scamper-pywarts

About pywarts

pywarts is a pure-python parsing library for the Warts format. Warts is an extensible binary format produced by Scamper, an Internet measurement tool from CAIDA, to store measurement results such as traceroutes and pings.

This library started off from the Python implementation from CMAND, by Robert Beverly, but has now vastly diverged. The parsing architecture is loosely inspired from the Ryu packet parser, although it is less complex because the requirements are less stringent.

Installation

pip install scamper-pywarts

Features

• pure-Python, no dependency, works with both python2 and python3
• can read all basic Warts data types (ping, traceroute)
• easily extensible for other Warts data types (patches are welcome)
• nice class-based interface
• reasonably good performance (a few minutes to parse a 80 MiB warts file with traceroute data)
• streaming-like interface: no more than one record is pulled in memory at any given time, so it should handle very large Warts file with a limited amount of memory. You can probably even consume data directly from the output of a running Scamper process.

Using pywarts

For now, the only public API is very low-level: it simply reads from a stream (for instance a file) and returns Warts records as Python objects.

To read records, call warts.parse_record repeatedly until it returns None. Remember to open your input Warts files in binary mode!

The returned value of warts.parse_record is an instance of an appropriate subclass (e.g. Traceroute), depending on the record type. Be aware that all optional attributes are set to None if not present in the input file. You should always check for this possibility in your user code.

Here is an example that opens a file, and repeatedly parses records until it finds a Traceroute record (warts files usually have a few initial records with mostly uninteresting data).

import warts
from warts.traceroute import Traceroute

with open('my_file.warts', 'rb') as f:
    record = warts.parse_record(f)
    while not isinstance(record, Traceroute):
        record = warts.parse_record(f)
    if record.src_address:
        print("Traceroute source address:", record.src_address)
    if record.dst_address:
        print("Traceroute destination address:", record.dst_address)
    print("Number of hops:", len(record.hops))
    print(record.hops)

To know which attributes are available, look at the definition of the relevant class (there will be real documentation at some point). For instance, for Traceroute, almost all attributes are optional and defined here: traceroute.py. Some attributes are not optional and are defined in the parse() method of the class. For instance, a traceroute object t always provides a list of TracerouteHop objects in t.hops.

If parsing fails, an instance of errors.ParseError is thrown. pywarts generally tries to clean up after itself, so the file descriptor should point to the next record even after a parsing error. Of course, this is not always possible, especially if the input file is incorrectly formatted.

Difference with the implementation from CMAND

Here is some points on which pywarts improves from the code from https://github.com/cmand/scamper:

• fully python3-compatible
• nicer class-based interface, instead of huge dicts with all flags
• properly handles unknown flags and options, by ignoring them
• attribute names have been generally made more readable (although that often means longer names)
• possibly quite a bit faster (it would need proper benchmarks), because of the way we parse flags and strings. Also, we read a whole record into memory before parsing it, which is a bit faster than calling read() repeatedly on very small amount of data.

However, there are some areas where the CMAND code does more things:

• pywarts does not implement the deprecated address format (it is quite complex and has been deprecated for several years)
• there are some nice scripts in https://github.com/cmand/scamper, for instance a script to attach to and control a running Scamper process

Developement

High-level

Some currently unanswered questions:

• What should the high-level API look like, and is there even a need for a higher-level API? Just an iterator of records? Allow to filter by record type? Try to parse further, for instance decode flags or produce different objects for UDP, TCP and ICMP traceroutes?
• Should we try to normalise values when parsing? For instance, should we use ipaddr objects for addresses? Some times are expressed in centiseconds, some in microseconds, some in seconds. Should we normalize that to a common base? Are floats acceptable for time values?
• What should we do when there is a parsing error? How can the user continue parsing the next record if he/she wants to?

Please open issues if you have ideas and thoughts on these questions.