人生如戏，全凭演技

• KOALA_INBOUND_ADDR: ip:port, the address inbound server will bind to
• KOALA_SUT_ADDR: ip:port, the address inbound will call
• KOALA_OUTBOUND_ADDR: ip:port, the address all outgoing traffic will be redirected to
• KOALA_LOG_FILE: STDOUT/STDERR/filepath, if using filepath, the log will rotate every hour
• KOALA_LOG_LEVEL: TRACE/DEBUG/INFO/ERROR/FATAL

• koala_go: for go application compiled with koala-go
• koala_replayer: enable replaying mode
• koala_recorder: enable recording mode

koala_replayer and koala_recorder can be enabled at the same time, to benchmark recording with replaying.

• intercept tcp send/recv
• associate send/recv data to same thread id as "session"
• request => response => request, so we can know when a "talk" (with request/response pair) is complete
• use udp 127.127.127.127:127 to inform recorder with helper information.
• for "system under test" using thread multiplexing (one thread doing more than one thing), map real thread id to virtual thread id by helper information.

replaying builds on same mechanism, but much more complex

• "system under test" is a process, "replayer inbound server" and "replayer outbound server" lives in same process. They are two tcp servers started by the .so loaded via LD_PRELOAD.
• session to replay is injected into the process via "replayer inbound server" tcp connection
• "replayer inbound server" call the "system under test" via tcp connection, store the "session id <=> inbound socket" mapping.
• "system under test" call external dependencies, which is intercepted to "replayer outbound server", store the "inbound socket <=> outbound socket" mapping
• "replayer outbound server" use its own socket to lookup the mapping, to find which session to replay

koala support two modes by different gateways (https://github.com/v2pro/koala/tree/master/gateway)

• libc: for application built with libc (c/c++/python/java)
• go: need to compile with https://github.com/v2pro/koala-go

• Long Connection: reusing connection sequentially is not a issue, just update the mapping
• Multiplexing: one thread handing multiple business processes at the same time, need "helper information"
• One Way Communication: request without response, need "helper information" to cut two requests out
• Greeting: protocol like mysql send greeting before request. use the ip:port or just guess, to decide if greeting is needed.