Python producer for AWS Kinesis Stream with record aggregation.
It uses Boto 3 and is tested on Python 2.7 and 3.4/3.5.
pip install 'kinesis_producer < 1'
Note: Kinesis Producer use semver: you should always freeze on the major version since it could mean breaking the API.
Send records aggregated up to 100KB, 200ms and joined with '\n':
from kinesis_producer import KinesisProducer
config = dict(
aws_region='us-east-1',
buffer_size_limit=100000,
buffer_time_limit=0.2,
kinesis_concurrency=1,
kinesis_max_retries=10,
record_delimiter='\n',
stream_name='KINESIS_STREAM_NAME',
)
k = KinesisProducer(config=config)
for record in records:
k.send(record)
k.close()
k.join()| aws_region: | AWS region for Kinesis calls (like us-east-1) |
|---|---|
| buffer_size_limit: | Approximative size limit for record aggregation (in bytes) |
| buffer_time_limit: | Approximative time limit for record aggregation (in seconds) |
| kinesis_concurrency: | Set the concurrency level for Kinesis calls. Set to 1 for no concurrency. Set to 2 and more to use a thread pool. |
| kinesis_max_retries: | Number of Kinesis put_records call attempt before giving up. This number should be between 4 and 10 if you want to handle temporary ProvisionedThroughputExceeded errors. |
| record_delimiter: | Delimiter for record aggregation |
| stream_name: | Name of the Kinesis Stream |
Kinesis calls are retried for ProvisionedThroughputExceeded error only. Retry use an exponential backoff logic (0.1s, 0.2s, 0.4s, 0.8s, 1.60s, 3.20s, 6.40s, 12.80s, 25.60s, 51.20s, 102.40s...)
Released under the MIT license.