Data corruption

[matiasandina]

This issue will try to put examples of data corruption affecting FEDWatcher functioning

1. Data corruption in FED number

FED 17 sent this string in the place of Session_type.

Pav\x83SPlusMinusl17

Problem

There's the more common data corruption of changing one character by \x[0-9]{2}. There's the speciffic problem of not splitting the 17 (Device_Number) and the program title. It then populated Device_Number with the value in Battery_Voltage

This happened in the middle of the session. Which means this row had to be manually inserted in the proper place on the proper table.

This error was not a one time event, several FEDWatcher had this. Not sure if related but the RPi were being pushed at 100% CPU usage. I would think it's data corruption on the emission not on the receiver side.

[matiasandina]

We could potentially implement a data cleanup routine. We could send each message 3 times and take the "average" of each character. This will likely reduce gibberish events. It will probably increase latency.

[matiasandina]

Here's an example. Instead of sending FreefFeed, 34, 3.97, ... It's sending FreeFeed\x06l34, which breaks the int() call somewhere else and stops the proper data saving

05/18/2023 06:02:32,1.9.2,FreeFeed\x06l34,3.97,1,1,Pellet,Left,31,76,59,0,20.49,31,nan
Process Process-2:
Traceback (most recent call last):
  File "/usr/lib/python3.7/multiprocessing/process.py", line 297, in _bootstrap
    self.run()
  File "/usr/lib/python3.7/multiprocessing/process.py", line 99, in run
    self._target(*self._args, **self._kwargs)
  File "/home/pi/FEDWatcher/fedwatcher/src/fedwatcher.py", line 234, in runHelper
    self._save_all_df()
  File "/home/pi/FEDWatcher/fedwatcher/src/fedwatcher.py", line 452, in _save_all_df
    self._save_to_csv(df_data)
  File "/home/pi/FEDWatcher/fedwatcher/src/fedwatcher.py", line 417, in _save_to_csv
    filename = f"FED{int(df_data[0]['Device_Number']):03d}_{timestr}_{self.session_num:02d}.csv"
ValueError: invalid literal for int() with base 10: '3.97'

[matiasandina]

It might have been that the issue was that there wasn't enough space for the termination character. Maybe your classic off-by-one error ? Will test before closing

Update

I see the errors continue only when the Event is LeftWithPellet or RightWithPellet. This is probably due to the length of the assigned character. So far, I haven't had any more breaking lines, but this issue is still not fully fixed.

[matiasandina]

This is not fully done. For example, this is 12 being passed as q2 and then the correct error handling and breaking of FEDWatcher

Error: Unable to convert 'Device_Number' to an integer.
Process Process-2:
Traceback (most recent call last):
  File "/home/pi/FEDWatcher/fedwatcher/src/fedwatcher.py", line 431, in _save_to_csv
    device_number = int(float(df_data[0]['Device_Number']))
ValueError: could not convert string to float: 'q2'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/usr/lib/python3.7/multiprocessing/process.py", line 297, in _bootstrap
    self.run()
  File "/usr/lib/python3.7/multiprocessing/process.py", line 99, in run
    self._target(*self._args, **self._kwargs)
  File "/home/pi/FEDWatcher/fedwatcher/src/fedwatcher.py", line 249, in runHelper
    self._save_all_df()
  File "/home/pi/FEDWatcher/fedwatcher/src/fedwatcher.py", line 479, in _save_all_df
    self._save_to_csv(df_data)
  File "/home/pi/FEDWatcher/fedwatcher/src/fedwatcher.py", line 435, in _save_to_csv
    raise ValueError(error_msg)
ValueError: Unable to convert 'Device_Number' to an integer.

[matiasandina]

Another example here 1w:44:54 gets parsed to 2023-12-07 01:44:54 and might affect how the pipeline works because of relying on datetime in other functions which might arrange by datetime (e.g., read_fed, recalculate_pellets).

# A tibble: 6 × 4
  `MM:DD:YYYY hh:mm:ss` Pi_Time             datetime            Pellet_Count
  <chr>                 <dttm>              <dttm>                     <dbl>
1 12/07/2023 17:43:43   2023-12-07 17:40:02 2023-12-07 17:43:43         1070
2 12/07/2023 17:44:14   2023-12-07 17:40:33 2023-12-07 17:44:14         1071
3 12/07/2023 1w:44:54   2023-12-07 17:41:14 2023-12-07 01:44:54         1072
4 12/07/2023 18:44:21   2023-12-07 18:40:40 2023-12-07 18:44:21         1073
5 12/07/2023 18:44:46   2023-12-07 18:41:06 2023-12-07 18:44:46         1074
6 12/07/2023 18:45:03   2023-12-07 18:41:23 2023-12-07 18:45:03         1075

[matiasandina]

There are predictable sources of "corruption" and somewhat unpredictable sources of corruption.

Predictable

durationStr seems to not handle RightWithPellet and LeftWithPellet

https://github.com/matiasandina/FEDWatcher/blob/9981d69dba178324212a1d3d678c5a07ff2bf3e8/sampleSketch/ITI_03/FED3.cpp#L918-L928

This could be changed to

char durationStr[20];
if (Event == "Pellet") {
    strcpy(durationStr, "nan");
} else if (Event == "LeftWithPellet" || Event == "Left") {
    sprintf(durationStr, "%.2f", leftInterval / 1000.0);
} else if (Event == "RightWithPellet" || Event == "Right") {
    sprintf(durationStr, "%.2f", rightInterval / 1000.0);
} else {
    strcpy(durationStr, "nan");
}

Unpredictable