Mqueue problem

Jon_Mills · September 20, 2000, 4:11pm

I am currently working on some code using POSIX style queues, and have
two processes, a client writing into the queue, and a server reading
from it. If I run these processes on the node I am logged into, this
works fine. If however I run one or both of the processes on another
node (e.g. “on -n4 RepServer”), the sender sends the data to the queue
(or seems to think it does), but the server does not see the data
arriving in the queue. I am using a proxy to signal the server when the
queue goes from empty to non-empty.

It appears that the queue is always being created on Node 1, which is
the node I am logged into. If I use the command “ls -l /dev/mqueue” from
Node 1, the queue can always be seen regardless of which node the
processes are running on. The command “ls -l //4/dev/mqueue” results in
no queue being seen. I have tried using the setenv() function to set the
environment variable MQ_OPEN to a value of 1 in both client and server,
but this does not help. Is this the correct use of MQ_OPEN? Should I set
it to a number (e.g. 4) or a node name?

Any ideas?

Best Regards, Jon Mills.

Barry_Millman · September 21, 2000, 1:23pm

Jon:

I don’t have my docs with me, but you got trapped into the problem of
software for the non-networked UNIX world being ported to QNX.

The solution to your problem is this…before you access a queue on a
remote node you must set an ENVIRONMENT VARIABLE which I believe is called
MQ_NODE (what a pain in the foot!). It would be nice (but currently is
illegal) if we could specify a node number when we do an mq_open().

Shameless advertisement: This is beautifully discussed (with examples
and hands-on exercises) in our Real-Time QNX Programming Computer-Based
Training Course…take a look at:
http://www.igetitnow.com

Barry

QNX Training at Your Desktop…with Your Success Guaranteed!
http://www.igetitnow.com

Jon Mills wrote:

I am currently working on some code using POSIX style queues, and have
two processes, a client writing into the queue, and a server reading
from it. If I run these processes on the node I am logged into, this
works fine. If however I run one or both of the processes on another
node (e.g. “on -n4 RepServer”), the sender sends the data to the queue
(or seems to think it does), but the server does not see the data
arriving in the queue. I am using a proxy to signal the server when the
queue goes from empty to non-empty.

It appears that the queue is always being created on Node 1, which is
the node I am logged into. If I use the command “ls -l /dev/mqueue” from
Node 1, the queue can always be seen regardless of which node the
processes are running on. The command “ls -l //4/dev/mqueue” results in
no queue being seen. I have tried using the setenv() function to set the
environment variable MQ_OPEN to a value of 1 in both client and server,
but this does not help. Is this the correct use of MQ_OPEN? Should I set
it to a number (e.g. 4) or a node name?

Any ideas?

Best Regards, Jon Mills.

–

Ken_Recchia · September 21, 2000, 3:24pm

Make sure you are running Mqueue 4.24A and not 4.24B.
The environment variable MQ_NODE will specify the node for mqueue to
use. Mqueue does not need to run on node2 if you are using MQ_NODE=1.

Jon Mills wrote:

I am currently working on some code using POSIX style queues, and have
two processes, a client writing into the queue, and a server reading
from it. If I run these processes on the node I am logged into, this
works fine. If however I run one or both of the processes on another
node (e.g. “on -n4 RepServer”), the sender sends the data to the queue
(or seems to think it does), but the server does not see the data
arriving in the queue. I am using a proxy to signal the server when the
queue goes from empty to non-empty.

It appears that the queue is always being created on Node 1, which is
the node I am logged into. If I use the command “ls -l /dev/mqueue” from
Node 1, the queue can always be seen regardless of which node the
processes are running on. The command “ls -l //4/dev/mqueue” results in
no queue being seen. I have tried using the setenv() function to set the
environment variable MQ_OPEN to a value of 1 in both client and server,
but this does not help. Is this the correct use of MQ_OPEN? Should I set
it to a number (e.g. 4) or a node name?

Any ideas?

Best Regards, Jon Mills.

Jon_Mills · September 22, 2000, 9:11am

Barry, Ken,

After further investigation, it seems that when we set MQ_NODE to 1, the
data is going to the queue on node 1. Unfortunately, we use a proxy to
generate a signal to the server when the queue becomes non-empty. This
signal does not appear to be being trasmitted across the network. I
guess we need some kind of virtual circuit. Is it more common to poll
for the queue being non-empty than to use a signal?

Thanks for you help.

Jon.

P.S. Ken, you mention that we should use 4.25A rather that 4.24B, but we
now appear to be using MQ_NODE and 4.24B without problems. What is the
problem with 4.24B?

-----Original Message-----
From: Barry Millman [mailto:bmillman@entrenet.com]
Posted At: 21 September 2000 14:24
Posted To: qnx
Conversation: Mqueue problem
Subject: Re: Mqueue problem

Jon:

I don’t have my docs with me, but you got trapped into the problem
of
software for the non-networked UNIX world being ported to QNX.

The solution to your problem is this…before you access a queue on
a
remote node you must set an ENVIRONMENT VARIABLE which I believe is
called
MQ_NODE (what a pain in the foot!). It would be nice (but currently is
illegal) if we could specify a node number when we do an mq_open().

Shameless advertisement: This is beautifully discussed (with
examples
and hands-on exercises) in our Real-Time QNX Programming Computer-Based
Training Course…take a look at:
http://www.igetitnow.com

Barry

QNX Training at Your Desktop…with Your Success Guaranteed!
http://www.igetitnow.com

Jon Mills wrote:

I am currently working on some code using POSIX style queues, and have
two processes, a client writing into the queue, and a server reading
from it. If I run these processes on the node I am logged into, this
works fine. If however I run one or both of the processes on another
node (e.g. “on -n4 RepServer”), the sender sends the data to the queue
(or seems to think it does), but the server does not see the data
arriving in the queue. I am using a proxy to signal the server when
the
queue goes from empty to non-empty.

It appears that the queue is always being created on Node 1, which is
the node I am logged into. If I use the command “ls -l /dev/mqueue”
from
Node 1, the queue can always be seen regardless of which node the
processes are running on. The command “ls -l //4/dev/mqueue” results
in
no queue being seen. I have tried using the setenv() function to set
the
environment variable MQ_OPEN to a value of 1 in both client and
server,
but this does not help. Is this the correct use of MQ_OPEN? Should I
set
it to a number (e.g. 4) or a node name?

Any ideas?

Best Regards, Jon Mills.

–

Barry_Millman · September 22, 2000, 1:21pm

Jon:

Yes, you need to create a virtual proxy. The QNX docs for
qnx_proxy_rem_attach() (used AFTER a qnx_proxy_attach() ) will create a
remote proxy valid on the node where the queue manager is running (you give
that node as a parm). When you ask the queue manager to inform of data in
the queue, the program should pass the remote proxy id to the queue manager.

Barry

QNX Training at Your Desktop…with Your Success Guaranteed!
http://www.igetitnow.com

Jon Mills wrote:

Barry, Ken,

After further investigation, it seems that when we set MQ_NODE to 1, the
data is going to the queue on node 1. Unfortunately, we use a proxy to
generate a signal to the server when the queue becomes non-empty. This
signal does not appear to be being trasmitted across the network. I
guess we need some kind of virtual circuit. Is it more common to poll
for the queue being non-empty than to use a signal?

Thanks for you help.

Jon.

P.S. Ken, you mention that we should use 4.25A rather that 4.24B, but we
now appear to be using MQ_NODE and 4.24B without problems. What is the
problem with 4.24B?

-----Original Message-----
From: Barry Millman [mailto:> bmillman@entrenet.com> ]
Posted At: 21 September 2000 14:24
Posted To: qnx
Conversation: Mqueue problem
Subject: Re: Mqueue problem

Jon:

I don’t have my docs with me, but you got trapped into the problem
of
software for the non-networked UNIX world being ported to QNX.

The solution to your problem is this…before you access a queue on
a
remote node you must set an ENVIRONMENT VARIABLE which I believe is
called
MQ_NODE (what a pain in the foot!). It would be nice (but currently is
illegal) if we could specify a node number when we do an mq_open().

Shameless advertisement: This is beautifully discussed (with
examples
and hands-on exercises) in our Real-Time QNX Programming Computer-Based
Training Course…take a look at:
http://www.igetitnow.com

Barry

Jon Mills wrote:

I am currently working on some code using POSIX style queues, and have
two processes, a client writing into the queue, and a server reading
from it. If I run these processes on the node I am logged into, this
works fine. If however I run one or both of the processes on another
node (e.g. “on -n4 RepServer”), the sender sends the data to the queue
(or seems to think it does), but the server does not see the data
arriving in the queue. I am using a proxy to signal the server when
the
queue goes from empty to non-empty.

It appears that the queue is always being created on Node 1, which is
the node I am logged into. If I use the command “ls -l /dev/mqueue”
from
Node 1, the queue can always be seen regardless of which node the
processes are running on. The command “ls -l //4/dev/mqueue” results
in
no queue being seen. I have tried using the setenv() function to set
the
environment variable MQ_OPEN to a value of 1 in both client and
server,
but this does not help. Is this the correct use of MQ_OPEN? Should I
set
it to a number (e.g. 4) or a node name?

Any ideas?

Best Regards, Jon Mills.

–

–
QNX Training at Your Desktop…with Your Success Guaranteed!
http://www.igetitnow.com

Ken_Recchia · September 22, 2000, 4:10pm

While using 4.24B, I had mysterious corrupted data and missing messages that
disappeared when I used 4.24A. Also, QNX clarified that 4.24A is the latest
version. (There was a small mistake with the versioning.)

MQ_NODE =1 will use Mqueue on node 1 and all mqueue’s will be created on
node 1. Also in this configuration you will not need to run Mqueue on any
other node. I wouldn’t poll the mqueue but you can see how many messages
are in it with mq_getattr().

I don’t use proxies so I don’t know how to advise about them - yet.

Do you need to handle the case where the client is on a different node than
the server? or will they always be on the same node? If they will always be
on the same node then set MQ_NODE=1 on node 1 and MQ_NODE=2 on node 2 and
see what happens (make sure Mqueue is running on each node).

Jon Mills wrote:

Barry, Ken,

After further investigation, it seems that when we set MQ_NODE to 1, the
data is going to the queue on node 1. Unfortunately, we use a proxy to
generate a signal to the server when the queue becomes non-empty. This
signal does not appear to be being trasmitted across the network. I
guess we need some kind of virtual circuit. Is it more common to poll
for the queue being non-empty than to use a signal?

Thanks for you help.

Jon.

P.S. Ken, you mention that we should use 4.25A rather that 4.24B, but we
now appear to be using MQ_NODE and 4.24B without problems. What is the
problem with 4.24B?

-----Original Message-----
From: Barry Millman [mailto:> bmillman@entrenet.com> ]
Posted At: 21 September 2000 14:24
Posted To: qnx
Conversation: Mqueue problem
Subject: Re: Mqueue problem

Jon:

I don’t have my docs with me, but you got trapped into the problem
of
software for the non-networked UNIX world being ported to QNX.

The solution to your problem is this…before you access a queue on
a
remote node you must set an ENVIRONMENT VARIABLE which I believe is
called
MQ_NODE (what a pain in the foot!). It would be nice (but currently is
illegal) if we could specify a node number when we do an mq_open().

Shameless advertisement: This is beautifully discussed (with
examples
and hands-on exercises) in our Real-Time QNX Programming Computer-Based
Training Course…take a look at:
http://www.igetitnow.com

Barry

QNX Training at Your Desktop…with Your Success Guaranteed!
http://www.igetitnow.com

Jon Mills wrote:

I am currently working on some code using POSIX style queues, and have
two processes, a client writing into the queue, and a server reading
from it. If I run these processes on the node I am logged into, this
works fine. If however I run one or both of the processes on another
node (e.g. “on -n4 RepServer”), the sender sends the data to the queue
(or seems to think it does), but the server does not see the data
arriving in the queue. I am using a proxy to signal the server when
the
queue goes from empty to non-empty.

It appears that the queue is always being created on Node 1, which is
the node I am logged into. If I use the command “ls -l /dev/mqueue”
from
Node 1, the queue can always be seen regardless of which node the
processes are running on. The command “ls -l //4/dev/mqueue” results
in
no queue being seen. I have tried using the setenv() function to set
the
environment variable MQ_OPEN to a value of 1 in both client and
server,
but this does not help. Is this the correct use of MQ_OPEN? Should I
set
it to a number (e.g. 4) or a node name?

Any ideas?

Best Regards, Jon Mills.

–

Davidfmarker · September 22, 2000, 5:47pm

I believe there is an environment variable MQ_NODE to specify what node the
Mqueue server is on. I have used Mqueue across many systems without a
problems.

Regards.

Dave Marker