Build Status Coverage Status PyPI Version

Amazon's VPC Flow Logs are analogous to NetFlow and IPFIX logs, and can be used for security and performance analysis. Observable Networks uses VPC Flow logs as an input to endpoint modeling for security monitoring.

This project contains:

The tools support reading Flow Logs from both CloudWatch Logs and S3. For S3 destinations, version 3 custom log formats are supported.

The library builds on boto3 and should work on Python 3.4+.

For information on VPC Flow Logs and how to enable them see this post at the AWS blog. You may use this library with the kinesis-logs-reader library when retrieving VPC flow logs from Amazon Kinesis.


You can get flowlogs_reader by using pip:

pip install flowlogs_reader

Or if you want to install from source and/or contribute you can clone from GitHub:

git clone
cd flowlogs-reader
python develop

CLI Usage

flowlogs-reader provides a command line interface called flowlogs_reader that allows you to print VPC Flow Log records to your screen. It assumes your AWS credentials are available through environment variables, a boto configuration file, or through IAM metadata. Some example uses are below.

Location types

flowlogs_reader has one required argument, location. By default that is interpreted as a CloudWatch Logs group.

To use an S3 location, specify --location-type='s3':

Printing flows

The default action is to print flows. You may also specify the ipset, findip, and aggregate actions:

You may combine the output of flowlogs_reader with other command line utilities:

Time windows

The default time window is the last hour. You may also specify a --start-time and/or an --end-time. The -s and -e switches may be used also:

Use the --time-format switch to control how start and end times are interpreted. The default is '%Y-%m-%d %H:%M:%S'. See the Python documentation for strptime for information on format strings.

Concurrent reads

Give --thread-count to read from multiple log groups or S3 keys at once:

AWS options

Other command line switches:

For CloudWatch Logs locations:

For S3 locations:

Module Usage

FlowRecord takes an event dictionary retrieved from a log stream. It parses the message in the event, which takes a record like this:

2 123456789010 eni-102010ab 443 49152 6 10 840 1439387263 1439387264 ACCEPT OK

And turns it into a Python object like this:

>>> flow_record.srcaddr
>>> flow_record.dstaddr
>>> flow_record.srcport
>>> flow_record.to_dict()
{'account_id': '123456789010',
 'action': 'ACCEPT',
 'bytes': 840,
 'dstaddr': '',
 'dstport': 49152,
 'end': datetime.datetime(2015, 8, 12, 13, 47, 44),
 'interface_id': 'eni-102010ab',
 'log_status': 'OK',
 'packets': 10,
 'protocol': 6,
 'srcaddr': '',
 'srcport': 443,
 'start': datetime.datetime(2015, 8, 12, 13, 47, 43),
 'version': 2}

FlowLogsReader reads from CloudWatch Logs. It takes the name of a log group and can then yield all the Flow Log records from that group.

>>> from flowlogs_reader import FlowLogsReader
... flow_log_reader = FlowLogsReader('flowlog_group')
... records = list(flow_log_reader)
... print(len(records))

S3FlowLogsReader reads from S3. It takes a bucket name or a bucket/prefix identifier.

By default these classes will yield records from the last hour.

You can control what's retrieved with these parameters:

When using FlowLogsReader with CloudWatch Logs:

When using S3FlowLogsReader with S3:


Start by importing FlowLogsReader:

from flowlogs_reader import FlowLogsReader

Find all of the IP addresses communicating inside the VPC:

ip_set = set()
for record in FlowLogsReader('flowlog_group'):

See all of the traffic for one IP address:

target_ip = ''
records = []
for record in FlowLogsReader('flowlog_group'):
    if (record.srcaddr == target_ip) or (record.dstaddr == target_ip):

Loop through a few preconfigured profiles and collect all of the IP addresses:

ip_set = set()
profile_names = ['profile1', 'profile2']
for profile_name in profile_names:
    for record in FlowLogsReader('flowlog_group', profile_name=profile_name):

Apply a filter for UDP traffic that was logged normally (CloudWatch Logs only):

    '[version="2", account_id, interface_id, srcaddr, dstaddr, '
    'srcport, dstport, protocol="17", packets, bytes, '
    'start, end, action, log_status="OK"]'

flow_log_reader = FlowLogsReader('flowlog_group', filter_pattern=FILTER_PATTERN)
records = list(flow_log_reader)

Retrieve logs from a list of regions:

from flowlogs_reader import S3FlowLogsReader

reader = S3FlowLogsReader('example-bucket/optional-prefix', include_regions=['us-east-1', 'us-east-2'])
records = list(reader)

You may aggregate records with the aggregate_records function. Pass in a FlowLogsReader or S3FlowLogsReader object and optionally a key_fields tuple. Python dict objects will be yielded representing the aggregated flow records. By default the typical ('srcaddr', 'dstaddr', 'srcport', 'dstport', 'protocol') will be used. The start, end, packets, and bytes items will be aggregated.

flow_log_reader = FlowLogsReader('flowlog_group')
key_fields = ('srcaddr', 'dstaddr')
records = list(aggregated_records(flow_log_reader, key_fields=key_fields))