Jump to content

[LVTN] Messenger Library


Recommended Posts

Your right, there is a reference leak here.  That Queue is a reference created in the actor's calling code, and is used to trigger the "Autoshutdown" message when the caller stops, thus invalidating the Queue.  There is an async component called the "ReferenceMonitor" which watches that Queue and sends the Autoshutdown message then closes.   ReferenceMonitor also polls the actor's address, so it can shut itself down if the actor dies.  However, I forgot to destroy the Queue in that case.   Here is an image where I have added the needed Close operation:

2146204797_FixReferenceLeak.png.3fb51efbe10eae86af6c78643cb176a8.png

 

Thank for spotting this,

-- James

Link to comment
  • 4 weeks later...

Hi Dr James,

Recently I've been thinking about the security when of TCP messengers. For example,
often I am communicating with actors from TestStand steps. Everything runs on the same PC,
works perfectly. However, I realised that if potential attacker knows the port of the actor,
he/she will be able to connect to my actors from outside and potentially do harm.

At the moment I see 2 solutions:

1. To configure Windows Firewall for the main application to only accept local connections.

2. To bind TCP socket to localhost by specifying "net address" here:

1474620694_netaddress.png.da8c77a00114195d927d82002e6eaf9c.png

The second solution will work regardless of firewall settings.
It would be great to hear your thoughts.Thank you!

Kind Regards,
Max

 

 

Link to comment

Here is a beta version, 1.11.0, with changes to TCP Messengers to support the various TCP options about ports and the like.  Also adds a Keep-alive to detect half-open connections.  Bitbucket Issues 26, 27, 33.  I have not had time to do more than a basic test; it would be a help if someone could give it a try.   Note that I've replaced some of the creation VIs for TCP Messengers in the pallette (originals should still work, but I wanted to change VI names).

drjdpowell_lib_messenging-1_11.0_116.vip

  • Thanks 1
Link to comment

Thank you for your work Dr Powell.

I did some testing of the following VIs from the palette:

1.PNG.dce311abb6ec3c0f928d99f9aea43b07.PNG

And I can confirm that I was able to establish connection using Service Name or Specific Port and
also it was possible to isolate actor from external network by setting net address to "2130706433" (127.0.0.1).

3.PNG.036d38be53c16f6b2938fafb97f533ab.PNG

However I noticed that the "Name the TCP Service.vi" is now missing from the palette.
I could still get it from the new "Create the TCP Server.vi" and wire it as a "event messenger subtype" to "Actor Startup" VI.

Kind Regards,
Max

Link to comment
  • 2 weeks later...

Dr Powell,

Thank you for the new release of the library. It is a beautiful piece of work.

Today I faced a new challenge (sorry for bombarding the forum). One of my actors is using TestStand API to execute test sequences. I noticed strange behavior, when this
actor is executing synchronous TestStand API calls that take a noticeable time (>200ms) other actors (5 of them, fully decoupled) are also being freezed until that API call is finished.
I started researching what could cause this. I believe it is a deadlock because preferred execution system of all actors is the same:
https://zone.ni.com/reference/en-XX/help/370052W-01/tsapiref/infotopics/lv_preferred_exec/

Ideally I would like to set specific "Preferred Execution System" for actors that require it. I tried to change this setting in VI properties
for ActorNR.vi. But doing this affects whole class hierarchy and propagates to all actors (as a result they stay under the same execution system).

Idea that I have in my mind is to create another actor type that is configured for specific execution system.
It would be great to hear how would you approach this issue?

Thank you!

Kind Regards,
Maksims

Link to comment
37 minutes ago, Maksim Kuznetsov said:

One of my actors is using TestStand API to execute test sequences. I noticed strange behavior, when this
actor is executing synchronous TestStand API calls that take a noticeable time (>200ms) other actors (5 of them, fully decoupled) are also being freezed until that API call is finished...

By default, subVI are set to "Same as Caller" execution system, but they can be a specific system instead.  I suspect it might be just the subVI that does the TestStand call that needs to be in a different Execution System, not the calling Actor.vi itself.  So try just changing the subVI.

  • Thanks 1
Link to comment
6 minutes ago, drjdpowell said:

By default, subVI are set to "Same as Caller" execution system, but they can be a specific system instead.  I suspect it might be just the subVI that does the TestStand call that needs to be in a different Execution System, not the calling Actor.vi itself.  So try just changing the subVI.

If that doesn't work you may have to separate the TestStandAPI calls out.  Are you using your TestStand Actor as a GUI or user interface?  If so you may have to create another Actor to separate out the TestStand API calls that are causing the log jam into a new Actor....That new actor should not have any property/invoke nodes which would force its VI into the UI thread.

  • Thanks 1
Link to comment

Thank you for all the suggestions! Since I based my actor on NI example of TestStand UI the SubVIs were already configured to run in a separate execution
and I still experienced blockings. That is why I thought maybe the actor itself has to run in a different execution and posted in this thread.
However today I found out that the freeze was caused by a dialog box that was created asynchronously by the TestStand API (e.g. when it noticed that sequence file was changed).
There is a "TestStand - Set TestStand Application Window.vi" that accepts VI reference and makes all the dialog boxes modal to that VI (I set it to TestStand ActorNR.vi reference).

It works perfectly fine when TestStand Actor's front panel is the front-most window, no blockings. However if another actor's front panel is currently in focus (front-most) there will be
no dialog box until I switch the focus back to TestStand Actor front panel. And in this case both actors get frozen until I acknowledge that dialog box.

P.S. Simple front panel drag around with a mouse pointer make actors unfreeze.

I think it is related to modality settings that I will explore in more details.
Thanks again James and bbean. I highly appreciate your time and support.

Link to comment
  • 5 weeks later...

How does your calling code keep track of the different clones?  Match that.  If the caller gives them specific names, such as storing the actor addresses in a cluster, then use those names in a JSON object.  If, instead, the actor addresses naturally are treated as an array, then store a JSON array.

Link to comment

Exactly.  You can even do the case where the types of actor is determined at run time, in which case the caller makes a JSON object like this:

{"Actor Class Name":"MyActorX.lvclass","Config":{...}} 

Then on application load, the caller launches an actor of that class, then sends it the "Config".

Link to comment
  • 3 weeks later...

Hello Dr Powell and everybody. I am facing some frustrating behavior, already spent many hours trying to figure it out, but no luck.
Maybe you could assist me. In my previous posts I mentioned that I have been working on TestStand Actor.

Current setup:
TestStand API wrapped in Messenger Library. It is calling Steps that also use Messenger Library to send messages (through TCP) to other actors
that control hardware.

Issue:
TestStand Actor and LabVIEW steps share the same application instance. Hence, I am getting a conflicts/errors when trying to run sequences.
The same steps work perfectly from normal TestStand provided by NI.

I am looking for a way to isolate LabVIEW steps code from the TestStand Actor. It can be accomplished by putting all the dependencies in "lvlib" or "lvlibp". This way Messenger Library
in custom steps will have a separate namespace. However when putting Messenger Library in "lvlib" or "lvlibp", my steps can no longer receive TCP replies. Only "Send Message"
without a reply address work (if reply address is provided the send message makes the destination address invalid) . Interestingly, the published events are received successfully.

Do you know why would "Send Message (without reply)" and "Receive Notification" work, however "Send Message (with reply)" wouldn't? (when Messenger Library is in put in .lvlib)

Thank you!

Kind regards,
Max

Edited by Maksim Kuznetsov
typo
Link to comment

Could you explain a little more what you are doing.  I am confused that you are using TCP (something usually used for communication BETWEEN applications) and having namespace conflicts (which happen when you have multiple paths to the same-named thing on the SAME instance).

On no condition would I recommend renamespacing the library, as TCP involves flattening the messages to strings, and that uses the class names, which you are changing.  The TCP Messengers override the usually LabVIEW flattening for some common messages for reasons of performance, replacing the class names with a much-shorter format using an enum for class type.  This is why some messages will still work.  But this is strictly limited.

I had imagined, when you first brought up Teststand and TCP Messengers, that you were building a separate LabVIEW application, and that Teststand (in a separate application) connects to the first via TCP.  That could not have conflicts.

Link to comment

Thank you for your reply. Indeed, I built a separate application (.exe) for TestStand. It is using the Messenger Library with TCP input messenger. Actors from other applications
connect to TestStand Actor and run sequences, subscribe for sequence finished events, etc. That part works well.

The conflict arises when the sequence file itself contains a LabVIEW step that connects to another actor (e.g. valve controller) and sends a command to it.
That step also utilises the Messenger Library, but loads it from the folder next to the sequence (to make sure it runs on systems with LabVIEW Run-Time only).
As a result I have a TestStand Actor that runs in LabVIEW Run-Time application instance, that actor loads a sequence that has steps that also run in LabVIEW Run-Time.
And that's where problem arises. Sequence can't load any Messenger Library VIs because they are already in memory (loaded by TestStand Actor).

Edited by Maksim Kuznetsov
typo
Link to comment

Yes, the steps are source.

I understand that ideally steps should be built and it can be achieved by building them into "packed library". But packed library
automatically brake the communication (same as when changing the namespace of the Messenger library).

Another possibility would be to build every step into an executable, but then returning errors, setting step input data would
become more complex.

Link to comment

Ah, so really your issue is about "plug in" architecture where you want to be able to drop new plugins to a folder, and there is the problem of the plugins depending on common code (Messenger Library).   I have only previously done dynamically-loaded code that is built as part of the same build as the EXE that loads them, where such issues don't come up.  You might be better to ask in another forum, as the issues you have are not Messenger-Library specific and will be common to other such plug-in designs.

Another option is to put the entire of Messenger Library in a PPL (changing namespace) but then have ALL your code use this PPL instead of VI.lib (so everything uses the same namespace).

Link to comment

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...

Important Information

By using this site, you agree to our Terms of Use.