The State Engine of lighttpd¶
- Table of contents
- The State Engine of lighttpd
The state-engine is currently made of 11 states which are walk-through for each connection. Some of them are specific for a special operation
and some may never be hit at all.
waiting for a connection
init the read-idle timer
read http-request-header from network
read http-request-content from network
handle the request internally (might result in sub-requests)
prepare response header
write response-header + content to network
cleanup environment, log request
reset connection (incl. close())
close connection (handle lingering close)
A simple GET request (green path)¶
The connection is idling in the 'connect' state waiting for a connection.
As soon as the connection is set up we init the read-timer in 'reqstart'
and start to read data from the network. As soon as we get the
HTTP-request terminator (CRLFCRLF) we forward the header to the parser.
The parsed request is handled by 'handlereq' and as soon as a decision about
the request is made it is sent to 'respstart' to prepare the
HTTP-response header. In the 'write' state the prepare content is sent out
to the network. When everything is sent 'respend' is entered to log the
request and cleanup the environment. After the close() call the connection
is set back to the 'connect' state again.
Keep-Alive (blue path)¶
The Keep-Alive handling is implemented by going from the 'respend'
directly to 'reqstart' without the close() and the accept() calls.
POST requests (grey path)¶
As requests might contain a request-body the state 'readpost' entered as
soon as the header is parsed and we know how much data we expect.
HTTP/1.1 supportes pipelining (sending multiple requests without waiting
for the response of the first request). This is handled transparently by
the 'read' state.
Unexpected errors (red path)¶
For really hard errors we use the 'error' state which resets the
connection and can be call from every state. It is only use if there is no
other way to handle the issue (e.g. client-side close of the connection).
If possible we should use http-status 500 ('internal server error') and
log the issue in the errorlog.
If we have to take care of some data which is coming in after we ran into
the error condition the 'close' state is used the init a half-close and
read all the delay packet from the network.
The FastCGI, CGI, ... integration is done by introducing a loop in
'handlereq' to handle all aspect which are neccesary to find out what has
to be sent back to the client.
Important functions used by the state-engine
connection_close()(if not keep-alive)
Updated by bwechner over 2 years ago · 12 revisions