ULog File Format

ULog is the file format used for logging system data. The format is self-describing, i.e. it contains the format and message types that are logged.

It can be used for logging device inputs (sensors, etc.), internal states (cpu load, attitude, etc.) and printf log messages.

The format uses Little Endian for all binary types.

Data types

The following binary types are used. They all correspond to the types in C:

Additionally all can be used as an array, eg. float[5]. In general all strings (char[length]) do not contain a '\0' at the end. String comparisons are case sensitive.

File structure

The file consists of three sections:

----------------------
|       Header       |
----------------------
|    Definitions     |
----------------------
|        Data        |
----------------------

Header Section

The header is a fixed-size section and has the following format (16 bytes):

----------------------------------------------------------------------
| 0x55 0x4c 0x6f 0x67 0x01 0x12 0x35 | 0x01         | uint64_t       |
| File magic (7B)                    | Version (1B) | Timestamp (8B) |
----------------------------------------------------------------------

Version is the file format version, currently 1. Timestamp is a uint64_t integer, denotes the start of the logging in microseconds.

Definitions Section

Variable length section, contains version information, format definitions, and (initial) parameter values.

The Definitions and Data sections consist of a stream of messages. Each starts with this header:

struct message_header_s {
    uint16_t msg_size;
    uint8_t msg_type
};

msg_size is the size of the message in bytes without the header (hdr_size= 3 bytes). msg_type defines the content and is one of the following:

• 'B': Flag bitset message.

  struct ulog_message_flag_bits_s {
    uint8_t compat_flags[8];
    uint8_t incompat_flags[8];
    uint64_t appended_offsets[3]; ///< file offset(s) for appended data if appending bit is set
  };
  

  This message must be the first message, right after the header section, so that it has a fixed constant offset.

  • compat_flags: compatible flag bits. None of them is currently defined and all must be set to 0. These bits can be used for future ULog changes that are compatible with existing parsers. It means parsers can just ignore the bits if one of the unknown bits is set.
  • incompat_flags: incompatible flag bits. The LSB bit of index 0 is set to one if the log contains appended data and at lease one of the appended_offsets is non-zero. All other bits are undefined und must be set to 0. If a parser finds one of these bits set, it must refuse to parse the log. This can be used to introduce breaking changes that existing parsers cannot handle.
  • appended_offsets: File offsets (0-based) for appended data. If no data is appended, all offsets must be zero. This can be used to reliably append data for logs that may stop in the middle of a message. A process appending data should do:
    • set the relevant incompat_flags bit,
    • set the first appended_offsets that is 0 to the length of the log file,
    • then append any type of messages that are valid for the Data section.

  It is possible that there are more fields appended at the end of this message in future ULog specifications. This means a parser must not assume a fixed length of this message. If the message is longer than expected (currently 40 bytes), the exceeding bytes must just be ignored.

• 'F': format definition for a single (composite) type that can be logged or used in another definition as a nested type.

  struct message_format_s {
    struct message_header_s header;
    char format[header.msg_size-hdr_size];
  };
  

  format: plain-text string with the following format: message_name:field0;field1; There can be an arbitrary amount of fields (at least 1), separated by ;. A field has the format: type field_name or type[array_length] field_name for arrays (only fixed size arrays are supported). type is one of the basic binary types or a message_name of another format definition (nested usage). A type can be used before it's defined. There can be arbitrary nesting but no circular dependencies.

  Some field names are special:

  • timestamp: every logged message (message_add_logged_s) must include a timestamp field (does not need to be the first field). Its type can be: uint64_t (currently the only one used), uint32_t, uint16_t or uint8_t. The unit is always microseconds, except for uint8_t it's milliseconds. A log writer must make sure to log messages often enough to be able to detect wrap-arounds and a log reader must handle wrap-arounds (and take into account dropouts). The timestamp must always be monotonic increasing for a message serie with the same msg_id.

  • Padding: field names that start with _padding should not be displayed and their data must be ignored by a reader. These fields can be inserted by a writer to ensure correct alignment.

    If the padding field is the last field, then this field will not be logged, to avoid writing unnecessary data. This means the message_data_s.data will be shorter by the size of the padding. However the padding is still needed when the message is used in a nested definition.

• 'I': information message.

  struct message_info_s {
    struct message_header_s header;
    uint8_t key_len;
    char key[key_len];
    char value[header.msg_size-hdr_size-1-key_len]
  };
  

  key is a plain string, as in the format message (can also be a custom type), but consists of only a single field without ending ;, eg. float[3] myvalues. value contains the data as described by key.

  Note that an information message with a certain key must occur at most once in the entire log. Parsers can store information messages as a dictionary.

  Predefined information messages are:

The format of ver_sw_release and ver_os_release is: 0xAABBCCTT, where AA is major, BB is minor, CC is patch and TT is the type. Type is defined as following: >= 0: development, >= 64: alpha version, >= 128: beta version, >= 192: RC version, == 255: release version. So for example 0x010402ff translates into the release version v1.4.2.

This message can also be used in the Data section (this is however the preferred section).

• 'M': information message multi.

  struct ulog_message_info_multiple_header_s {
    uint8_t is_continued; ///< can be used for arrays
    uint8_t key_len;
    char key[key_len];
    char value[header.msg_size-hdr_size-2-key_len]
  };
  

  The same as the information message, except that there can be multiple messages with the same key (parsers store them as a list). The is_continued can be used for split-up messages: if set to 1, it is part of the previous message with the same key. Parsers can store all information multi messages as a 2D list, using the same order as the messages occur in the log.

• 'P': parameter message. Same format as message_info_s. If a parameter dynamically changes during runtime, this message can also be used in the Data section. The data type is restricted to: int32_t, float.

This section ends before the start of the first message_add_logged_s or message_logging_s message, whichever comes first.

Data Section

The following messages belong to this section:

• 'A': subscribe a message by name and give it an id that is used in message_data_s. This must come before the first corresponding message_data_s.

  struct message_add_logged_s {
    struct message_header_s header;
    uint8_t multi_id;
    uint16_t msg_id;
    char message_name[header.msg_size-hdr_size-3];
  };
  

  multi_id: the same message format can have multiple instances, for example if the system has two sensors of the same type. The default and first instance must be 0. msg_id: unique id to match message_data_s data. The first use must set this to 0, then increase it. The same msg_id must not be used twice for different subscriptions, not even after unsubscribing. message_name: message name to subscribe to. Must match one of the message_format_s definitions.

• 'R': unsubscribe a message, to mark that it will not be logged anymore (not used currently).

  struct message_remove_logged_s {
    struct message_header_s header;
    uint16_t msg_id;
  };
  

• 'D': contains logged data.

  struct message_data_s {
    struct message_header_s header;
    uint16_t msg_id;
    uint8_t data[header.msg_size-hdr_size];
  };
  

  msg_id: as defined by a message_add_logged_s message. data contains the logged binary message as defined by message_format_s. See above for special treatment of padding fields.

• 'L': Logged string message, i.e. printf output.

  struct message_logging_s {
    struct message_header_s header;
    uint8_t log_level;
    uint64_t timestamp;
    char message[header.msg_size-hdr_size-9]
  };
  

  timestamp: in microseconds, log_level: same as in the Linux kernel:

• 'S': synchronization message so that a reader can recover from a corrupt message by searching for the next sync message (not used currently).

  struct message_sync_s {
    struct message_header_s header;
    uint8_t sync_magic[8];
  };
  

  sync_magic: to be defined.

• 'O': mark a dropout (lost logging messages) of a given duration in ms. Dropouts can occur e.g. if the device is not fast enough.

  struct message_dropout_s {
    struct message_header_s header;
    uint16_t duration;
  };
  

• 'I': information message. See above.

• 'M': information message multi. See above.

• 'P': parameter message. See above.

Requirements for Parsers

A valid ULog parser must fulfill the following requirements:

• Must ignore unknown messages (but it can print a warning).
• Parse future/unknown file format versions as well (but it can print a warning).
• Must refuse to parse a log which contains unknown incompatibility bits set (incompat_flags of ulog_message_flag_bits_s message), meaning the log contains breaking changes that the parser cannot handle.
• A parser must be able to correctly handle logs that end abruptly, in the middle of a message. The unfinished message should just be discarged.

• For appended data: a parser can assume the Data section exists, i.e. the offset points to a place after the Definitions section.

  Appended data must be treated as if it was part of the regular Data section.

Known Implementations

• PX4 Firmware: C++
  • logger module
  • replay module
  • hardfault_log module: append hardfault crash data.
• pyulog: python, ULog parser library with CLI scripts.
• FlightPlot: Java, log plotter.
• pyFlightAnalysis: Python, log plotter and 3D visualization tool based on pyulog.
• MAVLink: Messages for ULog streaming via MAVLink (note that appending data is not supported, at least not for cut off messages).
• QGroundControl: C++, ULog streaming via MAVLink and minimal parsing for GeoTagging.
• mavlink-router: C++, ULog streaming via MAVLink.
• MAVGAnalysis: Java, ULog streaming via MAVLink and parser for plotting and analysis.

File Format Version History

Changes in version 2

Addition of ulog_message_info_multiple_header_s and ulog_message_flag_bits_s messages and the ability to append data to a log. This is used to add crash data to an existing log. If data is appended to a log that is cut in the middle of a message, it cannot be parsed with version 1 parsers. Other than that forward and backward compatibility is given if parsers ignore unknown messages.