MCAP logging error: Message: invalid magic bytes in Footer: 0xD0FF0B0000000000

[scastro-bdai]

Hi @facontidavide!

I've been trying to get started with this tool and I'm probably doing something wrong with the MCAPSink.

So first of all, I was running into the chunking issue explained in #20 -- since I was only trying to log a single scalar double value over time, I wasn't seeing anything coming up on file until I logged a ton of data.

Next up, when I finally get a file that's partially full, I try to load it into PlotJuggler and get this error:

 Message: invalid magic bytes in Footer: 0xD0FF0B0000000000

Is there some special way to make sure the file/buffer are closed properly? Or something missing? Here is my code:

(in the constructor)

  mcap_sink_ = std::make_shared<DataTamer::MCAPSink>("ruckig_traj_executor_log.mcap");
  log_channel_ = DataTamer::LogChannel::create("trajectory");
  log_channel_->addDataSink(mcap_sink_);

  log_channel_->registerValue("joint_1", &joint_value_);

in some subscriber callback

  joint_value_ += 1.0;
  if (!log_channel_->takeSnapshot()) {
    RCLCPP_INFO_STREAM(get_node()->get_logger(), "LOGGING FAILED!");
  } else {
    RCLCPP_INFO_STREAM(get_node()->get_logger(), "LOGGING SUCCEEDED!");
  }

Confirming the "LOGGING SUCCEEDED" printout is happening.

Any ideas? Thanks!

[jlack1987]

I could be wrong, but I have also had similar issues and it seems like issues like that are due to the mcap writer not properly destructing and getting to finish writing the file.

[facontidavide]

yes, that is the problem... i need to find a way to make this less fragile to crashes

[scastro-bdai]

Yep -- ok, confirming that for whatever reason, this code is not properly destroying the object containing the MCAPSink. I blame the use of pluginlib in this whole mess :)

If I manually reset() the shared pointer to the sink during execution, the file reads correctly in PlotJuggler.

Not sure if you'd like to leave this open or closed, but I defer to you -- thanks, guys!

[jlack1987]

I will at the very least confirm that properly destructing "mcap stuff" can be finicky. I haven't really dug into it, but I think it may be the provided mcap writer or something bc i've written an mcap writer that uses a json schema and it suffers from similar issues in that you just have to be super careful to make sure it destructs properly