| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Add a warning if parallel_simple is passed an empty list of machines.
Signed-off-by: Scott Zawalski <scottz@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3314 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
object and recreate the temporary output dir if it notices they are
missing (say if the test code removes them, they are allowed to do
that). Also changed it to uninstall the sysinfo Autotest client and
remove the contents of the outputdir after each sysinfo step if the test
has set to True a flag attribute named 'disable_sysinfo_install_cache'.
Signed-off-by: Mihai Rusu <dizzy@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3311 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
as it is for clients
Signed-off-by: Martin Bligh <mbligh@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3310 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
instead of sys.std*. this involved refactoring the code out of ssh_host/paramiko_host/abstrash_ssh, which previously did the same thing (they still do, but now the use the common logic in utils)
* change packages.py to suppress command output properly. it was using the verbose option previously, which was added to ssh_host.run() and paramiko_host.run() specifically for this call site. in this context, verbose=False meant "dont print the command and dont print the output either). but this doesn't make the interface of utils.run(), for which verbose=False just means "dont print the command", because utils.run() suppresses output by default (while *host.run() doesn't). to resolve that, i made verbose on *host.run() match the interface of utils.run(), and i made packages.py suppress the output properly by passing stdout_tee=None.
* move tee.flush() calls in utils.BgJob code to where it makes more sense (and interacts with LoggingFiles better)
Signed-off-by: Steve Howard <showard@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3307 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
they we introduced long ago, while stderr of commands was redirected to stdout, making noise relatively harmless. but i don't think they've served a very significant purpose, and now they're just noise on stderr, which we'd like to promote to ERROR level logging.
Signed-off-by: Steve Howard <showard@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3305 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
--verbose sets the console to DEBUG level output (this is added to autotest client as well)
--no_console_prefix disables the logging prefix entirely on the console
Signed-off-by: Steve Howard <showard@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3301 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
* client.common_lib.debug
* abstract_ssh.LoggerFile
Changed {SSHHost,ParamikoHost}.run() default args for std{out,err}_tee to a constant instead of None. This preserves the existing behavior (teeing to the logging module by default), but allows callers to explicitly avoid this teeing if desired, by passing stdout_tee=None.
Signed-off-by: Steve Howard <showard@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3293 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
cleaning up verbose logs in general
Signed-off-by: Scott Zawalski <scottz@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3290 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
reduce the number of RPCs to the server significantly
Signed-off-by: Martin Bligh <mbligh@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3286 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Gregory Smith <gps@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3285 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
test kernel install stanza when >4 machines are supplied. This allows
for large multi-machine tests to proceed even if some of the hosts die
while failing to install or boot the kernel.
This is not the ideal implementation. We should really have some way
to get the max(sync_count) required by a control file's step_test()
before calling the function. Ideas welcome.
Signed-off-by: Gregory Smith <gps@google.com>
(this patch breaks a unittest. the fix is the patch i am mailing immediately after this, sorry. 2 mails is easier than merging them into one patch before mailing for me)
git-svn-id: svn://test.kernel.org/autotest/trunk@3284 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Gregory Smith <gps@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3271 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
debug dir. this is quite a bit more involved than the server-side change because of some buggy directory clearing code and various places that depended on the old configuration.
This also changes the TKO autoserv log viewers to point to the new .DEBUG and .ERROR autoserv logs.
Signed-off-by: Steve Howard <showard@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3266 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
CRITICAL -- going up to ERROR should suffice).
Signed-off-by: Steve Howard <showard@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3265 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
other autotest imports.
Signed-off-by: Steve Howard <showard@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3253 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
like to split them into separate changes, but the change in common_lib.test means it all has to change at once.
this replaces the old ini-style logging configurations with code-based configurations. i know the ini files were easier to read, but we're doing a lot more fancy stuff with logging and these give us much more power and less duplication. as one example, we'll probably move very soon to having separate .DEBUG, .INFO, .WARNING, and .ERROR logs. This will allow us to make a centralized change to make that happen, rather than duplicating a whole bunch of information across .ini files, and it'll make the information much more concise.
Risk: Gravely high
Signed-off-by: Steve Howard <showard@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3243 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
code that raised them to use these new exceptions. Completely reworked
the 3 repair phases to try to fix things by handling the specific errors
generated by verify (in a loop where it exits either by not being able
to fix an issue or if the last verify ran successfully). Reorganized
some code to support this new way to repair hosts.
The code tries to remember each repair handler it used to catch the case
where even after executing the handler the same verify failure is
reported and it may execute the repair handler in an infinite loop.
When it finds out that the same repair handler has already been executed
it will re-raise the original exception hoping it will be cought in
surrounding try/except contexts or the exception will make repair fail.
Signed-off-by: Mihai Rusu <dizzy@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3217 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
verify and repair tasks, as well as normal jobs.
Risk: Low
Visibility: Should avoid leaving host garbage behind after repair and
verify jobs.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3215 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
the hosts it uses. This should keep it from launching a bunch of
spurious warning monitor processes.
Risk: Low
Visibility: Eliminate some pointless warning monitor usage.
Signed-off-by: John Admanski <jadmanski@google.com>
Autotest@test.kernel.org
http://test.kernel.org/cgi-bin/mailman/listinfo/autotest
git-svn-id: svn://test.kernel.org/autotest/trunk@3214 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
rather than a separate one per host
Signed-off-by: Martin Bligh <mbligh@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3212 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
early on, since the results dir for it won't exist yet. Make sure we
create it if not.
Also, increase the timeout when waiting for the profiler to be ready
from 30 seconds to three minutes, since taking 30-60 seconds to go
from exec to "profiler is ready to run" is not unreasonable,
particularly if the profiler is being retrieved via the packaging
system.
Risk: Low
Visibility: More robust failure log collection added, and make the
profiler.ready wait less racy.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3211 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
called directly
Signed-off-by: Martin Bligh <mbligh@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3210 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
we make things a little more efficient - I'll try to rewrite
the poll loop later today
Signed-off-by: Martin Bligh <mbligh@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3208 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
and temporary work dir values and reuse them across the various sysinfo
steps. Still supports only a single host per sysinfo object instance
(which means it works with multiple machines if one uses
job.parallel_simple and such to run their server side test, it does not
work with server side tests that don't do that and work directly with
all the machines in their test class code).
Signed-off-by: Mihai Rusu <dizzy@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3207 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
if it fails for some reason we try to get the client.log from the
profiler. Otherwise there's no good way to determine why the profiler
failed, since we run it in temp dirs that get cleaned up by the end
of the job (and so nothing is available for a post-mortem).
Risk: Medium
Visibility: Adds some log collection during server profiler failures.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3204 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
AutoservHardwareRepairRequestedError but let the implementation decide
how to handle hardware repair (either by just requesting and raising
that exception or wait for the machine to be repaired and return
normally).
Signed-off-by: Mihai Rusu <dizzy@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3203 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
as a test, so we can see the kernel qual results more easily
Signed-off-by: Martin Bligh <mbligh@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3202 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Risk: Low
Visibility: Add a generic os.path.exists equivalent to Host
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3200 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
are often passed in as parameters to tests the repr will often
show up in test keyvals (where we log all params).
Risk: Low
Visibility: Eliminates the (generally unhelpful) default repr for
host objects.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3168 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Risk: Low
Visibility: Refactoring only.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3167 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
to make it easier to see what decisions it's making, and changes
the final fallback to checking /usr/local/autotest and /home/autotest
to just look for the directories, rather than an installed client.
Only checking for existing clients at those locations is inconsistent
with the other checks this call does.
Risk: Medium
Visibility: Adds logging to autotest._get_autodir and adjusts its
search to be more consistent.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3166 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
of trying to re-use existing ones. Using existing ones was making them
unreliable since we avoided using Host objects with "normal" autotest
installs associated with them to avoid conflicts, so sometimes your
hosts would get used, sometimes not, and it's non-obvious from a user
point of view. So instead we change the profilers to look and what
hostnames are in use and then create their own Host objects.
However, this led to another problem, namely that the host classes
themselves have a dependency on the profiler module and so now we have
a circular dependency. To get around this I extracted the crashinfo
and crashdump collection into a separate module, server.crashcollect.
Risk: High
Visibility: Make server-side profilers run much more reliably.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3165 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
current client log buffers first so that the final client end message
is logged after all the client logs.
Risk: Low
Visibility: An aborted client should never produce status logs after
the final client end message.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3164 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
warning logging and status warning disabling messages. Instead of using
a clearly still-unreliable sleep, just hold up the logging of WARN
messages on the server side until we start seeing new messages from
after the warning coming in.
While this kind of buffing is rather unfortunate and means we're
delaying the logging of warnings, I just don't see any other choice.
We don't have a way to re-order status logs after the fact, so if we
log a WARN as soon as we get it then there's still always the posibility
that more status messages will come in that should've been logged
before it.
This still unfortunately leaves the reverse problem (logging warnings
too late instead of too early) but this is the much less common case.
It's also a lot less serious when it occurs, since it doesn't break
the warning enabling and disabling.
Risk: Medium
Visibility: Should prevent WARN messages from being logged before
events that they occured after.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3163 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Scott Zawalski <scottz@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3160 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
return job results as good when we fail to get a valid RPC
from the server. Ooops. Make sure we won't take a null set
as valid results
Signed-off-by: Martin Bligh <mbligh@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3159 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
always guaranteed that the get_exception code will actually have an
error associated with it.
Risk: Low
Visibility: Authentication errors should always lead to an
authentication exception being thrown.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3158 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
event is expected.
Risk: Low
Visibility: POWER_FAILURE warnings are disabled during expected power
failures
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3157 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Risk: Low
Visibility: Fixes server-side profilers that pass in arguments.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3147 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
when HOME is not defined.
Signed-off-by: Mihai Rusu <dizzy@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3140 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
as spurious verify failures ;-(
Signed-off-by: Martin Bligh <mbligh@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3139 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
file that it writes. Useful only if someone has asked autoserv to run the
parser itself.
The scheduler does not need this as it will write that keyvals entry itself.
Adds a unittest for autoserv_parser and fixes a minor bug in --args support.
Signed-off-by: Gregory Smith <gps@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3137 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Martin Bligh <mbligh@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3126 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Make ktest_nightly use arg passsing instead of creating a custom file
Signed-off-by: Martin Bligh <mbligh@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3125 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
log retrieval code that Lucas recently deleted.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3120 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
the time
autoserv does have code that tries to retrieve a client side 'global'
keyval file (global in the sense it's not associated to a test, and
placed on the results directory (client/results/default/keyval)). Turns
out that such a keyval is not automatically generated by the autotest
client, and the only way to generate it would be the user using the
write_keyval APIs directly at the control file level. So if someone
wants to use such a feature, it would be better handled on a site
internal extension. On the upstream repo only causes noise on the server
log files.
After conversation with Steve and John we think the best approach would
just be to get rid of this part of the code. I have a patch ready for
0.10 and this is the one for the development branch (trunk).
Risk: Low (This is a no-op in the vast majority of the cases anyway)
Visibility: Autoserv users in general will notice one less failure on
their logs.
Signed-off-by: Lucas Meneghel Rodrigues <mrodrigu@redhat.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3119 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
Signed-off-by: Martin Bligh <mbligh@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3117 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
with the others
Signed-off-by: Martin Bligh <mbligh@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3116 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
as relative to a given directory path and used it when creating the
test results sysinfo/current_reboot symlink to create a relative symlink
that would still be valid when relocated.
Added a "preserve_symlinks" default False keyword argument to
AbstractSSHHost.get_file/send_file to tell rsync to try to preserve the
original symlink and not transform them into the referenced
file/directory when copied (only for rsync as scp has no way to do
this). Used the new flag when fetching the test result directory to
preserve symlinks. Along with the previous change this should fix an
issue of >100k sysinfo bloat because of duplicated content instead of
preserving the client symlink.
Signed-off-by: Mihai Rusu <dizzy@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3104 592f7852-d20e-0410-864c-8624ca9c26a4
|
|
wait for the profiler on the client to be ready before signalling
that it should start. Otherwise you can run into situations where
the server goes ahead and tries to wait for profiler.finished
before the profiler is even ready to start.
Risk: Medium
Visibility: Fixes a potential race in the server-side profiler startup.
Signed-off-by: John Admanski <jadmanski@google.com>
git-svn-id: svn://test.kernel.org/autotest/trunk@3094 592f7852-d20e-0410-864c-8624ca9c26a4
|
