Serialization with Avro without schema

Hi,

The scenario is a machine dynamically generates data, which consists
sections of binary data. We use Flume SDK to collect data and the sink is
HDFS(SequenceFile).

I'm curious what is in the sequence file, since Flume is unaware of schema.
i.e., How does Flume and Avro do serialization without schema? ( Directly
writing raw bytes to disk file may cause alignment issue).
http://stackoverflow.com/questions/18001818/avro-schema-storage is similar
to my question.

Also, how the key is determined in the sequence file? If my understanding
is not correct, please indicate correct usage of Flume with Avro.

Thank you for your clarification.

Cheers,
Blade