pegasus-keg — kanonical executable for grids
pegasus-keg [-a appname] [-t interval |-T interval] [-l logname]
[-P prefix] [-o fn [..]] [-i fn [..]] [-G sz [..]] [-m memory]
[-C] [-e env [..]] [-p parm [..]] [-u data_unit]
The kanonical executable is a stand-in for regular binaries in a DAG - but not for their arguments. It allows to trace the shape of the execution of a DAG, and thus is an aid to debugging DAG related issues.
Key feature of pegasus-keg is that it can copy any number of input files, including the generator case, to any number of output files, including the datasink case. In addition, it protocols the IPv4 and hostname of the host it ran upon, the current timestamp, and the run time from start til the point of logging the information, the current working directory and some information on the system environment. pegasus-keg will also report all input files, the current output files and any requested string and environment value.
The workflow of the Keg tool is as follows: - if -m - allocate a memory buffer of the specified amount - if -i - read all input files into the memory buffer - if -o - write either the input files content (or a generated content if -G) to output files - if -T - generate CPU load for the specified time period decreased by the time period spent on IO stuff; if the IO stuff time period exceeds the time period specified here the program exits with code status 3 - if -t - wait/sleep for the specified time period decreased by time periods spent on IO stuff (and CPU load generating if any); if the time period spent on previous activities exceeds the amount specified here the program exits with code status 3 - if -l - write info to the specified log file.
The -e, -i, -o, -p and -G arguments allow lists with arbitrary number of arguments. These options may also occur repeatedly on the command line. The file options may be provided with the special filename - to indicate stdout in append mode for writing, or stdin for reading. The -a, -l , -P , -T and -t arguments should only occur a single time with a single argument.
If pegasus-keg is called without any arguments, it will display its usage and exit with success.
- -a appname
- This option allows pegasus-keg to display a different name as its applications. This mode of operation is useful in make-believe mode. The default is the basename of argv.
- -e env [..]
- This option names any number of environment variables, whose value should be reported as part of the data dump. By default, no environment variables are reported.
- -i infile [..]
- The pegasus-keg binary can work on any number of input files. For each output file, every input file will be opened, and its content copied to the output file. Textual input files are assumed. Each input line is indented by two spaces. The input file content is bracketed between an start and end section, see below. By default, pegasus-keg operates in generator mode.
- -l logfile
- The logfile is the name of a file to append atomically the self-info, see below. The atomic write guarantees that the multi-line information will not interleave with other processes that simultaneously write to the same file. The default is not to use any log file.
- -o outfile [..]
- The pegasus-keg can work on any number of output files. For each output file, every input file will be opened, and its content copied to the output file. Textual input files are assumed. Each input line is indented by two spaces. The input file content is bracketed between an start and end section, see 2nd example. After all input files are copied, the data dump from this instance of pegasus-keg is appended to the output file. Without output files, pegasus-keg operates in data sink mode. Accept also <filename>=<filesize><data_unit> form, where <data_unit> is a character supported by the -u switch.
- -G size [..]
- If you want pegasus-keg to generate a lot of output, the generator option will do that for you. Just specify how much, in bytes (but you can change it with -u switch), you want. You can specify more than 1 value here if you specify more than 1 output file. Subsequent values specified here will correspond to sizes of subsequent output files. This option is off by default.
- -u data_unit
- By default, the output data generator (the -G switch) generates the specified amount of data in Bytes. You can alter this behavior with this switch. It accepts one of the following characters as data_unit value: B for Bytes, K for KiloBytes, M for MegaBytes, and G for GigaBytes.
- This option causes pegasus-keg to list all environment variables that start with the prefix \_CONDOR The option is useful, if .B pegasus-keg is run as (part of) a Condor job. This option is off by default.
- -p string [..]
- Any number of parameters can be reported, without being specific on their content. Effectively, these strings are copied straight from the command line. By default, no extra arguments are shown.
- -P prefix
- Each line from every input file is indented with a prefix string to visually emphasize the provenance of an input files through multiple instances of pegasus-keg. By default, two spaces are used as prefix string.
- -t interval
- The interval is an amount of sleep time that the pegasus-keg executable is to sleep in seconds. This can be used to emulate light work without straining the pool resources. If used together with the -T spin option, the sleep interval comes before the spin interval. The default is no sleep time.
- -T interval
- The interval is an amount of busy spin time that the pegasus-keg executable is to simulate intense computation in seconds. The simulation is done by random julia set calculations. This option can be used to emulate an intense work to strain pool resources. If used together with the -t sleep option, the sleep interval comes before the spin interval. The default is no spin time.
- -m memory
- The amount of memory ([MB]) the Keg process should use. This option can be used to emulated application’s memory requirements. The default is not to allocate anything.
Execution as planned will return 0. The failure to open an input file will return 1, the failure to open an output file, including the log file, will return with exit code 2. If the time spent on IO exceeds the specified time CPU load period with -T or the time spent on IO and CPU load exceeds the specified wall time with -T the return code will be 3.
The example shows the bracketing of an input file, and the copy produced on the output file. For illustration purposes, the output file is connected to stdout :
$ date > xx $ pegasus-keg -i xx -p a b c -o - --- start xx ---- Thu May 5 10:55:45 PDT 2011 --- final xx ---- Timestamp Today: 20110505T105552.910-07:00 (1304618152.910;0.000) Applicationname: pegasus-keg [3661M] @ 128.9.xxx.xxx (xxx.isi.edu) Current Workdir: /opt/pegasus/default/bin/pegasus-keg Systemenvironm.: x86_64-Linux 2.6.18-238.9.1.el5 Processor Info.: 4 x Intel(R) Core(TM) i5 CPU 750 @ 2.67GHz @ 2660.068 Load Averages : 0.298 0.135 0.104 Memory Usage MB: 11970 total, 8089 free, 0 shared, 695 buffered Swap Usage MB: 12299 total, 12299 free Filesystem Info: / ext3 62GB total, 20GB avail Filesystem Info: /lfs/balefire ext4 1694GB total, 1485GB avail Filesystem Info: /boot ext2 493MB total, 447MB avail Output Filename: - Input Filenames: xx Other Arguments: a b c
The input file must be textual files. The behaviour with binary files is unspecified.
The host address is determined from the primary interface. If there is no active interface besides loopback, the host address will default to 0.0.0.0. If the host address is within a virtual private network address range, only (VPN) will be displayed as hostname, and no reverse address lookup will be attempted.
The processor info line is only available on Linux systems. The line will be missing on other operating systems. Its information is assuming symmetrical multi processing, reflecting the CPU name and speed of the last CPU available in /dev/cpuinfo .
There is a limit of 4 * page size to the output buffer of things that .B pegasus-keg can report in its self-info dump. There is no such restriction on the input to output file copy.
Jens-S. Vöckler <voeckler at isi dot edu>
Pegasus - http://pegasus.isi.edu/