You are here: Home / Devel Env / Buildbot / Particulars Buildbot

Particulars of the use of buildbot for ABINIT

WARNING: this page is partially outdated
update in progress...

1. Triggering the execution of the automatic tests.

2. How to handle failures ?

3. Accessing directly some slaves.

1. Triggering the execution of the automatic tests

Although the ETSF test farm might be accessed independently of buildbot, the usual mode of interaction with the test farm is through the buildbot system.

In the following, we suppose that you are a well-identified developer of ABINIT, hence, you have your own bzr branches.

There are two ways to trigger the tests on the test farm:

  • a merge request of your branch to trunk/develop will trigger nightly (the bunch starts around 10:30pm European time) the launch of automatic tests on all the "nightly" slaves ;
  • a general "on-demand" Web page on bbportal. You can contact Xavier if you think you are sufficiently expert to use it, provided (of course) you need it for your development.

You will receive by mail the results of the execution of the tests, concerning each nightly slave, one for each slave. Congratulation if you compilation succeeded, all the tests succeeded, etc ! 

2. How to handle failures ?

Let's now analyze what to do in case of failure. The mail from buildbot is rather self-explanatory, and one is quickly lead, through hyperlinks ("further details") to a list of the steps that were executed by buildbot, the failing one being indicated in red. 

If it is step "10 make_core", you have to click on the "10.2 make" or "10.3 stderr" hyperlink, and you will see the usual log/err file produced by the compiler. In case you would like to have direct access on the machine, this is possible, see below.

If it is step "11 newtests", you have to click on the "11.4 summary" hyperlink, to identify whether tests "passed" or "failed", and to identify the test directories where such things happened. Remember : on the reference machine "testf", all tests must "succeed", while on the other machines, no one test can "fail", but a "pass" is OK. More execution information is found in the "11.2 xreport" hyperlink.

In order to correct the test in the other series, use "11.5 All Results" hyperlink to identify precisely the tests that must be corrected.

The links "11.1 stdio" and "11.3 testsbot" are not very often used in the debugging process by standard users, but more by the maintainers.

11.1 stdio     = the stdout of buildbot (Usually NOT useful for debugging)
11.2 xreport   = detailed output of execution
11.3 testbot = output of the configuration use to drive // tests (Usually NOT useful for debugging)
11.4 summary = a short table, summarizing the status (succeeded, passed, failed) of the tests, per directory
11.5 All Results = all results grouped by number of "cpu" (np= 1,2,4,10,...).

In order to correct the test in the "20. abirules" or "21. buildsys" series executed with abiref_gnu_5.3_debug, the "diff" hyperlink will likely be the most useful for you to correct your source code. Perhaps you will need also to complete the information from  "20.4 log" by a look to the "19.2 make" hyperlink. Do not hesitate to contact Xavier Gonze to find some reasonably elegant way to get rid of the "unused variable" or "unused arguments" problems in delicate cases.

3. Accessing directly some slave.


If you have git branches, you are entitled to have interactive access to each of the slaves. Moreover, you can login as "buildbot" user, and place yourself as if you were the one who just made the run reported in the mail that you received. Hence, you are directly placed in an environment ready to debug.
 
  1. Login  on testf slave (well, the first time, contact Jean-Michel Beuken, to allow your machine to connect to the private network).
    ssh testf.pcpm.ucl.ac.be
    [At that stage, you can already work on testf : you might simply download, compile, etc ABINIT in your home dir].
    In addition you can take the hand over buildbot as follows.
  2. Execute :
    sudo su - buildbot
    This will place you in the buildbot directory of testf. If you need to access other machines of the testfarm, use the supplementary step ssh <name_of_the_machine>, where <name_of_the_machine> might be testf, ubu, coba2, etc. e.g.
    ssh abiref
    The first part of the name of the bot is the name_of_the_machine.
  3. cd ABINIT_GIT
  4. cd    to the directory for the particular slave that you want to debug (the machine might support more than one slave) ( ex: cd abiref_gnu_4.7_openmpi )
  5. cd    to the directory for the particular branch that you want to debug (the one bearing your name)  ( ex : cd trunk/develop )
  6. Use the command "module load NAME_OF_BUILDER" to select the compiler that you want to use (type "module avail" to see the list of options).
    Try to solve your problem ... ! You can modify the files, recompile, etc ...
  7. Port your modifications (debugging) to your own branch on your own machine (this must be done on your machine, not on the slave, the latter is not intended to support the commits to your own branch). Also, it is advised to test your modifications on your own machine, before using again the test farm.