Dumps a paje trace file in a csv-like textual format
pj_dump [OPTIONS] [FILE]
The pj_dump(1) command translates the paje trace file FILE to a CSV-like textual format (described below). It is a useful program to analyze the behavior of parallel and distributed applications that were traced using some library that generates trace files in the Paje file format. Once you dump the contents of the Paje trace file in a CSV-like manner, you are free to analyze the contents of the trace the way you want. You can use R for example to draw scatter plots and gantt charts. If FILE is not provided, then the standard input is used.
By default, pj_dump will read the trace file from the beginning until the end of file is found. During this process, pj_dump relies on the Paje library to recreate in memory the behavior that is registered in the trace files. This means that pj_dump will put in memory all the contents of the trace file, even if the input is very large. Once all the contents of the trace file are simulated, pj_dump dumps the information in the CSV-like textual format described below in the OUTPUT DESCRIPTION section.
You can change the default behavior of pj_dump by providing the parameters --start=START and --end=END where START and END are valid timestamps of the input trace. If provided, it dumps only the contents of the trace between START and END. Note that even if used, pj_dump will simulate the whole trace file to keep the same semantic of behavior. Another way to change the default behavior is through the --stop-at=TIME parameter. If provided, pj_dump will read the trace file up to timestamp TIME (considering that the trace file is completely time ordered) and dumps what has been simulated until then. The --no-strict switch should be avoided and can be used only with old Paje trace files with old field names in event definitions. The --ignore-incomplete-links switch make pj_dump ignore incomplete links silently. More details on this switch below, in the OUTPUT DESCRIPTION section.
pj_dump accepts the following options:
-a, --stop-at=TIME
Stop the trace simulation at TIME.
-s, --start=START
Dump starts at timestamp START (instead of timestamp 0).
-e, --end=END
Dump ends timestamp END (instead of End Of File).
-n, --no-strict
Support old field names in event definitions.
-z, --ignore-incomplete-links
Ignore incomplete links without warnings.
-?, --help
Show all the available options.
--usage
Give a short usage message.
The pj_dump(1) command expects an input that follows the Paje file format (as described in the PDF document listed in the RESOURCES section of this page). If FILE is not provided, pj_dump(1) will try to read from the standard input.
It\(cqs easier to understand what is written here if you are acquainted to the Paje terminology (Container, State, Variable, Link, Event and the information attached to each of these). Take a look to the description of the Paje File Format (link below in the RESOURCES section) for further details.
The contents of the lines generated by the pj_dump(1) command are separated by commas, defining the columns. So, a line like this:
Container, 0, LINK, 0, 4.48514, 4.48514, 9
has seven columns. The first column is always one of: Container, State, Variable, Event or Link. The remaining columns of the line have specific information depending on the first column. Here\(cqs a synthetic description of the five different types of lines you\(cqll find as output of pj_dump(1):
Container, parentContainer, containerType, startTime, endTime, duration, name State, container, stateType, startTime, endTime, duration, imbrication, value Variable, container, variableType, startTime, endTime, duration, value Event, container, eventType, time, value Link, container, linkType, startTime, endTime, duration, value, startContainer, endContainer
See below a detailed description with examples for each of them.
All lines starting with Container look like this:
Container, 0, HOST, 0, 4.48514, 4.48514, Tremblay
"Container"
"0" - The name of the parent container
"HOST" - The type of this container
"0" - The starting time
"4.48514" - The finish time
"4.48514" -The duration
"Tremblay" - The name of this container
All lines starting with State look like this:
State, node48, SERVICE, 691, 692, 1, 0, booked
"State"
"node48" - The name of the container
"SERVICE" - The type of this state
"691" - The starting time
"692" - The finish time
"1" - The duration
"0" - The imbrication level
"booked" - The value of the state
All lines starting with Variable look like this:
Variable, Tremblay, pcompute, 2.15357, 2.17013, 0.016554, 9.8095e+07
"Variable"
"Tremblay" - The name of the container
"pcompute" - The name of the variable
"2.15357" - The starting time
"2.17013" - The ending time
"0.016554" - The duration
"9.8095e+07" - The value of the variable
All lines starting with Event look like this:
Event, Tremblay, msmark, 3.4286, finish_send_tasks
"Event"
"Tremblay" - The name of the container
"msmark" - The name of the event
"3.4286" - The instant in time when this event took place
"finish_send_tasks" - The value of the event
All lines starting with Link look like this:
Link, 0, 0-HOST1-LINK4, 0, 0, 0, G, Tremblay, 9
"Link"
"0" - The name of the container
"0-HOST1-LINK4" - The type of this link
"0" - The starting time
"0" - The ending time
"0" - The duration
"G" - The value of this link
"Tremblay" - The starting container
"9" - The ending container
Incomplete Links
According to the description of the Paje File Format, a link is formed by two events: PajeStartLink and PajeEndLink. These events are matched by the Paje Simulator using a key that is provided in the trace file. If one of these two events are missing for some arbitrary reason and the trace file ends (or the container is destroyed), you\(cqll have a simulation with incomplete links. Generally, the Paje Simulator, and by consequence pj_dump, consider these links as errors, and list them in the following manner:
$ pj_dump ~/tracefile.paje List of incomplete links in container '0': Link, 0, MSG_PROCESS_TASK_LINK, 0, -1, 0, SR, broadcaster-12, NULL Link, 0, MSG_PROCESS_TASK_LINK, 0.00013, -1, 0, SR, broadcaster-13, NULL Link, 0, MSG_PROCESS_TASK_LINK, 0.002868, -1, 0, SR, broadcaster-13, NULL (...) PajeLinkException: Incomplete links at the end of container with name '0'
The best action when this happens is to fix the tracer or the converter that generated the trace, since it indicates probably some error during the execution. If you think that this error is acceptable, you can provide the -z switch to pj_dump to tell the Paje Simulator to ignore incomplete links. All the trace file will be dumped and all errors concerning incomplete links will be silently ignored. Use with caution.
Description of the Paje trace file: http://paje.sourceforge.net/download/publication/lang-paje.pdf
Main web site: http://github.com/schnorr/pajeng/
Report pj_dump bugs to <http://github.com/schnorr/pajeng/issues>
Copyright (C) 2012-2014 Lucas M. Schnorr. Free use of this software is granted under the terms of the GNU General Public License (GPL).