vrmlogger: Logging breaks if malformed sqlite3 file on internal storage

Scenario:

External logging is in place, to an sd-card probably
An error occurs on logging to the sd-card, so vrmlogger tries to switch to an internal buffer
The internal buffer is also bad, so a stacktrace is logged, and that is the end of it.

    @4000000064d0bf9230351a44     main()
    @4000000064d0bf9230353984   File "/opt/victronenergy/vrmlogger/vrmlogger.py", line 681, in main
    @4000000064d0bf92305c841c     start_logging(state)
    @4000000064d0bf92305d91a4   File "/opt/victronenergy/vrmlogger/vrmlogger.py", line 440, in start_logging
    @4000000064d0bf9230773424     endpoint.send(VrmCommandType.ANNOUNCE, get_announce())
    @4000000064d0bf92307812cc   File "/opt/victronenergy/vrmlogger/http_endpoint.py", line 494, in send
    @4000000064d0bf9230950cc4     self._put_in_buffer_with_fallback_on_internal(data, int(time.time()))
    @4000000064d0bf92309602dc   File "/opt/victronenergy/vrmlogger/http_endpoint.py", line 449, in _put_in_buffer_with_fallback_on_internal
    @4000000064d0bf9230b0ed7c     self._active_buffer.put(data, timestamp)
    @4000000064d0bf9230b1c83c   File "/opt/victronenergy/vrmlogger/nonvolatilequeueramdiskwrapper.py", line 58, in put
    @4000000064d0bf9230bae05c     return super(NonVolatileQueueRamDiskWrapper, self).put(data, timestamp)
    @4000000064d0bf9230bbf1cc   File "/opt/victronenergy/vrmlogger/nonvolatilequeue.py", line 186, in put
    @4000000064d0bf9230cae204     cursor.execute("DELETE FROM queue WHERE timestamp < (strftime('%s', 'now') - ?)", (self._max_age,))
    @4000000064d0bf9230cbeba4 sqlite3.DatabaseError: database disk image is malformed

The code around nonvolatilequeueramdiskwrapper.py, line 58 (where the error is raised) will switch to a RAM-disk on some errors. But not this one.

According to documentation, sqlite3.DatabaseError is an unexpected exception to show up here. We would expect an OperationalError.

Nevertheless, it seems one easy improvement, is to catch the base error here, so we can more smoothly switch to RAM-disk. Then we at least continue logging. So:

[x] Make switch to RAM-disk mode more reliable, by catching more errors, including the malformed disk image one.
[x] Come up with a way to recover from a malformed disk image, which will probably be simply by deleting it under very well defined and controlled conditions.

victronenergy / venus

vrmlogger: Logging breaks if malformed sqlite3 file on internal storage #1128