Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

RIPE Atlas Sagan Build Status Documentation Status

A parsing library for RIPE Atlas measurement results

Why this exists

RIPE Atlas generates a lot of data, and the format of that data changes over time. Often you want to do something simple like fetch the median RTT for each measurement result between date X and date Y. Unfortunately, there are dozens of edge cases to account for while parsing the JSON, like the format of errors and firmware upgrades that changed the format entirely.

To make this easier for our users (and for ourselves), we wrote an easy to use parser that's smart enough to figure out the best course of action for each result, and return to you a useful, native Python object.

How to install

The stable version should always be in PyPi, so you can install it with pip:

$ pip install ripe.atlas.sagan

Better yet, make sure you get ujson and sphinx installed with it:

$ pip install ripe.atlas.sagan[fast,doc]


Some setups (like MacOS) have trouble with building the dependencies required for reading SSL certificates. If you don't care about SSL stuff and only want to use sagan to say, parse traceroute or DNS results, then you can do the following:

$ SAGAN_WITHOUT_SSL=1 pip install ripe.atlas.sagan

Quickstart: How To Use It

You can parse a result in a few ways. You can just pass the JSON-encoded string:

from ripe.atlas.sagan import PingResult

my_result = PingResult("<result string from RIPE Atlas ping measurement>")



You can do the JSON-decoding yourself:

from ripe.atlas.sagan import PingResult

my_result = PingResult(
    json.loads("<result string from RIPE Atlas ping measurement>")



You can let the parser guess the right type for you, though this incurs a small performance penalty:

from ripe.atlas.sagan import Result

my_result = Result.get("<result string from RIPE Atlas ping measurement>")



What it supports

Essentially, we tried to support everything. If you pass in a DNS result string, the parser will return a DNSResult object, which contains a list of Response's, each with an abuf property, as well as all of the information in that abuf: header, question, answer, etc.

from ripe.atlas.sagan import DnsResult

my_dns_result = DnsResult("<result string from a RIPE Atlas DNS measurement>")
my_dns_result.responses[0].abuf  # The entire string
my_dns_result.responses[0].abuf.header.arcount  # Decoded from the abuf

We do the same sort of thing for SSL measurements, traceroutes, everything. We try to save you the effort of sorting through whatever is in the result.

Which attributes are supported?

Every result type has its own properties, with a few common between all types.

Specifically, these attributes exist on all *Result objects:

  • created An datetime object of the timestamp field
  • measurement_id
  • probe_id
  • firmware An integer representing the firmware version
  • origin The from attribute in the result
  • is_error Set to True if an error was found

Additionally, each of the result types have their own properties, like packet_size, responses, certificates, etc. You can take a look at the classes themselves, or just look at the tests if you're curious. But to get you started, here are some examples:

# Ping
ping_result.packets_sent  # Int
ping_result.rtt_median    # Float, rounded to 3 decimal places
ping_result.rtt_average   # Float, rounded to 3 decimal places

# Traceroute                   # 4 or 6
traceroute_result.total_hops           # Int
traceroute_result.destination_address  # An IP address string

dns_result.responses                        # A list of Response objects
dns_result.responses[0].response_time       # Float, rounded to 3 decimal places
dns_result.responses[0].headers             # A list of Header objects
dns_result.responses[0].headers[0].nscount  # The NSCOUNT value for the first header
dns_result.responses[0].questions           # A list of Question objects
dns_result.responses[0].questions[0].type   # The TYPE value for the first question
dns_result.responses[0].abuf                # The raw, unparsed abuf string

# SSL Certificates                        # 4 or 6
ssl_result.certificates              # A list of Certificate objects
ssl_result.certificates[0].checksum  # The checksum for the first certificate

# HTTP                      # 4 or 6
http_result.uri                     # A URL string
http_result.responses               # A list of Response objects
http_result.responses[0].body_size  # The size of the body of the first response

# NTP                          # 4 or 6
ntp_result.stratum                     # Statum id
ntp_result.version                     # Version number
ntp_result.packets[0].final_timestamp  # A float representing a high-precision NTP timestamp
ntp_result.rtt_median                  # Median value for packets sent & received

What it requires

As you might have guessed, with all of this magic going on under the hood, there are a few dependencies:

Additionally, we recommend that you also install ujson as it will speed up the JSON-decoding step considerably, and sphinx if you intend to build the documentation files for offline use.

Running Tests

There's a full battery of tests for all measurement types, so if you've made changes and would like to submit a pull request, please run them (and update them!) before sending your request:

$ python test

You can also install tox to test everything in all of the supported Python versions:

$ pip install tox
$ tox

Further Documentation

Complete documentation can always be found on Read the Docs, and if you're not online, the project itself contains a docs directory -- everything you should need is in there.

Who's Responsible for This?

Sagan is actively maintained by the RIPE NCC and primarily developed by Daniel Quinn, while the abuf parser is mostly the responsibility of Philip Homburg with an assist from Bert Wijnen and Rene Wilhelm who contributed to the original script. Andreas Stirkos did the bulk of the work on NTP measurements and fixed a few bugs, and big thanks go to Chris Amin, John Bond, and Pier Carlo Chiodi for finding and fixing stuff where they've run into problems.


But why "Sagan"? The RIPE Atlas team decided to name all of its modules after explorers, and what better name for a parser than that of the man who spent decades reaching out to the public about the wonders of the cosmos?