search

ami-iit / yarp-openmct

Repo for YARP and OpenMCT integration.
BSD 3-Clause "New" or "Revised" License
6 stars 1 forks source link

Camera view poor performance: fake framegrabber image appears as scattered lines instead of a single one #102

Open nunoguedelha opened 2 years ago

nunoguedelha commented 2 years ago

When visualization the fake framegrabber generated image, we see a swiping set of scattered lines instead of a single one.

nunoguedelha commented 2 years ago

At some point we suspected that the line scattering observed in the fake-framegrabber rendering was due to the simultaneous display of multiple frames, and thus related to the Historical requests (refer to this documentation section). This seems now to be a wrong lead, as we can see in the screenshot below that even a single frame includes multiple lines instead of a singe one for the fake framegrabber.

image

Originally posted by @nunoguedelha in https://github.com/ami-iit/yarp-openmct/issues/71#issuecomment-1063569388

nunoguedelha commented 2 years ago

Logging the received image data on a yarp port

We publish the image data received through the plugin RealtimeTelemetryPlugin on a Yarp port /yarpjs/cameraEcho:o of type bottle.

For this we added the support of socket.io and yarp.js features in the visualizer plugin installation page:

https://github.com/ami-iit/yarp-openmct/blob/3d6105dc8efc7de4eefeef9ed27feba60bc21eba/openmctStaticServer/index.html#L12-L13

https://github.com/ami-iit/yarp-openmct/blob/3d6105dc8efc7de4eefeef9ed27feba60bc21eba/openmctStaticServer/index.html#L28-L33

https://github.com/ami-iit/yarp-openmct/blob/3d6105dc8efc7de4eefeef9ed27feba60bc21eba/openmctStaticServer/index.html#L143

https://github.com/ami-iit/yarp-openmct/blob/3d6105dc8efc7de4eefeef9ed27feba60bc21eba/openmctStaticServer/realtime-telemetry-plugin.js#L5-L16

We then use the stream video plugin example of yarp.js repository to display the streamed images. We can see in the stream the same scattered lines instead of a single one, as we see in the Open-MCT based visualizer.

We can conclude that the line scattered lines issue occurs in the data transmission between the telemetry server and the Open-MCT based visualizer client.

Sending a fixed frame

If we break the telemetry server we get a constant sample and thus a fixed frame on the visualizer, similar to the output illustrated on https://github.com/ami-iit/yarp-openmct/issues/71#issuecomment-1063569388, i.e with scattered lines.

image

This confirms the observation from the previous experiment. We then implement some changes in order to have the server send the raw image data buffer, and the client to apply the conversion to the URI format and base 64 (yarp.getImageSrc(...)): https://github.com/ami-iit/yarp-openmct/blob/34cae51cd90077c7d7d63f01f937562d5839ae15/openmctStaticServer/realtime-telemetry-plugin.js#L10-L21

This way it's easier to compare the data sent and received.

I'm suspecting an issue with the way the data transmission is scheduled, which could cause a race condition leading to concurrent data access in the case of big chunks of data, typically image data.

Originally posted by @nunoguedelha in https://github.com/ami-iit/yarp-openmct/issues/71#issuecomment-1068398512

nunoguedelha commented 2 years ago

This way it's easier to compare the data sent and received.

  • We log the data received on port /yarpjs/camLeftEye:i into a yarpview port
  • After running the telemetry server and client visualizer, we disconnect the framegrabber ouput: 
    yarp disconnect /icubSim/camLeftEye /yarpjs/camLeftEye:i

    This way, we get a fixed frame on the visualizer, and also on the yarpview display as well:

image

We can see that the outputs are well aligned. We then break the telemetry server and the client visualizer in order to get in the debugger the sent and received buffers.

The buffers are identical.

The problem is not in the transmission of the data between the server and client, but rather in the implementation of the yarp image type port reading.

From the yarp.js repository examples, we know that the implementation of image type port reading through websockets works pretty well. For that reason we'll try reading the port directly via yarp socket in the visualizer client (no history shall be supported).

Originally posted by @nunoguedelha in https://github.com/ami-iit/yarp-openmct/issues/71#issuecomment-1068702205

nunoguedelha commented 2 years ago

A debug version of the fix is implemented in https://github.com/ami-iit/yarp-openmct/tree/e5e0f28fbe8708954ac8240bfc43819b2955ff02.

This change broke the shutdown sequence. Fixing...

The proper closure of the iCubTelemVizServer server was blocked by the openmctStaticServer server hanging at closure itself.

Received SIGINT ...
{
  status: 'WRPLY',
  message: 'Process (PID 36504) OpenMCT Server stopping (signal SIGINT) ...'
}
iCub Telemetry Server closing: no further connection requests accepted.
Control Console Server closing: no further connection requests accepted.
[OPEN-MCT STATIC SERVER] stdout: Received SIGINT ...
Open-MCT Visualizer Server closing: no further incoming requests accepted. Refreshing the visualizer web page will fail.
Open-MCT Visualizer Server closed: all sockets closed.

The closure process resumes normally after we kill the nodejs process running the openmctStaticServer server.

iCub Telemetry Server closing: no further "subscribe"/"unsubscribe" requests accepted.
[INFO] |yarp.os.impl.PortCoreInputUnit| Removing input from /icubSim/inertial to /yarpjs/inertial:i
[INFO] |yarp.os.impl.PortCoreInputUnit| Removing input from /icubSim/left_leg/stateExt:o to /yarpjs/left_leg/stateExt:o
[INFO] |yarp.os.impl.PortCoreInputUnit| Removing input from /icubSim/camLeftEye to /yarpjs/camLeftEye:i
[INFO] |yarp.os.impl.PortCoreInputUnit| Removing input from /walking-coordinator/logger/data:o to /yarpjs/walking-coordinator-logger/data:i
iCub Telemetry Server closing: disconnected network ports.
Data transmission ended.
Closing all Telemetry Server and Control Console sockets!
[
  'iCub Telemetry Server closed: all sockets closed.',
  'Control Console Server closed: all sockets closed.'
]
Waiting for the debugger to disconnect...

Process finished with exit code 0

Originally posted by @nunoguedelha in https://github.com/ami-iit/yarp-openmct/issues/71#issuecomment-1069401564

nunoguedelha commented 2 years ago

Checks...

The previous analysis was verified by checking:

Quick Fix

A quick fix ( cd07da9dc6c5c25d9ae70acf80561ad7d6d5d634 ) consists in adding a process.exit() as a final SIGINT callback , as done for the iCubTelemVizServer task https://github.com/ami-iit/yarp-openmct/blob/e5e0f28fbe8708954ac8240bfc43819b2955ff02/iCubTelemVizServer/terminationHandler.js#L123-L124

We shall later implement a proper fix which does a backup of the original termination callbacks and restores them at the end, as done for the iCubTelemVizServer task https://github.com/ami-iit/yarp-openmct/blob/e5e0f28fbe8708954ac8240bfc43819b2955ff02/iCubTelemVizServer/iCubTelemVizServer.js#L197-L203 https://github.com/ami-iit/yarp-openmct/blob/e5e0f28fbe8708954ac8240bfc43819b2955ff02/iCubTelemVizServer/terminationHandler.js#L25-L30 We could actually re-use the class TerminationHandler as a base, and derived classes for processing the different closures (iCubTelemVizServer, openmctStaticServer).

Note on the Never Ending Main Processes

openmctStaticServer process keeps running most probably for the same reason as iCubTelemVizServer. We suspect a never expiring timer to be the cause in the node.js internal engine.

  function processTimers(now) {
    debug('process timer lists %d', now);
    nextExpiry = Infinity;

    let list;
    let ranAtLeastOneList = false;
    while (list = timerListQueue.peek()) {
      if (list.expiry > now) {
        nextExpiry = list.expiry;
        return refCount > 0 ? nextExpiry : -nextExpiry;
      }
      if (ranAtLeastOneList)
        runNextTicks();
      else
        ranAtLeastOneList = true;
      listOnTimeout(list, now);
    }
    return 0;
  }