I’ve got a strange situation with an extension on which I’m working.
When it is loaded, the extension starts up a long-lived object called
CommandRunner on a separate thread. As the extension does its work it
push objects into a queue that the CommandRunner then services
serially. The objects pushed into that queue each have their own file
descriptor (created with pipe()) that they use to communicate back to
the extension to say they’re done running. The extension, once it has
created the object and enqueues it, calls rb_thread_wait_fd() on that
file descriptor and, when the object sends the notify, SHOULD wake up
and continue executing.
This code has worked in past, when I had each of those objects create
their own thread of execution to run. But this was a painfully slow
process since creating a thread takes time. My goal is to have a
single thread (eventually perhaps a few parallel threads) to service
these objects. But, for whatever reason, right now everything seems to
run (I see the objects doing their work). But the code immediately
after rb_thread_wait_fd() never executes.
I did a simple test in IRB:
require ‘qpid’
=> trueworker = Thread.new do
?> conn = Qpid::Messaging::Connection.newconn.open
session = conn.create_session
sender = session.create_sender “my-queue;{create:always}”
receiver = session.create_receiver “my-queue”
outgoing = Qpid::Messaging::Message.new :content => “This is an outgoing
message. #{Time.new}”sender.send outgoing
incoming = receiver.get Qpid::Messaging::Duration::IMMEDIATE
puts “Incoming message: #{incoming.content}”
conn.close
end
What I found is that when I took the reference to this separate thread
(worker) and forced it to start, it picked up with the line after
rb_thread_wait_fd() and worked as expected.
Any suggestions on how to debug this problem?