chenbk85 / reactor-8 Goto Github PK

A reactor/demultiplexer event micro-framework for Linux in C

• event demux for timers, signals, sockets, stdin/stdout, serial etc
• support for application user-contexts in the callback handler
• specify timeout for any io action
• synchronous management of signal interupts
• clean cancel/terminate semantics for easy cleanup of resources
• no hiding/abstracting ioctl(), tty_ioctl() etc
• no queue lag for simpler lifetime management

  $ cd ~/reactor
  $ ./build.sh
  $ ./examples/timeout.out
  ...

• https://en.wikipedia.org/wiki/Reactor_pattern
• Refs proactor versus reactor - http://www.artima.com/articles/io_design_patterns2.html
  • Reactor
    • dmux and dispatch to handler on read or write readiness
    • dmux dispatching always synchronous
  • Proactor
    • additionally handle the read or write asynchronously in kernel and then dispatch with data to handler
    • strands, apartments thread contexts...
• For the most part a proactor can be built around a reactor
• http://gngrwzrd.com/libgwrl/pod.html

• parser for config /etc/resolve?. just a simple closure. with regex. and or etc.

• handle a single queue - based on a pipe - inside

  • handlers would have a predicate taking the messgae and returning whether they can handle the message
  • encode len, so that can always extract entire message, then pass to handler.
  • handler then co-exists with timeout
  • use for signals, - and would enable per signal handlers
  • also thread pool

  // No the way to do it, is to have another module entire. // with one reader, that dequeues the message, and always rebinds itself // and that maintains it's own handler set. // that it delegates to (perhaps with predicates). // this keeps it entirely out of the demux core.

• non-blocking dns resolution on a consumer thread.

  • may be able to do with fifo queue similar to interupts.

  • http://discuss.joelonsoftware.com/default.asp?joel.3.475011.13

  • getaddrinfo_a() is async

  • http://stackoverflow.com/questions/22013421/non-blocking-network-address-resolution-gethostbyname-or-getaddrinfo

  • pool.c - push job onto pool with context and callback. then write pipe, with callback handler. POSIX.1 says that write(2)s of less than PIPE_BUF bytes must be atomic: the output data is written to the pipe as a contiguous sequence. Writes of more than PIPE_BUF bytes may be nonatomic: the kernel may interleave the data with data written by other processes. POSIX.1 requires PIPE_BUF to be at least 512 bytes. (On Linux, PIPE_BUF is 4096 bytes.)

• maybe rename Event to ReactorEvent or use reactor_event_t etc... becomes more complicated

• maybe rename to use lowercase underscore t

• maybe support cancelling individual handlers (lookup by fd and type)

  • or just return the handler for user to call action on?

• investigate if ok to mix fopen(stdout) with open(1) for logging... (better to use stderr?)

• signal deregistration

• fix device echo issue with serial comms

• investigate - for proactor ioctl for sockets to discover number bytes in buffer already/yet to transmit. similar to boost asio

• network examples

  • websockets - would be really good. could then create a shell.
  • postgres async connection
  • http
  • dns async - server/forwarder and client
  • telnet socket bind/listen

• better enum variable prefixes - eg. prefix with 'reactor'

• done - move signal fifo handling outside pure reactor, also logging. for modularity. delegate
• no - rename d-> vars to r->
• no - maybe rename ureactor or udemux?
• done - combine the read/writ callback, then type the event we're interested in
  • will simplify, and make rebind(e) possible.
  • simultaneous read and write handler binding
  • READ_READY, WRITE_READY, EXCEPTION, TIMEOUT, CANCELLED
• done - issue with binding both read/write is may want to cancel one in response to the other?
  • reason to encode the action... rather
  • makes cancelling, as easy as not rebinding
• done - example simple serial comms
  • ioctl tty_ioctl set baud, rts, dtr
• done - get bind() and listen() and check that can spawn new contexts easily
• done - fix cancel_all
  • make it easy to do a cancel/shutdown from any callback
• done - investage issue when using serial-comm and the double \n\n
• done - use gettimeofday() instead of time for millisec prec.
• done - compute smallest timeout value across all handlers and use as select() sleep timeout
• done - change dispatcher name to reactor or demux? or separate demux and handler dispatch?
• done - signals using fifo (maybe outside the core dispatcher)
• done - move tests into examples...
• done - control over debug stream
• done - pure timers not associated with fd (outside core dispatcher if can? ).
  • eg. read stdin -> write to /dev/ttyUSB0, read /dev/ttyUSB write stdout,
• done - timer / timeouts
• done - cancel all, and call all callbacks
• done - write handler

• /blog - task esp8266, and control registers. screen, minicom, miniterm that I use this led, to writing a demux . handling of signals.