Man Page: rmr_call¶

RMR LIBRARY FUNCTIONS¶

NAME¶

rmr_call

SYNOPSIS¶

#include <rmr/rmr.h>

extern rmr_mbuf_t* rmr_call( void* vctx, rmr_mbuf_t* msg );

DESCRIPTION¶

The rmr_call function sends the user application message to a remote endpoint, and waits for a corresponding response message before returning control to the user application. The user application supplies a completed message buffer, as it would for a rmr_send call, but unlike with the send, the buffer returned will have the response from the application that received the message.

Messages which are received while waiting for the response are queued internally by RMR, and are returned to the user application when rmr_rcv_msg is invoked. These messages are returned in the order received, one per call to rmr_rcv_msg.

Call Timeout¶

The rmr_call function implements a timeout failsafe to prevent, in most cases, the function from blocking forever. The timeout period is not based on time (calls to clock are deemed too expensive for a low latency system level library), but instead the period is based on the number of received messages which are not the response. Using a mechanism which is not time based for timeout prevents the async queue from filling (which would lead to message drops) in an environment where there is heavy message traffic.

When the threshold number of messages have been queued without receiving a response message, control is returned to the user application and a nil pointer is returned to indicate that no message was received to process. Currently the threshold is fixed at 20 messages, though in future versions of the library this might be extended to be a parameter which the user application may set.

Retries¶

The send operations in RMR will retry soft send failures until one of three conditions occurs:

The message is sent without error
The underlying transport reports a hard failure
The maximum number of retry loops has been attempted

A retry loop consists of approximately 1000 send attempts without any intervening calls to sleep() or usleep(). The number of retry loops defaults to 1, thus a maximum of 1000 send attempts is performed before returning to the user application. This value can be set at any point after RMR initialisation using the rmr_set_stimeout() function allowing the user application to completely disable retires (set to 0), or to increase the number of retry loops.

Transport Level Blocking¶

The underlying transport mechanism used to send messages is configured in non-blocking mode. This means that if a message cannot be sent immediately the transport mechanism will not pause with the assumption that the inability to send will clear quickly (within a few milliseconds). This means that when the retry loop is completely disabled (set to 0), that the failure to accept a message for sending by the underlying mechanisms (software or hardware) will be reported immediately to the user application.

It should be noted that depending on the underlying transport mechanism being used, it is extremely likely that retry conditions will happen during normal operations. These are completely out of RMR’s control, and there is nothing that RMR can do to avoid or mitigate these other than by allowing RMR to retry the send operation, and even then it is possible (e.g., during connection reattempts), that a single retry loop is not enough to guarantee a successful send.

RETURN VALUE¶

The rmr_call function returns a pointer to a message buffer with the state set to reflect the overall state of call processing (see Errors below). In some cases a nil pointer will be returned; when this is the case only errno will be available to describe the reason for failure.

ERRORS¶

These values are reflected in the state field of the returned message.

RMR_OK

The call was successful and the message buffer references the response message.

RMR_ERR_CALLFAILED

The call failed and the value of errno, as described below, should be checked for the specific reason.

The global “variable” errno will be set to one of the following values if the overall call processing was not successful.

ETIMEDOUT

Too many messages were queued before receiving the expected response

ENOBUFS

The queued message ring is full, messages were dropped

EINVAL

A parameter was not valid

EAGAIN

The underlying message system was interrupted or the device was busy; the message was not sent, and the user application should call this function with the message again.

EXAMPLE¶

The following code snippet shows one way of using the rmr_call function, and illustrates how the transaction ID must be set.

int retries_left = 5;               // max retries on dev not available
int retry_delay = 50000;            // retry delay (usec)
static rmr_mbuf_t*  mbuf = NULL;    // response msg
msg_t*  pm;                         // application struct for payload

// get a send buffer and reference the payload
mbuf = rmr_alloc_msg( mr, sizeof( pm->req ) );
pm = (msg_t*) mbuf->payload;

// generate an xaction ID and fill in payload with data and msg type
snprintf( mbuf->xaction, RMR_MAX_XID, "%s", gen_xaction() );
snprintf( pm->req, sizeof( pm->req ), "{ \\"req\\": \\"num users\\"}" );
mbuf->mtype = MT_REQ;

msg = rmr_call( mr, msg );
if( ! msg ) {               // probably a timeout and no msg received
    return NULL;            // let errno trickle up
}

if( mbuf->state != RMR_OK ) {
    while( retries_left-- > 0 &&             // loop as long as eagain
           errno == EAGAIN &&
           (msg = rmr_call( mr, msg )) != NULL &&
           mbuf->state != RMR_OK ) {

        usleep( retry_delay );
    }

    if( mbuf == NULL || mbuf->state != RMR_OK ) {
        rmr_free_msg( mbuf );        // safe if nil
        return NULL;
    }
}

// do something with mbuf