ULog File Format
Last updated
Last updated
ULog is the file format used for logging messages. The format is self-describing, i.e. it contains the format and uORB message types that are logged. This document is meant to be the ULog File Format Spec Documentation. It is intended especially for anyone who is interested in writing a ULog parser / serializer and needs to decode / encode files.
PX4 uses ULog to log uORB topics as messages related to (but not limited to) the following sources:
Device inputs: Sensors, RC input, etc.
Internal states: CPU load, attitude, EKF state, etc.
String messages: printf statements, including PX4_INFO() and PX4_ERR().
The format uses little endian memory layout for all binary types (the least significant byte (LSB) of data type is placed at the lowest memory address).
The following binary types are used for logging. They all correspond to the types in C.
int8_t, uint8_t
1
int16_t, uint16_t
2
int32_t, uint32_t
4
int64_t, uint64_t
8
float
4
Additionally the types can be used as an array: e.g. float[5].
Strings (char[length]) do not contain the termination NULL character '\0' at the end.
:::note String comparisons are case sensitive, which should be taken into account when comparing message names when adding subscriptions. :::
ULog files have the following three sections:
----------------------
| Header |
----------------------
| Definitions |
----------------------
| Data |
----------------------A description of each section is provided below.
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) |
----------------------------------------------------------------------File Magic (7 Bytes): File type indicator that reads "ULogXYZ where XYZ is the magic bytes sequence 0x01 0x12 0x35"
Version (1 Byte): File format version (currently 1)
Timestamp (8 Bytes): uint64_t integer that denotes when the logging started in microseconds.
The Definitions and Data sections contain a number of messages. Each message is preceded by 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.
msg_type defines the content, and is a single character.
:::note Message sections below are prefixed with the character that corresponds to it's msg_type. :::
The definitions section contains basic information such as software version, message format, initial parameter values, and so on.
The message types in this section are:
:::note This message must be the first message right after the header section, so that it has a fixed constant offset from the start of the file! :::
This message provides information to the log parser whether the log is parsable or not.
struct ulog_message_flag_bits_s {
struct message_header_s header; // msg_type = 'B'
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
};compat_flags: compatible flag bits
These flags indicate the presence of features in the log file that are compatible with any ULog parser.
compat_flags[0]: DEFAULT_PARAMETERS (Bit 0): if set, the log contains default parameters message
The rest of the bits are currently not defined and must be set to 0. These bits can be used for future ULog changes that are compatible with existing parsers. For example, adding a new message type can be indicated by defining a new bit in the standard, and existing parsers will ignore the new message type. It means parsers can just ignore the bits if one of the unknown bits is set.
incompat_flags: incompatible flag bits.
incompat_flags[0]: DATA_APPENDED (Bit 0): if set, the log contains appended data and at least one of the appended_offsets is non-zero.
The rest of the bits are currently not defined and must be set to 0. This can be used to introduce breaking changes that existing parsers cannot handle. For example, when an old ULog parser that didn't have the concept of DATA_APPENDED reads the newer ULog, it would stop parsing the log as the log will contain out-of-spec messages / concepts. If a parser finds any of these bits set that isn't specified, it must refuse to parse the log.
appended_offsets: File offset (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. For example, crash dumps.
A process appending data should do:
set the relevant incompat_flags bit
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 msg_size is bigger than expected (currently 40), any additional bytes must be ignored/discarded.
Format message defines a single message name and it's inner fields in a single string.
struct message_format_s {
struct message_header_s header; // msg_type = 'F'
char format[header.msg_size];
};format is a plain-text string with the following format: message_name:field0;field1;
There can be an arbitrary amount of fields (minimum 1), separated by ;.
A field has the format: type field_name, or for an array: type[array_length] field_name is used (only fixed size arrays are supported).
A 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.
e.g. The message MessageA:MessageB[2] msg_b can come before the MessageB:uint_8[3] data
There can be arbitrary nesting but no circular dependencies
e.g. MessageA:MessageB[2] msg_b & MessageB:MessageA[4] msg_a
Some field names are special:
timestamp: every Subscription Message must include a timestamp 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 in uint8_t where the unit is in milliseconds.
The timestamp must always be monotonic increasing for a message series with the same msg_id.
Optionally, when using a small timestamp datatype such as uint8_t, the log writer must make sure to log messages often enough to be able to detect wrap-arounds (when the timestamp overflows the data type and goes back to 0)
In that case, the log reader must handle wrap-arounds as well, and take into account dropouts.
_padding{}: field names that start with _padding (e.g. _padding[3]) 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 may not be logged, to avoid writing unnecessary data.
The Information message defines a dictionary type definition key : value pair for any information, including but not limited to Hardware version, Software version, Build toolchain for the software, etc.
struct ulog_message_info_header_s {
struct message_header_s header; // msg_type = 'I'
uint8_t key_len;
char key[key_len];
char value[header.msg_size-1-key_len]
};key_len: Length of the key value
key: Contains the key string. eg. char[value_len] sys_toolchain_ver
value: Contains the data (with the length value_len) corresponding to the key e.g. 9.4.0.
:::note A key : value pair defined in the Information message should be unique. Meaning there shouldn't be more than one definition with the same key value! :::
Parsers can store information messages as a dictionary.
Predefined information messages are:
char[value_len] sys_name
Name of the system
"PX4"
char[value_len] ver_hw
Hardware version (board)
"PX4FMU_V4"
char[value_len] ver_hw_subtype
Board subversion (variation)
"V2"
:::note value_len represents the data size of the value. This is described in the key. :::
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.
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).
Multi information message serves the same purpose as the information message, but for long messages or multiple messages with the same key.
struct ulog_message_info_multiple_header_s {
struct message_header_s header; // msg_type = 'M'
uint8_t is_continued; // can be used for arrays
uint8_t key_len;
char key[key_len];
char value[header.msg_size-2-key_len]
};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.
Parameter message in the Definitions section defines the parameter values of the vehicle when logging is started. It uses the same format as the Information Message.
struct message_info_s {
struct message_header_s header; // msg_type = 'P'
uint8_t key_len;
char key[key_len];
char value[header.msg_size-1-key_len]
};If a parameter dynamically changes during runtime, this message can also be used in the Data section as well.
The data type is restricted to int32_t and float.
The default parameter message defines the default value of a parameter for a given vehicle and setup.
struct ulog_message_parameter_default_header_s {
struct message_header_s header; // msg_type = 'Q'
uint8_t default_types;
uint8_t key_len;
char key[key_len];
char value[header.msg_size-2-key_len]
};default_types is a bitfield and defines to which group(s) the value belongs to.
At least one bit must be set:
1<<0: system wide default
1<<1: default for the current configuration (e.g. an airframe)
A log may not contain default values for all parameters. In those cases the default is equal to the parameter value, and different default types are treated independently.
This message can also be used in the Data section, and the data type is restricted to int32_t and float.
This section ends before the start of the first Subscription Message or Logging message, whichever comes first.
The message types in the Data section are:
A: Subscription MessageSubscribe a message by name and give it an id that is used in Logged data Message. This must come before the first corresponding Logged data Message.
struct message_add_logged_s {
struct message_header_s header; // msg_type = 'A'
uint8_t multi_id;
uint16_t msg_id;
char message_name[header.msg_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 Logged data Message data. The first use must set this to 0, then increase it.
The same msg_id must not be used twice for different subscriptions.
message_name: message name to subscribe to. Must match one of the definitions.
R: Unsubscription MessageUnsubscribe a message, to mark that it will not be logged anymore (not used currently).
struct message_remove_logged_s {
struct message_header_s header; // msg_type = 'R'
uint16_t msg_id;
};struct message_data_s {
struct message_header_s header; // msg_type = 'D'
uint16_t msg_id;
uint8_t data[header.msg_size-2];
};msg_id: as defined by a Subscription Message
data contains the logged binary message as defined by Format Message
See above for special treatment of padding fields.
Logged string message, i.e. printf() output.
struct message_logging_s {
struct message_header_s header; // msg_type = 'L'
uint8_t log_level;
uint64_t timestamp;
char message[header.msg_size-9]
};timestamp: in microseconds
log_level: same as in the Linux kernel:
EMERG
'0'
System is unusable
ALERT
'1'
Action must be taken immediately
CRIT
'2'
Critical conditions
struct message_logging_tagged_s {
struct message_header_s header; // msg_type = 'C'
uint8_t log_level;
uint16_t tag;
uint64_t timestamp;
char message[header.msg_size-9]
};tag: id representing source of logged message string. It could represent a process, thread or a class depending upon the system architecture.
For example, a reference implementation for an onboard computer running multiple processes to control different payloads, external disks, serial devices etc can encode these process identifiers using a uint16_t enum into the tag attribute of struct as follows:
enum class ulog_tag : uint16_t {
unassigned,
mavlink_handler,
ppk_handler,
camera_handler,
ptp_handler,
serial_handler,
watchdog,
io_service,
cbuf,
ulg
};timestamp: in microseconds
log_level: same as in the Linux kernel:
EMERG
'0'
System is unusable
ALERT
'1'
Action must be taken immediately
CRIT
'2'
Critical conditions
Synchronization message so that a reader can recover from a corrupt message by searching for the next sync message.
struct message_sync_s {
struct message_header_s header; // msg_type = 'S'
uint8_t sync_magic[8];
};sync_magic: [0x2F, 0x73, 0x13, 0x20, 0x25, 0x0C, 0xBB, 0x12]
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; // msg_type = 'O'
uint16_t duration;
};Since the Definitions and Data Sections use the same message header format, they also share the same messages listed below:
For the Data section, the Parameter Message is used when the parameter value changes
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 Flag Bits 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 discarded.
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.
PX4-Autopilot: C++
: append hardfault crash data.
: python, ULog reader and writer library with CLI scripts.
: C++, ULog reader and writer library.
: Java, log plotter.
: Messages for ULog streaming via MAVLink (note that appending data is not supported, at least not for cut off messages).
: C++, ULog streaming via MAVLink and minimal parsing for GeoTagging.
: C++, ULog streaming via MAVLink.
: Java, ULog streaming via MAVLink and parser for plotting and analysis.
: C++/Qt application to plot logs and time series. Supports ULog since version 2.1.3.
: Javascript, ULog reader and parser outputs log in JSON object format.
Addition of Multi Information Message and Flag Bits Message 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.
appended_offsets that is currently 0 to the length of the log file without the appended data, as that is where the new data will startappend any type of messages that are valid for the Data section.
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.
double
8
bool, char
1
char[value_len] ver_sw
Software version (git tag)
"7f65e01"
char[value_len] ver_sw_branch
git branch
"master"
uint32_t ver_sw_release
Software version (see below)
0x010401ff
char[value_len] sys_os_name
Operating System Name
"Linux"
char[value_len] sys_os_ver
OS version (git tag)
"9f82919"
uint32_t ver_os_release
OS version (see below)
0x010401ff
char[value_len] sys_toolchain
Toolchain Name
"GNU GCC"
char[value_len] sys_toolchain_ver
Toolchain Version
"6.2.1"
char[value_len] sys_mcu
Chip name and revision
"STM32F42x, rev A"
char[value_len] sys_uuid
Unique identifier for vehicle (eg. MCU ID)
"392a93e32fa3"...
char[value_len] log_type
Type of the log (full log if not specified)
"mission"
char[value_len] replay
File name of replayed log if in replay mode
"log001.ulg"
int32_t time_ref_utc
UTC Time offset in seconds
-3600
ERR
'3'
Error conditions
WARNING
'4'
Warning conditions
NOTICE
'5'
Normal but significant condition
INFO
'6'
Informational
DEBUG
'7'
Debug-level messages
ERR
'3'
Error conditions
WARNING
'4'
Warning conditions
NOTICE
'5'
Normal but significant condition
INFO
'6'
Informational
DEBUG
'7'
Debug-level messages