X hour run for Tribler

[whirm]

Create a Jenkins job that runs Tribler for N hours, logging all the usual stuff from the experiments. And keep it running in a loop.

[NielsZeilemaker]

Ok, this is how I would do it:

• Create a file in Tribler/Main/vwxGUI similar to tribler_profiler.py, e.g. import the run method of tribler.
• Only before calling run(), spawn a thread which is going to wait for dispersy to be running, e.g. similarly to startTest in TestGuiAsServer in test_as_server.py
• After you got a reference to dispersy, you only need to wait for lm.initComplete, enable the statistics using

dispersy.statistics.enable_debug_statistics(True)

• Schedule the do_log method from script.py. It's probably going to be a bit difficult, maybe move it to another class and let ScenarioScriptBase inherit the method from there.

Then run the new tribler.py file from within the process_monitor, and parse the files as usual afterwards.

[synctext]

@NielsZeilemaker, is your proposed approach reusing stuff from Jenkins FullGUI-test?

So we additionally need a representative workload.
There are several options we can focus on or have multiple results sequentially.
We can measure Tribler with a 1 hour workload of:

• idle and no searches and no channel joins
• do a search in the search community every 1 minute (use tag cloud for terms)
• sync with AllChannel and join the channel with most votes, next join after 5 minutes (5-55min = 9 joins). NOTE: no pollution of the real votes on channels, hopefully we can join without a +1 vote on AllChannels.
• Start a new file download with Libtorrent every 5 minutes (VODO stuff)
• Start a new file download with Libswift every 5 minutes (what content? seed on localhost?)

[NielsZeilemaker]

More or less, it's basically doing your first "workload" the idle and no searches thingy.
If we want to have all these different workloads, I suggest creating a unit_test which uses the GuiAsServer class.

It's maybe even easier this way:

• Copy the test_remote_search.py file
• Modify the startTest method to schedule the do_log method and maybe enable the debug statistics (if we're going to run in optimized mode)
• Run the test from within the process_monitor etc.

Using the GuiAsServer class you can easily create the workloads as described by Johan above. Maybe even call the annotate method to have the lines in the cpu/resource figures showing the different parts of the workload.

[whirm]

A quick update on the status of this one.
We already have the experiment running and collecting data. The reporting part is almost there.

[whirm]

Assigning the issue to @corpaul as he's working on the reporting.

[NielsZeilemaker]

@corpaul lets create daily 1 hour tribler run, simply calling tribler.py from within a systemtab enabled environment should result in enough data.

[corpaul]

We used to have this working but I think the updates over time killed it. Will check with @whirm when he is in

[corpaul]

should be up and running again http:https://jenkins.tribler.org/job/Test_performance_Tribler_idle_1H_nightly_devel so this can be closed