Thrift binary protocol

Revision as of 05:24, 20 May 2019

Apache Thrift, a serialization/RPC system with much more emphasis on RPC than serialization, calls its serialization formats "protocols"; the oldest Thrift-specific protocol is the "binary" protocol, while the newer one is the "compact" protocol.^[1]

Identification

In the new^[2] ("strict") header type, binary protocol files start with a one-byte message type, a zero byte, and then the version indicator 0x01 0x00 0x00 0x08.^[3]. (This version indicator has been steady since at least as far back as 2011.) This is followed by an encoded string (a 32-bit length and series of bytes) and a 32-bit sequence ID.^[4]

In the old^[2] (non-"strict") header type, the message begins with the encoded string of the message name, followed by the type byte, followed by a 32-bit sequence ID.^[5]

Links

• Thrift website
• The original Thrift description
• A brief description of protocols

References

1. ↑ https://thrift.apache.org/docs/concepts#protocol
2. ↑ ^{2.0 2.1} Commit 589859984f43af1ed0f9ee45ce8374a0d1640813
3. ↑ Thrift source → /lib/java/src/org/apache/thrift/protocol/TBinaryProtocol.java → definition of VERSION_MASK and VERSION_1, and of writeMessageBegin(). https://stackoverflow.com/questions/8402479/error-while-compiling-thrift-server indicates that the version/size integer is presumably stored little-endian.
4. ↑ Thrift source → /lib/java/src/org/apache/thrift/protocol/TBinaryProtocol.java → readMessageBegin(); /lib/java/src/org/apache/thrift/protocol/TMessage.java
5. ↑ Thrift source → /lib/java/src/org/apache/thrift/protocol/TBinaryProtocol.java → readMessageBegin()