Finally :) I managed to integrate the proactor in my program loop by using pn_proactor_get carefully correlated with pn_proactor_done.
The main mistake I did, was that I used to call pn_proactor_done after getting a flow event and just before calling send. Many thanks for advise and for patience. Adrian On Thu, Jun 18, 2020, 7:50 AM Adrian Florea <florea....@gmail.com> wrote: > OK, so it must be pn_proactor_wait() or can I achieve similar behavior > with pn_proactor_get() ? I assume not. > > It would be nice if pn_proactor_wait would have a timed_wait option. > > > > On Thu, Jun 18, 2020, 7:44 AM Ted Ross <tr...@redhat.com> wrote: > >> If you look at the examples supplied with Proton, you will see simple >> applications that behave as you desire. Sends are immediate. >> >> Changing your idle timeout is only altering the timing of the bad behavior >> of your app. You need to find a way to incorporate pn_proactor_wait into >> your logic. >> >> On Thu, Jun 18, 2020 at 1:07 AM Adrian Florea <florea....@gmail.com> >> wrote: >> >> > So, based on this email chain and looking at what the idle timeout is >> > intended for, I think that is true ... proton is "woke up" by these >> > heartbeats, like you said. Playing with transport timeout values, just >> > increased their frequency. >> > >> > I will look at other possibilities to obtain an "immediate send" effect. >> > >> > >> > On Wed, Jun 17, 2020, 3:26 PM Adrian Florea <florea....@gmail.com> >> wrote: >> > >> > > Some news. >> > > >> > > After setting up the transport (SSL and all), I added a call to >> > > pn_transport_set_idle_timeout, with 20000ms. >> > > >> > > This provides great improvement, as now I can see my messages going >> out >> > > every few seconds, definitely sooner than 20s. >> > > >> > > As a side note, I tried to set the timeout to a subsecond value, >> doesn't >> > > work. >> > > Said it must be min 10000. Setting it to 10000 is causing a subsequent >> > > error with the connection timeout. The connection timeout becomes 5000 >> > ... >> > > so I ended up setting transport timeout to 20000 to achieve a >> cinnection >> > > timeout of 10000. >> > > >> > > As I said, this provides great improvement but it would be nice if the >> > > send can be "flushed" immediately. >> > > >> > > Adrian >> > > >> > > On Wed, Jun 17, 2020, 2:40 PM Ted Ross <tr...@redhat.com> wrote: >> > > >> > >> Proactor is a single-threaded, event-driven API for messaging. It >> owns >> > >> the >> > >> main execution loop and uses the pn_proactor_wait() execution to do >> > >> background work like sending your message out the connection. >> > >> >> > >> I don't know what your application looks like, but I assume that you >> > have >> > >> your own main loop and you don't ever give proactor a chance to run. >> > Your >> > >> message is probably being sent when a heartbeat frame arrives from >> > >> whatever >> > >> you're connected to. This is the PN_TRANSPORT event you are seeing. >> > >> >> > >> -Ted >> > >> >> > >> On Wed, Jun 17, 2020 at 3:00 PM Adrian Florea <florea....@gmail.com> >> > >> wrote: >> > >> >> > >> > Yeah... forget my last mention. Looking at what pn_proactor_done >> does, >> > >> it >> > >> > doesn't make sense to call it when the batch of events is null. >> > >> > >> > >> > On Wed, Jun 17, 2020, 1:50 PM Adrian Florea <florea....@gmail.com> >> > >> wrote: >> > >> > >> > >> > > Yes. >> > >> > > I don't call it when the pn_proactor_get() returns null. >> > >> > > >> > >> > > I should probably call it in this case as well.. >> > >> > > >> > >> > > >> > >> > > On Wed, Jun 17, 2020, 1:30 PM Ted Ross <tr...@redhat.com> wrote: >> > >> > > >> > >> > >> On Wed, Jun 17, 2020 at 2:19 PM Adrian Florea < >> > florea....@gmail.com> >> > >> > >> wrote: >> > >> > >> >> > >> > >> > Hi, thanks. >> > >> > >> > I am using the proactor. >> > >> > >> > I need a way to clearly send a message out. >> > >> > >> > My program has a loop and everytime it loops, I tried this: >> > >> > >> > >> > >> > >> > - call pn_proactor_wait --> this ends up blocking my loop, >> which >> > >> is >> > >> > not >> > >> > >> > good. >> > >> > >> > >> > >> > >> > - call pn_proactor_get -- this does not block and returns no >> > event >> > >> > for a >> > >> > >> > long while, when suddenly it gets a PN_TRANSPORT event and >> all my >> > >> > >> messages >> > >> > >> > are really sent out. >> > >> > >> > >> > >> > >> >> > >> > >> Are you calling pn_proactor_done() after processing the batch of >> > >> events >> > >> > >> from pn_proactor_get()? >> > >> > >> >> > >> > >> >> > >> > >> > >> > >> > >> > Adrian >> > >> > >> > >> > >> > >> > On Wed, Jun 17, 2020, 12:36 PM Ted Ross <tr...@redhat.com> >> > wrote: >> > >> > >> > >> > >> > >> > > Hi Adrian, >> > >> > >> > > >> > >> > >> > > What is your program doing after it calls pn_message_send? >> > That >> > >> > >> function >> > >> > >> > > queues the message for delivery but the delivery isn't >> actually >> > >> > >> > transferred >> > >> > >> > > until the application yields the control back to the Proton >> > >> reactor >> > >> > >> (via >> > >> > >> > > pn_proactor_wait). If the application is doing other >> > processing >> > >> or >> > >> > >> > waiting >> > >> > >> > > on a condition or mutex, the delivery won't go out the door >> > >> > >> immediately. >> > >> > >> > > >> > >> > >> > > -Ted >> > >> > >> > > >> > >> > >> > > On Wed, Jun 17, 2020 at 1:11 PM Adrian Florea < >> > >> florea....@gmail.com >> > >> > > >> > >> > >> > > wrote: >> > >> > >> > > >> > >> > >> > > > Hi, >> > >> > >> > > > >> > >> > >> > > > Any idea is welcome on this one. >> > >> > >> > > > >> > >> > >> > > > I am trying to send messages (via a sender link) at >> various >> > >> > moments >> > >> > >> in >> > >> > >> > > the >> > >> > >> > > > life of a program. I am using pn_message_send. >> > >> > >> > > > >> > >> > >> > > > I have set the outgoing window size to 1, on the session. >> > >> > >> > > > >> > >> > >> > > > The current behavior is: >> > >> > >> > > > >> > >> > >> > > > 1. pn_message_send completes OK >> > >> > >> > > > 2. nothing is actually sent >> > >> > >> > > > 3. after a while (I guess this is where I miss something) >> I >> > see >> > >> > that >> > >> > >> > the >> > >> > >> > > > proactor gets an event of type PN_TRANSPORT and I can see >> all >> > >> > >> messages >> > >> > >> > > > being really sent. >> > >> > >> > > > >> > >> > >> > > > Is there a way to achieve a "send immediate" behavior ? >> > >> > >> > > > >> > >> > >> > > > When a message send is invoked, I need it to really go >> out. >> > >> > >> > > > >> > >> > >> > > > many thanks for pointing me in the right direction, >> > >> > >> > > > >> > >> > >> > > > Adrian >> > >> > >> > > > >> > >> > >> > > >> > >> > >> > >> > >> > >> >> > >> > > >> > >> > >> > >> >> > > >> > >> >
