| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
| | |
| | |
| | |
| | | |
avoid looping so much when calculating MAD and variance
|
| | |
| | |
| | |
| | | |
Add methods to retrieve status of services and priorities of kthreads
|
| | | |
|
| | | |
|
| | | |
|
| | |
| | |
| | |
| | |
| | | |
updated option processing, changed stats to use less memory,
changed debug output to prefix with module name
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Rundata::reduce was using a range() statement, which generated a
potentially HUGE list on long runs, many times exhausting memory.
Avoid the range statment and just manually iterate
Also update the cyclictest::debug() method to prefix all output
with cyclictest:
|
| | |
| | |
| | |
| | |
| | | |
updated the rteval::debug method to prefix output with reval: to
differentiate output from other objects.
|
| | |
| | |
| | |
| | |
| | |
| | | |
updated the load::debug method to print the object name as
well as the input string, then updated call sites for this
method in rteval/hackbench.py and rteval/kcompile.py
|
| | |
| | |
| | |
| | |
| | |
| | | |
rearranged order of command line and config file processing so that
command line is processed, then config file and finally command line
parameters override the config file values.
|
| | | |
|
| | |
| | |
| | |
| | | |
updated %changelog section to reflect changes in cyclictest.py
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Calculate statistics directly from histogram (removed the temporary
array which was eating *tons* of memory in Cyclictest::reduce())
Changed sampling interval to 100us (-i100 option to cyclictest)
Added -a option to force affinity for measurement threads
|
| | | |
|
| | |
| | |
| | |
| | |
| | | |
added units to mode parameter
added mean absolute deviation
|
| | |
| | |
| | |
| | | |
__find_config to fail (and allow command line to specify new value)
|
| | |
| | |
| | |
| | |
| | |
| | | |
added code to change xslt_report path when -i specified
added synchronization code with cyclictest to avoid trying
to generate XML before cyclictest has finished stats calcuations
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
- keep data as histogram instead of sample array
- change stats calculation to deal with histogram
- add Mean Absolute Deviation calculation
- print variance
- add synchronization with rteval to prevent race
|
| | | |
|
| | | |
|
| | |
| | |
| | |
| | |
| | | |
This way, the rteval_parserd will also shutdown almost instantly
unless one of the threads are in the progress of parsing a report.
|
| | |
| | |
| | |
| | |
| | |
| | | |
The cyclic_statistics.num_samples and cyclic_histogram.value can in some
situations become incredibly high. This fix should give a little bit more
room for higher numbers. PostgreSQL bigint uses 8 bytes for the storage.
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Both the process_submission_queue() and parserthread() functions
now make use of db_ping() to make sure they have an alive database
connection. And if not, trigger the needed notifications.
If the process_submission_queue() looses the database connection
completely, it will shutdown rteval_parserd immediately.
If the parserthread()s looes the database connection, it will
reduce the activethreads counter. And if this counter reaches 0,
it will signal the main thread which initiates a complete shutdown
of rteval_parserd.
Btw. db_ping() will indeed try to restore the database connection
before presuming the database is unavailable.
|
| | | |
|
| | | |
|
| | | |
|
| | | |
|
| | | |
|
| | |
| | |
| | |
| | |
| | | |
And fixed an error when the program exits before POSIX MQ queue
is opened and initialised.
|
| | | |
|
| | | |
|
| | | |
|
| | | |
|
| | |
| | |
| | |
| | |
| | | |
Makes it easier to implement log parameters from argument parsing
or config file
|
| | | |
|
| | | |
|
| | | |
|
| | |
| | |
| | |
| | |
| | | |
The memory region needed for the dbconn structure was not released
in db_disconnect(). This leak was introduced in commit 9526e7d8844a47fa
|
| | | |
|
| | | |
|
| | | |
|
| | | |
|
| | | |
|
| | | |
|
| | | |
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
This introduces async waiting for report parsing. Commit
2584a3c36c97c757dc80108b898eede52b91dc44 introduced sending
a NOTIFY whenever an INSERT is done into the submissionqueue table.
With the db_wait_notification() the process_submission_queue() function
will sleep until a a notification is received.
The LISTEN, UNLISTEN and NOTIFY SQL commands are PostgreSQL dependent.
Other database implementations may use what that database supports, or
just simply do a sleep() to change the behaviour to a polling model.
|
| | |
| | |
| | |
| | |
| | | |
This new field references submissionqueue.submid, to keep a link
between parsed rtevalruns and the original submission.
|
| | |
| | |
| | |
| | | |
Also added missing error check of the result of the function call
|
| | |
| | |
| | |
| | |
| | | |
This will NOTIFY all clients listening to 'rteval_submq', to trigger
processing of this table after INSERTs have completed.
|
| | | |
|