Age | Commit message (Collapse) | Author | Files | Lines |
|
expose in global_config
Confusingly, timeouts for the same thing are set differently in multiple
places. Also, many of the host timeouts are far too long for Chrome OS.
As such, I've unified a couple of the host timeouts and exposed the
default values through global config.
Signed-off-by: Dale Curtis <dalecurtis@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@5434 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
global_config
The four hour timeout for crash collection is way too long for Chrome
OS, so we moved this parameter into global config for easier control.
Signed-off-by: Dale Curtis <dalecurtis@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@5433 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
There have been several cases where we need to modify the SSH command
line for Chrome OS. This patch adds site extensibility to the
make_ssh_command function in abstract_ssh.
Signed-off-by: Dale Curtis <dalecurtis@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@5432 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@5379 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Fix unit test for server/autotest.py broke in r5317.
git-svn-id: svn://test.kernel.org/autotest/trunk@5323 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
I need to override the logic to read fetch_location from global_config with
site_autotest implementation. So I refactored it out to its own function
inside server/autotest.py.
Signed-off-by: Eric Li <ericli@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@5317 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Since its introduction on r2440, server/frontend.py has a call to
generate_control_file, present on frontend/afe/rpc_interface.py.
This method apparently never had a do_push_packages parameter,
but for some reason we were passing it in case we provide a specific
kernel with -k. Remove all references to do_push_packages.
With this, bug #47 is fixed, and it is possible to specify -k through
the cli.
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@5287 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Jongki Suwandi <jsuwandi@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@5257 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Jongki Suwandi <jsuwandi@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@5233 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
the default for my local emacs configuration).
Signed-off-by: David Rochberg (rochberg@chromium.org)
BUG=chromiumos:7716
TEST=stty tostop, then run with and without this change. Without the change, tests will hang in a SIGTTOU/syscall-restarted loop.
Change-Id: Ib5317e238b68089d71e8e57418717e7553fd64d9
Review URL: http://codereview.chromium.org/6368022
git-svn-id: svn://test.kernel.org/autotest/trunk@5216 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
I got confused by the error messages printed by autoserv when I tried to
use it, so I tried to improve it a little bit.
Instead of printing error messages to stdout (and sometimes simply
dumping help text without any explanation), use the OptionParse.error()
method that aborts the program automatically after printing a short
command-line help text and an error message.
Signed-off-by: Eduardo Habkost <ehabkost@raisama.net>
git-svn-id: svn://test.kernel.org/autotest/trunk@5067 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
The current logging code assumes that 'root' can be used to log
into the client machine, which isn't always the case. This patch
makes the logger user the user and port settings saved in the
host object.
This patch also removes the used-exactly-once run_cmd_on_host() function
and inlined it since it doesn't seem like there is any benefit to splitting
it out. It instead adds a _make_ssh_cmd() method to the abstract_ssh
class which respects the username and port settings.
It also eliminates the redirection of stderr from the subprocess.Popen()
call. If the command fails, then I want to see the output appear in
the log.
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
git-svn-id: svn://test.kernel.org/autotest/trunk@5011 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
When using an unprivileged account on a client, need to make sure that
the autotest directory is actually writable before trying to use it.
Otherwise the test will fail.
This problem was encountered when trying to use an unprivileged
account on the same machine as the autotest server (to use it for
running builds). Without this patch, autotest would try to delete the
server files while installing the needed client side files in
/usr/local/autotest.
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
git-svn-id: svn://test.kernel.org/autotest/trunk@5010 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
When connecting to an unprivileged account on a client, /var/lib will
not be writable. Store the snapshot of /var/log/messages in /var/tmp
instead which is far more likely to be writable.
v2:
- Moved storage location from /tmp to /var/tmp so that it survives over
reboot
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
git-svn-id: svn://test.kernel.org/autotest/trunk@5009 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
This patch simplifies the implementation of base_utils.parse_machine()
and adds test cases to make sure it is parsing strings correctly.
v2:
- Reordered with 'server: user parse_machine directly inside
create_host()' to avoid creating an non-bisectable commit.
- modified code to use the form "if '@' in string:" which is clearer
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
git-svn-id: svn://test.kernel.org/autotest/trunk@5008 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Currently, parse_machine() is called explicitly by the
control_segments hooks before calling create_host, but is not used by
anything else which means that a machine that is specified in
[user[:pass]@]host[:port] format will not get the username, password
and port preferences applied consistently (the hooks will use it
correctly, but other connections will not).
This patch moves the call to parse_machine directly inside create_host
so that all references to clients can also specify a
username/port/password (for instance, in the server's list of hosts).
At the same time, reorder the arguments so it emits the values in the
same order that they are passed.
v2:
- reordered with 'server: simplify parse_machine ...'
- fixed non-bisectable commit by fixing return result order
- cleaned up spacing around '=' in keyword arguments
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
git-svn-id: svn://test.kernel.org/autotest/trunk@5007 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Jean-Marc Eurin <jmeurin@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@5000 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Jean-Marc Eurin <jmeurin@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4965 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Cleaning up and using the base code for the host.
In the profiler code, the hostid should be a hostname.
Added some client barrier tests and unittests.
Signed-off-by: Jean-Marc Eurin <jmeurin@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4964 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
We found the previous patch only fixed a part of our problem. We just found
there are occurrences inside server test control files which called
hosts.create_host(hostname) without properly passed in ssh options. And I do
not want to touch the test control files at all.
So here is a solution, my previous changes to SSHHost is reverted in this
patch, since it is not necessary any more.
I tested this change with both client and server side sleeptest against a
remote VM ( on port other than 22).
Signed-off-by: Eric Li <ericli@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4944 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
When we started using autotest to run test against a VM based chromium os,
we found autotest was not able to handle ssh port other than 22, autoserv
--ssh-port did not work as it suppose to.
It turned our the ssh options were not properly passed into SSHHost class.
Signed-off-by: Eric Li <ericli@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4939 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Adds a barriertest_2client server side test that tests barrier
functionality between two client tests without any server involvement.
Signed-off-by: Gregory Smith <gps@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4919 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Jean-Marc Eurin <jmeurin@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4908 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Once again I went through the autotest tree with my automated script to
check all autotest source, and discovered several places that needed
permission and indentation (including trailing whitespace) fixes. I have
checked the generated patch and indeed it looks OK. Will commit this
shortly, as preparation for autotest 0.13.
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4907 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Gregory Smith <gps@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4872 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Make it possible to specify a builddir for the install() method
of the GitKernel class.
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4847 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
So people don't have to actually provide the weburl in
control files.
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4845 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Turns out it was wrong.
git-svn-id: svn://test.kernel.org/autotest/trunk@4844 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
This patch fixes a long, long standing bug (in fact, I
started to dig and gave up when I looked the last 5
changes to the file, dating 8 months ago): The method
use_rsync(), used to verify whether autotest should
be copied to the client using scp or rsync does not
return in case we detect that the client machine does
have rsync installed, so the first use_rsync() check
will allways return None. Subsequent checks will
return True because during the first run an attribute
will be set to True, making the function return properly.
Bottom line, we will *allways* end up copying all the
client code to the client machine using scp instead of rsync,
which should be way faster after the 1st time the client
got copied to the machine. We probably didn't notice this
because:
1) Most of the time we'll run with the key
[PACKAGES]
serve_packages_from_autoserv: True
Which makes the amount of data being copied way smaller,
so we won't bother
2) Other times we'd be installing the client using the
packaging system.
Anyway, it's a bug, should be fixed and will probably
help people using autoserv to install autotest in remote
machines (as I was doing this week :))
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4820 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
The original code of git_kernel.py was attributing an
incorrect path to the source code directory in the
client. Fixing that.
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4811 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
During the change made on r4792, I (lmr) forgot to
change the call to the git parent class to use super().
Let's fix that.
Signed-off-by: Frank Becker <fb@alien8.de>
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4810 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
(,) within labels properly.
Chromium OS autotest project had assigned labels with comma to test hosts
and this situation makes the current host_keyvals file impossible to parse
out correct label values. This change will fix/enhance that.
git-svn-id: svn://test.kernel.org/autotest/trunk@4793 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Introduce a library with classes to abstract interaction
with version control system repositories. Right now, there is
only GitRepo implemented. Made server/git.py to use the
common_lib implementation (the server git class represents
the install of a git repo in a host).
Now, test writers and can use a shared API to check out and
update git code conveniently.
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4792 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Fix for my last patch (r4780). Thanks to John Admanski
<jadmanski@google.com>.
Signed-off-by: Frank Becker <fb@alien8.de>
git-svn-id: svn://test.kernel.org/autotest/trunk@4786 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
when it is raised within the server control file and not inside of a job.
Currently these exceptions are only logged into the debug/ logs and not into
the status log (and thus not into the database).
It adds support by logging an INFO message with a special job_abort_reason
field when a server job fails with an exception, and the parser then
special-cases these info lines (much like it already does with kernel version
INFO lines) and updates the SERVER_JOB entry.
Unfortunately this can't easily be done with a "normal" status log entry since
when a failure occurs outside of a test the special SERVER_JOB entry is the
only place where this reason information actually belongs, but the parser has
no existing mechanism for modifying this entry (since it's automatically
generated by the parser, rather than being generated in response to a specific
log). Adding this (relatively small) special case seemed more practical than
either adding yet another special entry or trying to perform a major
restructuring of the parser.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4785 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
This patch does some cleaning up on the git checkout class:
1 - Used the logging module
2 - Made docstrings comply with coding standards
Next steps on the work of getting the git class suitable
for common autotest usage are:
* Decouple interaction with git repo from the concept of
host (which is only available on the autotest server)
* Move the libraries to the common lib
That will be made on a separate patchset.
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4782 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
It is not necessary to resort to the git web frontend
to verify whether the local repo is outdated with regards
to origin. This patch simplifies the methods using built
in git commands, thus eliminating the need of a web URL.
The interface was maintainted, and now people who
still specify a web URL on their control file will have
a deprecationwarning message on their logs.
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4781 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
I need to checkout kernel revisions by branch, commit id, or tag before
building and testing them.
Here is my patch to extend server/git.py and server/git_kernel.py
I'm very open for any kind of suggestions.
The following new methods exist in GitKernel()
* checkout - Check out a certain branch, tag or revision. Optional
a local branch name can be given for -b
* show_branch - print the current local branch
* show_branches - print the local and/or remote branches
* show_revision - print the current tip commit id (checked out revision)
* extended install - optional a commit id / revision that is checked out
* before patching, configure, and build
Therefore the following methods were added to Git()
* get_revision() - To print the current tip commit id (aka revision).
* checkout() - Check out a certain branch, tag or revision.
* get_branch() - return the branches depending on options
Risk: Low (Just extends functionality, doesn't change existing one.)
Visibility: Users who build Linux kernels may profit.
Signed-off-by: Frank Becker <fb@alien8.de>
git-svn-id: svn://test.kernel.org/autotest/trunk@4780 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Adding a flag to allow halting before powercycling.
Signed-off-by: Fa Yoeu <fayoeu@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4757 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
If we can't install the client using the autotest internal
package management system, give the user a less intimidating
message, after all we'll fall back to use other methods to
install the client.
Signed-off-by: Lucas Meneghel Rodrigues <lmr@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4749 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
it a little easier to override more reliably. Specifically:
- Switch the partition imports from execfile back to an import * style of loading, ultimately import seems to be the most reliable and portable
- Clean up the base+site definitions in autotest.py to make Site overrides more consistent
- Drop the explicit failure if the Autotest.source_material is a file and not a directory, there's no reason to outright fail and it just introduces some unnecessary fragility
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4736 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
test with just host.run_test, instead of having to setup an autotest
client directly and generate a control file.
Note that since server-client job state is managed automatically, this also
supports profilers out of the box. If you do a job.profilers.add before calling
host.run_test then the client test will automatically be run with the
appropriate profilers.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4725 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Darren Kuo <darrenkuo@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4681 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4632 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4631 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
after the server_job.record refactoring.
The original record implementation just dumped whatever status logs
the client sent it into the status.log, so if the client died on you
there was never a problem with the client leaving the server-side
logging in a funky state. Now that client job records are being
properly routed though the server-side job.record a client death can
leave the server logging in a broken state.
This implementation adds a special get_record_context method which can
be used by the autotest.py code to retrieve the current job.record
context at the start of the client job, and then restore it at the end
so that the server side logging is put back into a known-good state.
It also removes a bunch of self._record_prefix uses in the server_job
code which were rendered useless by the job.record changes, but
weren't removed.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4606 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
record.
The two main features that the server job code added on top of the
basic job record were the ability to inject client job records into
the server job record, and the ability to inject console/vlm warnings
into the record.
The Autotest injection was implemented by adding a parse classmethod
to the status_log_entry class that does the inverse of the render
method, so that the Autotest class can construct status_log_entries
from the raw text status logs it gets from the client. The server then
re-records the messages, rather than using a low-level method to
bypass the old job.record rendering.
The warning injection was implemented by using the record_hook; to
make it work better it was moved to be called before the entry is
logged, not after. This required a slightly more involved hook, as it
was useful to be able to call job.record directly from the hook and so
the hook was wrapped in a callable class that prevents recursion.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4605 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
ABORT error.
Signed-off-by: Gregory Smith <gps@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4600 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
frontend.AFE.run_test_suites().
Signed-off-by: Mihai Rusu <dizzy@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4578 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Josh Gilkerson <jwg@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@4566 592f7852-d20e-0410-864c-8624ca9c26a4
|