mirror of
https://github.com/nasa/trick.git
synced 2024-12-21 06:03:10 +00:00
337 lines
14 KiB
Markdown
337 lines
14 KiB
Markdown
|
|
||
|
Data Recording provides the capability to specify any number of data recording groups,
|
||
|
each with an unlimited number of parameter references, and with each group recording
|
||
|
at different frequencies to different files in different formats.
|
||
|
|
||
|
All data is written to the simulation output directory.
|
||
|
|
||
|
### Format of Recording Groups
|
||
|
|
||
|
Trick allows recording in three different formats. Each recording group is readable by
|
||
|
different external tools outside of Trick.
|
||
|
|
||
|
- DRAscii - Human readable and compatible with Excel.
|
||
|
- DRBinary - Readable by previous Trick data products.
|
||
|
- DRHDF5 - Readable by Matlab.
|
||
|
|
||
|
DRHDF5 recording support is off by default. To enable DRHDF5 support Trick must be built with HDF5 support.
|
||
|
Go to http://www.hdf5group.org and download the latest pre-built hdf5 package for your system. Source packages are
|
||
|
available as well. We recommend getting the static library packages above the shared. Static packages make
|
||
|
your executable larger, but you will not have to deal with LD_LIBRARY issues. The HDF5 package may be installed
|
||
|
anywhere on your system. To tell Trick you have HDF5 run ${TRICK_HOME}/configure --with-hdf5=/path/to/hdf5.
|
||
|
Re-compile Trick to enable HDF5 support.
|
||
|
|
||
|
### Creating a New Recording Group
|
||
|
|
||
|
To create a new recording group, in the Python input file instantiate a new group by format name:
|
||
|
<tt><variable_name> = trick.<data_record_format>() ;</tt>
|
||
|
|
||
|
For example:
|
||
|
|
||
|
```
|
||
|
drg = trick.DRBinary() ;
|
||
|
```
|
||
|
|
||
|
Note: drg is just an example name. Any name may be used.
|
||
|
|
||
|
### Adding a Variable To Be Recorded
|
||
|
|
||
|
To add variables to the recording group call the <tt>drg.add_variable("<string_of_variable_name">)</tt> method of the recording group.
|
||
|
For example:
|
||
|
|
||
|
```python
|
||
|
drg.add_variable("ball.obj.state.output.position[0]")
|
||
|
drg.add_variable("ball.obj.state.output.position[1]")
|
||
|
```
|
||
|
|
||
|
An optional alias may also be specified in the method as <tt>drg.add_variable("<string_of_variable_name"> [, "<alias>"])</tt>.
|
||
|
If an alias is present as a second argument, the alias name will be used in the data recording file instead of the actual variable name.
|
||
|
For example:
|
||
|
|
||
|
```python
|
||
|
drg.add_variable("ball.obj.state.output.position[0]", "x_pos")
|
||
|
drg.add_variable("ball.obj.state.output.position[1]", "y_pos")
|
||
|
```
|
||
|
|
||
|
### Changing the Recording Rate
|
||
|
|
||
|
To change the recording rate call the <tt>set_cycle()</tt> method of the recording group.
|
||
|
|
||
|
```python
|
||
|
drg.set_cycle(0.01)
|
||
|
```
|
||
|
|
||
|
### Buffering Techniques
|
||
|
|
||
|
Data recording groups have three buffering options:
|
||
|
|
||
|
- DR_Buffer - the group will save recorded data to a buffer and use a separate thread to write recorded
|
||
|
data to disk. This will have little impact to the performance of the simulation. The downside
|
||
|
is that if the simulation crashes, the most recent recorded points may not be written to disk in time.
|
||
|
DR_Buffer is the default buffering technique. (For backwards compatibility, DR_Buffer can also be called DR_Thread_Buffer).
|
||
|
- DR_No_Buffer - the group will write recorded data straight to disk. All data is guaranteed to be written
|
||
|
to disk at simulation termination time. The downside of this method is that it is performed in
|
||
|
the main thread of the simulation and could impact real-time performance.
|
||
|
- DR_Ring_Buffer - the group will save a set number of records in memory and write this data to disk during
|
||
|
a graceful simulation termination. The advantage of this method is that there is only a set, usually
|
||
|
small, number of records written. The downside of this method is that if the simulation terminates
|
||
|
ungracefully, all recorded data may be lost.
|
||
|
|
||
|
To set the buffering technique call the <tt>set_buffer_type(trick.<buffering_option>)</tt> method of the recording group.
|
||
|
For example:
|
||
|
|
||
|
```python
|
||
|
drg.set_buffer_type(trick.DR_Buffer)
|
||
|
```
|
||
|
|
||
|
All buffering options (except for DR_No_Buffer) have a maximum amount of memory allocated to
|
||
|
holding data. See Trick::DataRecordGroup::set_max_buffer_size for buffer size information.
|
||
|
|
||
|
### Recording Frequency: Always or Only When Data Changes
|
||
|
|
||
|
Data recording groups have three recording frequency options:
|
||
|
|
||
|
- DR_Always - the group will record the variable value(s) at every recording cycle. (This is the default).
|
||
|
- DR_Changes - the group will record the variable value(s) only when a particular watched parameter (or parameters) value changes.
|
||
|
- DR_Changes_Step - like DR_Changes, except that a before and after value will be recorded for each variable,
|
||
|
creating a stair step effect (instead of point-to-point) when plotted.
|
||
|
|
||
|
To set the recording frequency call the <tt>set_freq(trick.<frequency_option>)</tt> method of the recording group. For example:
|
||
|
|
||
|
```python
|
||
|
drg.set_freq(trick.DR_Changes)
|
||
|
```
|
||
|
|
||
|
For DR_Changes or DR_Changes_Step, to specify parameter(s) to watch that will control when the variables added with <tt>add_variable</tt> are recorded,
|
||
|
call the <tt>add_change_variable(string)</tt> method of the recording group. For example:
|
||
|
|
||
|
```python
|
||
|
drg.add_change_variable("ball.obj.state.output.velocity[0]")
|
||
|
```
|
||
|
|
||
|
So if we assume the <tt>add_variable</tt> statements from the example in @ref S_7_8_3 "7.8.3" combined with the above <tt>add_change_variable</tt> statement,
|
||
|
then <tt>ball.obj.state.output.position[0]</tt> and <tt>ball.obj.state.output.position[1]</tt> will be recorded only when
|
||
|
<tt>ball.obj.state.output.velocity[0]</tt> changes. Multiple parameters may be watched by adding more change variables, in which case
|
||
|
data will be recorded when any of the watched variable values change.
|
||
|
|
||
|
### Turn Off/On and Record Individual Recording Groups
|
||
|
|
||
|
At any time during the simulation, model code or the input processor can turn on/off individual
|
||
|
recording groups as well as record a single point of data.
|
||
|
|
||
|
```c++
|
||
|
/* C code */
|
||
|
dr_enable_group("<group_name">) ;
|
||
|
dr_disable_group("<group_name">) ;
|
||
|
dr_record_now_group("<group_name">) ;
|
||
|
```
|
||
|
|
||
|
This is the Python input file version:
|
||
|
|
||
|
```python
|
||
|
# Python code
|
||
|
trick.dr_enable_group("<group_name">) ; # same as <group_name>.enable()
|
||
|
trick.dr_disable_group("<group_name">) ; # same as <group_name>.disable()
|
||
|
trick.dr_record_now_group("<group_name">) ;
|
||
|
```
|
||
|
|
||
|
### Changing the thread Data Recording runs on.
|
||
|
|
||
|
To change the thread that the data recording group runs on use the DataRecordGroup::set_thread
|
||
|
method. The thread number follows the same numbering as the child threads in the S_define file.
|
||
|
This must be done before the add_data_record_group function is called. Trick does not
|
||
|
provide data locks for data record groups. It is up to the user to ensure that the data
|
||
|
recorded on *any* thread (including the master) is ready in order for data recording to
|
||
|
record a time homogeneous set of data.
|
||
|
|
||
|
```python
|
||
|
drg.set_thread(<thread_number>)
|
||
|
```
|
||
|
|
||
|
### Changing the Job Class of a Data Record Group
|
||
|
|
||
|
The default job class of a data record group is "data_record". This job class is run after all
|
||
|
of the cyclic job classes have completed. The job class of the data record group can be
|
||
|
changed through the set_job_class method. The data recording job will be added to the end of
|
||
|
the job class queue it is set.
|
||
|
|
||
|
```python
|
||
|
drg.set_job_class(<string class_name>)
|
||
|
```
|
||
|
|
||
|
|
||
|
### Changing the Max File Size of a Data Record Group (Ascii and Binary only)
|
||
|
|
||
|
The default size of a data record is 1 GiB. A new size can be set through the set_max_file_size method. For unlimited size, pass 0.
|
||
|
|
||
|
```python
|
||
|
drg.set_max_file_size(<uint64 file_size_in_bytes>)
|
||
|
```
|
||
|
|
||
|
### Example Data Recording Group
|
||
|
|
||
|
This is an example of a data recording group in the input file
|
||
|
|
||
|
```python
|
||
|
# Data recording HDF5 test
|
||
|
drg0 = trick.DRHDF5("Ball")
|
||
|
drg0.add_variable("ball.obj.state.output.position[0]")
|
||
|
drg0.add_variable("ball.obj.state.output.position[1]")
|
||
|
drg0.add_variable("ball.obj.state.output.velocity[0]")
|
||
|
drg0.add_variable("ball.obj.state.output.velocity[1]")
|
||
|
drg0.add_variable("ball.obj.state.output.acceleration[0]")
|
||
|
drg0.add_variable("ball.obj.state.output.acceleration[1]")
|
||
|
drg0.set_cycle(0.01)
|
||
|
drg0.freq = trick.DR_Always
|
||
|
trick.add_data_record_group(drg0, trick.DR_Buffer)
|
||
|
|
||
|
# This line is to tell python not to free this memory when drg0 goes out of scope
|
||
|
drg0.thisown = 0
|
||
|
```
|
||
|
|
||
|
### User accessible routines
|
||
|
|
||
|
Create a new data recording group:
|
||
|
|
||
|
```c++
|
||
|
Trick::DRAscii::DRAscii(string in_name);
|
||
|
Trick::DRBinary::DRBinary(string in_name);
|
||
|
Trick::DRHDF5::DRHDF5(string in_name);
|
||
|
```
|
||
|
|
||
|
This list of routines is for all recording formats:
|
||
|
|
||
|
```c++
|
||
|
int dr_disable_group( const char * in_name );
|
||
|
int dr_enable_group( const char * in_name );
|
||
|
int dr_record_now_group( const char * in_name );
|
||
|
|
||
|
int Trick::DataRecordGroup::add_variable
|
||
|
int Trick::DataRecordGroup::add_change_variable
|
||
|
int Trick::DataRecordGroup::disable
|
||
|
int Trick::DataRecordGroup::enable
|
||
|
int Trick::DataRecordGroup::set_cycle
|
||
|
int Trick::DataRecordGroup::set_freq
|
||
|
int Trick::DataRecordGroup::set_job_class
|
||
|
int Trick::DataRecordGroup::set_max_buffer_size
|
||
|
|
||
|
```
|
||
|
This list of routines provide file size configuration for Ascii and Binary:
|
||
|
|
||
|
```c++
|
||
|
|
||
|
int set_max_size_record_group (const char * in_name, uint64_t bytes ) ;
|
||
|
int dr_set_max_file_size ( uint64_t bytes ) ;
|
||
|
|
||
|
int Trick::DataRecordGroup::set_max_file_size
|
||
|
|
||
|
```
|
||
|
|
||
|
This list of routines provide some additional configuration for DR_Ascii format only:
|
||
|
|
||
|
```c++
|
||
|
int Trick::DRAscii::set_ascii_double_format
|
||
|
int Trick::DRAscii::set_ascii_float_format
|
||
|
int Trick::DRAscii::set_delimiter
|
||
|
int Trick::DataRecordGroup::set_single_prec_only
|
||
|
```
|
||
|
|
||
|
### DRAscii Recording Format
|
||
|
|
||
|
The DRAscii recording format is a comma separated value file named log_<group_name>.csv. The contents
|
||
|
of this file type are readable by the Trick Data Products packages, ascii editors, and Microsoft Excel.
|
||
|
The format of the file follows. Users are able to change the comma delimiter to another string. Changing
|
||
|
the delimiter will change the file extension from ".csv" to ".txt".
|
||
|
|
||
|
```
|
||
|
name_1 {units_1},name_2 {units_2},etc...
|
||
|
value1,value2,etc...
|
||
|
value1,value2,etc...
|
||
|
value1,value2,etc...
|
||
|
value1,value2,etc...
|
||
|
```
|
||
|
|
||
|
### DRBinary Recording Format
|
||
|
|
||
|
The DRBinary recording format is a Trick simulation specific format. Files written in this format are named
|
||
|
log_<group_name>.trk. The contents of this file type are readable by the Trick Data Products packages from
|
||
|
Trick 07 to the current version. The format of the file follows.
|
||
|
|
||
|
<table>
|
||
|
<tr><th>Value</th><th>Description</th><th>Type</th><th>Bytes</th></tr>
|
||
|
<tr><td colspan=4 align=center>START OF HEADER</td></tr>
|
||
|
<tr><td>Trick-\<vv\>-\<e\></td><td>\<vv\> is trick version, 2 characters (e.g. 07)
|
||
|
\<e\> is endianness, 1 character: L for little endian, B for big endian</td><td>string</td><td>10</td></tr>
|
||
|
<tr><td>\<numparms\></td><td>Number of parameters recorded</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td>17</td><td>parameter \#1 Name string length</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td>sys.exec.out.time</td><td>parameter \#1 Name (1st parameter is always the system time)</td><td>string</td><td>17</td></tr>
|
||
|
<tr><td>1</td><td>parameter \#1 Units Name string length</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td>s</td><td>parameter \#1 Units Name</td><td>string</td><td>1</td></tr>
|
||
|
<tr><td>10</td><td>parameter \#1 Type (see Table 15)</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td>8</td><td>parameter \#1 Size (number of bytes the value occupies)</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td>\<namelen\></td><td>parameter \#2 Name string length</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td>\<name\></td><td>parameter \#2 Name</td><td>string</td><td>\<namelen\></td></tr>
|
||
|
<tr><td>\<unitlen\></td><td>parameter \#2 Units Name string length</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td>\<unit\></td><td>parameter \#2 Units Name</td><td>string</td><td>\<unitlen\></td></tr>
|
||
|
<tr><td>\<type\></td><td>parameter \#2 Type</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td><size></td><td>parameter \#2 Size</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td></td><td>.</td><td></td><td></td></tr>
|
||
|
<tr><td></td><td>.</td><td></td><td></td></tr>
|
||
|
<tr><td></td><td>.</td><td></td><td></td></tr>
|
||
|
<tr><td>\<namelen\></td><td>parameter \#n Name string length</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td>\<name\></td><td>parameter \#n Name</td><td>string</td><td>\<namelen\></td></tr>
|
||
|
<tr><td>\<unitlen\></td><td>parameter \#n Units Name string length</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td>\<unit\></td><td>parameter \#n Units Name</td><td>string</td><td>\<unitlen\></td></tr>
|
||
|
<tr><td>\<type\></td><td>parameter \#n Type</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td><size></td><td>parameter \#n Size</td><td>int</td><td>4</td></tr>
|
||
|
<tr><td colspan=4 align=center>END OF HEADER, START OF RECORDED DATA</td></tr>
|
||
|
<tr><td>\<value\></td><td>parameter \#1 Value</td><td>10</td><td>8</td></tr>
|
||
|
<tr><td>\<value\></td><td>parameter \#2 Value</td><td>\<type\></td><td>\<size\></td></tr>
|
||
|
<tr><td></td><td>.</td><td></td><td></td></tr>
|
||
|
<tr><td></td><td>.</td><td></td><td></td></tr>
|
||
|
<tr><td></td><td>.</td><td></td><td></td></tr>
|
||
|
<tr><td>\<value\></td><td>parameter \#n Value</td><td>\<type\></td><td>\<size\></td></tr>
|
||
|
<tr><td colspan=4 align=center>REPEAT RECORDED DATA FOR EACH CYCLE</td></tr>
|
||
|
<tr><td colspan=4 align=center>END OF RECORDED DATA</td></tr>
|
||
|
</table>
|
||
|
|
||
|
### DRHDF5 Recording Format
|
||
|
|
||
|
HDF5 recording format is an industry conforming HDF5 formatted file. Files written in this format are named
|
||
|
log_<group_name>.hd5. The contents of this file type are readable by the Trick Data Products packages from
|
||
|
Trick 07 to the current version. The contents of the file are binary and is not included here. The HDF5 layout
|
||
|
of the file follows.
|
||
|
|
||
|
```
|
||
|
GROUP "/" {
|
||
|
GROUP "header" {
|
||
|
DATASET "file_names" {
|
||
|
"param_1_file_name", "param_2_file_name", etc...
|
||
|
}
|
||
|
DATASET "param_names" {
|
||
|
"param_1_name", "param_2_name", etc...
|
||
|
}
|
||
|
DATASET "param_types" {
|
||
|
"param_1_type", "param_2_type", etc...
|
||
|
}
|
||
|
DATASET "param_units" {
|
||
|
"param_1_units", "param_2_units", etc...
|
||
|
}
|
||
|
}
|
||
|
DATASET "parameter #1" {
|
||
|
value1 , value2 , value3 , etc...
|
||
|
}
|
||
|
DATASET "parameter #2" {
|
||
|
value1 , value2 , value3 , etc...
|
||
|
}
|
||
|
.
|
||
|
.
|
||
|
.
|
||
|
DATASET "parameter #n" {
|
||
|
value1 , value2 , value3 , etc...
|
||
|
}
|
||
|
}
|
||
|
```
|
||
|
|
||
|
[Continue to Checkpointing](Checkpoints)
|