| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
| |
Don't depend on the 'run' attribute for a load section since
it was just added and old reports do not contain it.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
| |
Return None if there were no jobs, otherwise return the command
line used to start hackbench
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
| |
Add the 'run' attribute for load summary so that appropriate text
can be generated if a load doesn't run
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
| |
If hackbench determines the ratio of available memory to active
cpucores is less than 1GB/core, it will not be run.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
| |
default the sourcedir and buildir values of the load base class
to the internal "../loadsource" and "../build" directories.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
|
| |
Remove system parameter access methods that were moved to the
util.py module and replace callsites with calls into the util
versions.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
| |
refactor some methods from rteval into a commmon utility module
for use by loads and cyclictest.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
| |
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
|
|
|
|
| |
In SMT systems like Power PC, while we turn off or control multithreading,
the processor id in cpu info need not be in sequence, for the available
cores. This patch adds a minor check that if the core id we range, is
really found in cpu info.
Signed-off-by: Gowrishankar <gowrishankar.m@in.ibm.com>
Tested-by: Gowrishankar <gowrishankar.m@in.ibm.com>
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
|
|
| |
The directory called 'rteval-YYYYMMDD-#' and the 'logs' sub-directory
was created in the RtEval constructor. Moved this directory creation to
before the RtEval::measure() function was called to avoid creating empty
directories when using --summarize or --raw-histogram.
Signed-off-by: David Sommerseth <davids@redhat.com>
|
|
|
|
|
|
|
|
|
|
| |
Some computers start with physical_package_id=1, which would return
two CPU sockets with the former CPU socket counting method.
This is fixed by using a more failsafe method, by counting unique
physical_package_id references.
Signed-off-by: David Sommerseth <davids@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Removed support for setting 'thread' workunit
This implementation needs to be re-implemented better, as it
will easily use more than 1024 file descriptors.
- Changed dynmic hackbench loops to a static value
To make the load more comparable no matter how many CPU cores
available. Using 100 as the default value.
- Reduced data size from 128 to 100
This is to match the old hackbench value.
Signed-off-by: David Sommerseth <davids@redhat.com>
|
|
|
|
| |
Signed-off-by: David Sommerseth <davids@redhat.com>
|
|
|
|
|
|
|
|
| |
This is is a feature regression going back to commit e5d01f24e6fb6883
where CPU topology was introduced. The displaying itself was never
updated.
Signed-off-by: David Sommerseth <davids@redhat.com>
|
|
|
|
| |
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
| |
detect when out-of-memory sleep time exceeds 60 seconds
and cap it at that value.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
|
| |
Catch out-of-memory errors and sleep for an increasing number
of seconds (capped at 60 seconds) to let the system recover
before kicking off another hackbench.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
| |
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
| |
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
| |
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
|
|
| |
Added logic to loads.py, hackbench.py and kcompile.py that adjusts
the number of jobs run based on the gigabytes per core ratio. If
the ratio is less than 1, then only do one job per core, else do
two jobs per core.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
| |
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
| |
Add the memory size, number of numa nodes and the run duration
to the parameter array passed to loads and cyclictest.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
| |
Go back to using threads rather than processes to save some
memory.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
| |
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
| |
Changed message size from 1024 to 128 (to reduce memory usage) and
changed work unit from threads to processes.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
| |
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
| |
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
|
| |
In the previous rework of passing in parameters via a dictionary,
I presumed that numcores was passed in as part of the parameter
dictionary. That was incorrect.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
| |
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
|
| |
Added parameters to rteval.conf and changed logic in hackbench.py
to deal with the behavior and new capabilities of the new
hackbench executable we're using.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
| |
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
| |
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
| |
Added an option (-L/--logging) to rteval that will turn on
logging of load output to a file in the report directory.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
|
|
|
|
| |
Put all relevant data for the run in the params section passed to the
loads and to cyclictest. This simplifies argument lists and
centralizes where information may be found.
Signed-off-by: Clark Williams <williams@redhat.com>
|
|
|
|
| |
Signed-off-by: David Sommerseth <davids@redhat.com>
|
|
|
|
|
| |
If rteval is used against an older python-dmidecode, it will raise
an Exception() due to dmidecode.get_warnings() not being available.
|
|
|
|
|
|
|
|
|
|
| |
Added a generic unit-test framework and rewritten the self-test parts in
some rteval modules where this was appropriate.
The unit-test contains a list of modules to load and test. It will run a
function named unit_test() in the imported modules. It is expected that
the unit_test() function will return 0 on success, otherwise it is logged
as an error.
|
| |
|
|\
| |
| |
| |
| |
| | |
Conflicts:
rteval/cyclictest.py
rteval/rteval.py
|
| |
| |
| |
| |
| |
| |
| |
| | |
As cyclictest handles SIGTERM and SIGINT, after sending these signals
rteval can proceed to extract the data from the pipe. A wait() there
would cause rteval to sit and wait forever.
Signed-off-by: Luis Claudio R. Gonçalves <lgoncalv@redhat.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Changes:
* Closed /dev/null after use on each function opening it.
* Make use of subprocess.wait() calls where appropriate.
* Imported the missing SIGKILL symbol in hackbench.py
* Removed the code snippet below from hackbench.py:runload():
count = 30
while count > 0 and p.poll() == None:
time.sleep(1.0)
count -= 1
if p.poll() == None:
os.kill(p.pid, SIGKILL)
Signed-off-by: Luis Claudio R. Gonçalves <lgoncalv@redhat.com>
|
| |
| |
| |
| |
| |
| | |
If SIGINT is received during the measurement phase, it will stop the
load and measurement threads and complete the report, with a shortened
run time.
|
| | |
|
| |
| |
| |
| |
| |
| |
| | |
This is to avoid DMI decode warnings popping up when calling
rteval --help. The other dmi.ProcessWarnings() call as the
RtEval object may be used from another module as well - as v7
does.
|
| | |
|
| |
| |
| |
| |
| | |
This will allow non-root users to use the --summary and --raw-ristogram
features.
|
| |
| |
| |
| |
| |
| |
| | |
This is to make automated tests using rteval behave better, where
rteval runs might go for several hours - and the XML-RPC server might
become available during that time. In addition, rteval will try
several times to send the report as well.
|
| |
| |
| |
| |
| |
| | |
This logic will try up to 5 times to get contact with the given XML-RPC
server before exiting. It will wait attempts * 15 seconds for each
attempt.
|