Syslog and GELF relay to AWS Kinesis Firehose. Supports UDP, TCP, and TLS; RFC3164, RFC5424, RFC5425, RFC6587, GELF v1.1.

pip install kinesyslog==2.0.1



PyPI version Build Status

Syslog and GELF relay to AWS Kinesis Firehose. Supports UDP, TCP, and TLS; RFC3164, RFC5424, RFC5425, RFC6587, GELF v1.1.


This module requires Python 3.5 or better, due to its use of the asyncio framework.

This module contains optional functionality (Prometheus monitoring and process title rewriting) that require additional modules. When installing, specify kinesyslog[all] in order to include all available functionality.

This module uses Boto3 to make API calls against the Kinesis Firehose service. You should have a working AWS API environment (~/.aws/credentials, environment variables, or EC2 IAM Role) that allows calling Kinesis Firehose's put-record-batch method against the stream specified on the command line.

Release History

Release 2.0.0 makes several breaking changes from earlier versions:

  • Messages are passed from the main listener to the worker process via shared memory ringbuffer instead of unix sockets. This significantly increases performance, at the cost of a noticably larger memory footprint.
  • Message listeners no longer respond to HTTP requests for statistics. Use the Prometheus monitoring port instead.
  • Event IDs are now random digit strings instead of GUIDs. This better aligns with AWS, while improving performance.


Usage: kinesyslog listen [OPTIONS]

  --stream TEXT              Kinesis Firehose Delivery Stream Name. [required]
  --address TEXT             Bind address.  [default:]
  --prometheus-port INTEGER  Bind port for Prometheus statistics listener; 0 to disable. May be repeated.  [default: 0]
  --udp-port INTEGER         Bind port for UDP listener; 0 to disable. May be repeated.  [default: 0]
  --tcp-port INTEGER         Bind port for TCP listener; 0 to disable. May be repeated.  [default: 0]
  --tls-port INTEGER         Bind port for TLS listener; 0 to disable. May be repeated.  [default: 6514]
  --cert PATH                Certificate file for TLS listener.
  --key PATH                 Private key file for TLS listener.
  --proxy-protocol INTEGER   Enable PROXY protocol v1/v2 support on the selected TCP or TLS port; 0 to disable. May be repeated.  [default: 0]
  --spool-dir DIRECTORY      Spool directory for compressed records prior to upload.  [default: /tmp]
  --region TEXT              The region to use. Overrides config/env settings.
  --profile TEXT             Use a specific profile from your credential file.
  --gelf                     Listen for messages in Graylog Extended Log Format (GELF) instead of Syslog.
  --group-prefix TEXT        Use the specified LogGroup prefix.  [default: /kinesyslog]
  --debug                    Enable debug logging.
  --debug-asyncio            With --debug, enable debugging of asyncio. This significantly decreases performance.
  --help                     Show this message and exit.

Options can also be passed via environment variables. Repeated option values should be separated by whitespace.

export KINESYSLOG_UDP_PORT="12201"
export KINESYSLOG_TCP_PORT="12201 12202"
export KINESYSLOG_STREAM="kinesyslog-gelf-stream"

kinesyslog listen


Follow standard steps to install Python packages using pip. It is recommended that you use a virtualenv:

virtualenv /opt/kinesyslog
source /opt/kinesyslog/bin/activate
pip install --upgrade pip
pip install kinesyslog[all]

Once kinesyslog has been installed, you can generate and install a systemd unit. You will likely need to run the install command with sudo or as root, and provide the full path to the executable if it is in a virtualenv: sudo `which kinesyslog` install

Usage: kinesyslog install [OPTIONS]

  --user TEXT             Configure the service unit to run as specified user. If not specified, the service runs as root.
  --system-dir DIRECTORY  Install the service unit to the specified systemd system directory  [default: /etc/systemd/system]
  --help                  Show this message and exit.

You should run systemctl edit kinesyslog to adjust the service configuration variable specific to your environment - ports, certificates, streams, etc.

If not running on EC2, you will need to ensure that credentials are available to the service; the easiest way to do this is to install awscli and run aws configure as the user that the service will run as.

Record Format

When delivered to S3, the objects will contain multiple concatenated GZip-compressed records in JSON format. The records are near-identical to those created by CloudWatch Logs subscriptions with a few caveats:

  • The 'logGroup' and 'subscriptionFilter' fields are set to <PREFIX>/<FORMAT>/<PORT>, where:
    • <PREFIX> is the specified prefix; '/kinesyslog' by default.
    • <FORMAT> is the message format, either 'syslog' or 'gelf'.
    • <PORT> is the TCP or UDP port on which the message was received.
  • The 'logStream' field contains the IP address that the message was received from. Note that this is probably NOT the same as the 'source' field in the payload, since that's (hopefully) a FQDN.

Sample Record

   "owner" : "123456789012",
   "logGroup" : "/kinesyslog/syslog/514",
   "logStream" : "",
   "subscriptionFilters" : [
   "messageType" : "DATA_MESSAGE",
   "logEvents" : [
         "id" : "363031368935654431292225237233232307",
         "timestamp" : 1519247270240,
         "message" : "<13>1 2018-02-21T21:07:50.239881+00:00 user 4326 - [timeQuality tzKnown=\"1\" isSynced=\"0\"] Hello, World!"
         "id" : "123534373972454565292590045404680758",
         "timestamp" : 1519247271713,
         "message" : "<13>1 2018-02-21T21:07:51.712636+00:00 user 4327 - [timeQuality tzKnown=\"1\" isSynced=\"0\"] I, for one, welcome our new insect overlords"

Data Flow

Syslog events are received via UDP, TCP, or TLS and parsed in the main listener process. When a message boundary is found, the reassembled message is immediately handed off to a background worker process via shared memory ring buffer, freeing the main process to continue accepting messages.

The background worker extracts event timestamps, assigns events a unique ID, and stores them into a buffer, grouped by source address and destination port. When this buffer reaches a preset threshold of size (4 MB) or message age (60 seconds), the messages are serialized into a GZip-compressed JSON data structure, which is written out to the on-disk message spool. 4MB has been found to frequently compress down to near 1000 KB, which is the maximum record size for Kinesis Firehose. Buffers will be split into multiple records (along message boundaries) if the compressed size exceeds the maximum Firehose record size.

Every 60 seconds, another background worker flushes spooled records to Firehose in batches that do not exceed 4 MB or 500 records. Spooled records are removed from disk only when Firehose acknowledges successful upload by assigning a Record ID.


A HTTP listener serving Prometheus statistics can be enabled with the --prometheus-port option. This endpoint is also suitable for monitoring by load-balancer health checks.

# HELP kinesyslog_batch_bytes Kinesis batch request size
# TYPE kinesyslog_batch_bytes histogram
# HELP kinesyslog_batch_record_failed Kinesis batch record failures
# TYPE kinesyslog_batch_record_failed counter
# HELP kinesyslog_batch_records Kinesis batch record count
# TYPE kinesyslog_batch_records histogram
# HELP kinesyslog_http_requests_total Total HTTP requests
# TYPE kinesyslog_http_requests_total counter
# HELP kinesyslog_listener_count The number of message listeners
# TYPE kinesyslog_listener_count gauge
# HELP kinesyslog_message_bytes_total Message bytes received
# TYPE kinesyslog_message_bytes_total counter
# HELP kinesyslog_message_count_total Message records received
# TYPE kinesyslog_message_count_total counter
# HELP kinesyslog_record_bytes Kinesis record size
# TYPE kinesyslog_record_bytes histogram
# HELP kinesyslog_spool_age Kinesis batch spool record age
# TYPE kinesyslog_spool_age gauge
# HELP kinesyslog_spool_count Kinesis batch spool record count
# TYPE kinesyslog_spool_count gauge

Vendored Modules

This module include a ctypes-based ringbuffer module that is not available on PyPI.


  • Client certificate validation?