ASCII letter or an underscore (not a number).

let underscoreOrColon = satisfy (fun c -> c = '_' || c = ':')
let ascii_alpha_numeric = (asciiLetter <|> digit)
let pname = manyChars2 (asciiLetter <|> underscoreOrColon) (ascii_alpha_numeric <|> underscoreOrColon)

These parsers can then be combined in other ways. The code below combines the pname parser with the "zero or more" whitespace parser but because the period is on the left of the .>> operator it takes only the result of pname (.>> and .>>. are available). The |>> operator returns a parser that takes the result of the parser to the left and applies the function to the right.

let private metric_name = pname.>> ws0 |>> MetricName

This is just a tiny taste of how you can build up a complex parser from simpler ones. Combining these you can start to build up a grammar for your parsers. Next we look at building this further with our Prometheus parser.

Prometheus parser

As a reminder, the Prometheus format is a text-based format.

# This is a comment but the following 2 have meaning
# HELP http_requests_total The total number of HTTP requests.
# TYPE http_requests_total counter
http_requests_total{method="post",code="200"} 1027 1395066363000
http_requests_total{method="post",code="400"}    3 1395066363000

You can read up on the exposition format here.

The model looks like this:

// details and types excluded for brevity
type MetricLine = {
    Name : MetricName
    Labels : Label list
    Value : MetricValue
    Timestamp : Timestamp option

type Line =
    | Help of (MetricName*DocString)
    | Comment of string
    | Type of (MetricName*MetricType)
    | Metric of MetricLine
    | Blank

You can see the full model on the GitHub repository

Any Prometheus log line can be Help information, a normal comment, type information, a metric, or a blank line. From a parsing point of view, I categorize comments, TYPE line, and HELP line all as comments since the # is a common first character. This is not reflected in the model.

So let's break down the Prometheus text and how it relates to the model above.

  1. A line in some Prometheus text can be blank for a Prometheus log line
  2. A Prometheus log line can be a comment or a metric
  3. A comment can be a just a normal comment, TYPE information, or HELP information
  4. A metric requires a name and value
  5. A metric can optionally have labels and a timestamp

Let's look at a few select parsers and see how they match with our description above. We will focus on the comment line of TYPE and how it fits in.

// TYPE from point 3
let typeLine = (``TYPE``>>.metric_name.>>.metric_type) |>> Line.Type
let comment = comment_prefix >>.ws0 >>.(typeLine <|> helpLine <|> commentLine)
// Point 2
let line = ws0 >>.(comment <|> metric)
// Point 1
ws0 >>.(line <|> emptyLine)


So that was a little under the hood of Fennel. What does it look like from the outside?

We can turn Prometheus log text into strongly typed models.

open Fennel

let input = """
# Finally a summary, which has a complex representation, too:
# HELP rpc_duration_seconds A summary of the RPC duration in seconds.
# TYPE rpc_duration_seconds summary
rpc_duration_seconds{quantile="0.01"} 3102
rpc_duration_seconds{quantile="0.05"} 3272
rpc_duration_seconds{quantile="0.5"} 4773
rpc_duration_seconds{quantile="0.9"} 9001
rpc_duration_seconds{quantile="0.99"} 76656
rpc_duration_seconds_sum 1.7560473e+07
rpc_duration_seconds_count 2693

let lines = Prometheus.parseText input

Each of these lines can match a specific line type:

match line with
| Help (name, doc) -> printfn "Help line %A" (name, doc)
| Comment txt -> printfn "Comment line %s" txt
| Type (name, t) -> printfn "Type line %A" (name, t)
| Metric m -> printfn "Metric line %A" m
| Blank -> printfn "Blank line"

And we can create an object that represents a Prometheus log line.

open Fennel

let prometheusString = Prometheus.metric "http_requests_total" 1027. [("method","post");("code","200")] DateTimeOffset.UtcNow


So that was a little peek into creating my first parser. Have you used FParsec? If not, was this helpful?
Do you have plenty of experience with it? What can I improve?
Leave a comment or create an issue or PR on the repo.

Photo by Gleb Lukomets on Unsplash

blog comments powered by Disqus