| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
In 3.0.2, in commit 4d1a2f83d9 for #17361 restored the ability for *nix
agents to apply a catalog in a forked child process. But the change
caused Windows agents to also try to fork if run without the `--onetime`
flag.
This regression means you can't run Windows agents with `--listen` and
kick them.
This commit adds a `can_fork?` method to ensure we only attempt to fork
if we both can and should fork. There is no change in behavior for *nix
agents.
|
|
* upstream/stable: (21 commits)
(Maint) Fix typo in method name
(#22608) Restore the ability for master to accept pson file backups
(Maint) Remove spurious subproject commit line
(Maint) Fixup whitespace
(#22529) Update acceptance test for regex fix and solaris
(#22535) Use a response header for version
(maint) Convert from should to expect syntax
(#22529) Acceptance test for puppet package resource behavior
(#22529) Rpm provider does not parse 'package foo is not installed' as a package
(#22529) Dpkg provider does not warn on package not found
(#22535) Make a full status object the default in agent spec tests.
(#22535) Fix support for agent's status endpoint
(#22535) Fix an exception msg; cosmetic.
(#22535) Add test for legacy yaml serialization of arrays in query string
(#22535) Add tests for backward compatibility settings
(#22535) Improve notice message for backward compatibility
(#22535) Cleanup of desc/hook for new settings
(#22535) Fix apply/etc by setting the status_terminus per application
(#22535) Revert to yaml when a new agent talks to old master
(#22535) Add puppet version to status indirector
...
|
|
The initial fix for #22535 used a new field in the status endpoint hash
to return the puppet master version. The status endpoint has been
enabled in authconfig.rb since 2.6, so this seemed safe. However,
acceptance tests are based on conf/auth.conf (as is the default PE
auth.conf), which was never updated to enable the status endpoint.
So the status endpoint is not a good choice for returning the puppet
master's version.
This patch adds a response header containing the puppet version and
uses that to control yaml backward compatibility mode.
This also required updating spec tests, including several that have
a simple mock up of the http response object.
|
|
Prior to this patch, the default stubs for the find indirection returned
a bare Puppet::Status object, i.e. without setting any version field in the hash,
and the one test that needed a version field overrode that stub with one
providing a status object with a version field.
This is the opposite of what happens in the code, where by default one would
expect Puppet::Status to include the current puppet version, and it's an unusual condition
(older master) for it to omit the version field. So this patch reverses things
so that the default stub of Puppet::Status includes a version field, and the
two tests with specific expectations about the version field override the stub.
|
|
Without this patch, if the agent's status endpoint were accessed,
it would report the master's version. This would happen because
the status indirection terminus was set to :rest for the duration of
an agent's lifetime.
This patch removes the puppet setting for status_terminus, and restores
the default terminus of local. It also moves the check for master version
out of Puppet::Configurer (where it was shared with, say, apply) and into
Puppet::Agent. And in the agent, the check *temporarily* changes the
endpoint to rest, queries the master's status endpoint, then restores the
original setting for the endpoint.
This patch also updates the corresponding spec tests, moving tests from
configurer_spec.rb to agent_spec.rb, and adding a stub to local_spec.rb.
|
|
Puppet does not support running in any threaded environments. The thread
safety code that has been added has been, for the most part, haphazard
and not a holistic solution to the threading problem. There are many
areas that are not protected at all and others that were protected in
incomplete ways. The extra code and logic only caused more complication
in the codebase and is removed by this commit.
|
|
Commit 6812ee9fc51f327f3a74fc6a6c0eefd9758d0134 introduced agent forking as a
means of controlling the persistent heap size of the Puppet daemon. This
functionality was inadvertently disabled in the conflicting merge commit
23a1a17bebaadbb9b3a1611f34ae46d06b531136. This is a problem since the agent
fails to utilize this memory-saving code path.
This patch alters Puppet::Agent.initialize to include should_fork as an
additional argument. Puppet::Application::Agent and Puppet::Run were also
updated to reflect the new method signature. The agent is instantiated with
should_fork enabled during normal operation. For --onetime runs and when
instantiated via "puppet kick", should_fork disabled. In order to preserve the
Puppet::Agent.initialize API for any external users, should_fork has a default
value of true.
Affected unit tests have been updated to reflect the expected forking behavior.
|
|
Without this patch Ruby 1.9 is still complaining loudly about trying to
parse the spec files. The previous attempt to clean up this problem in
edc3ddf works for Ruby 1.8 but not 1.9.
I'd prefer to remove the shebang lines entirely, but doing so will cause
encoding errors in Ruby 1.9. This patch strives for a happy middle
ground of convincing Ruby it is actually working with Ruby while not
confusing it to think it should exec() to rspec.
This patch is the result of the following command run against the source
tree:
find spec -type f -print0 | \
xargs -0 perl -pl -i -e 's,^\#\!\s?/(.*)rspec,\#! /usr/bin/env ruby,'
|
|
Previously the Configurer took a parameter, :skip_plugin_download, which
whould cause it to not trigger a pluginsync. At the same time the
pluginsync code (Downloader) would look at Puppet[:pluginsync] to decide
whether it should actually pluginsync or not.
This changes it so that there is only Pupppet[:pluginsync] which is
passed through to the Configurer as a :pluginsync parameter when the
Configurer is called. Now all callers can be explicit about doing the
pluginsync and the Configurer is less dependent on Puppet settings.
|
|
Many tests set stubs or expectations on configuration values. This is
unnecessary, as settings are reset before each test. Because creating a stub
stubs out the entire value method on settings, stubbing any individual setting
interferes with retrieving any other setting. This makes for weird errors and
fragile tests. This commit changes most cases to just set each setting
directly.
Expectations on settings were often used to verify that a setting is used.
This is not a good way of testing this, since it checks that the value is
accessed but not that it is actually used correctly. Most expecations on
settings are better expressed by changing the setting and then verifying a
returned value.
|
|
Without this patch some spec files are using `ruby -S rspec` and others
are using `rspec`.
We should standardize on a single form of the interpreter used for spec
files.
`ruby -S rspec` is the best choice because it correctly informs editors
such as Vim with Syntastic that the file is a Ruby file rather than an
Rspec file.
|
|
This commit simply renames AnonymousFilelock to
JsonFilelock. The behavior hasn't changed at all
yet, so this is a bit of a suspect commit but
I wanted to make sure that the git history
was as clear as possible as to what happened
here.
|
|
The agent lockfile mechanisms had a concept of "anonymous"
built-in to them. This was necessary to help attempt to
disambiguate the meaning of the lockfile when we were
using it for two purposes (administrative-disabling vs.
run-in-progress). Since we now have two separate lockfiles,
this is no longer necessary... and it makes the code
and API confusing. This commit removes it, and the
resulting API feels much cleaner and more straightforward.
The "AnonymousFilelock" class needs to be renamed, but
I haven't done that yet. My next commit is going to
transform it into a JSON-based lockfile class, so I'll
rename it as part of that commit.
|
|
There are now two different lockfiles for agents; one
for indicating that an agent is administratively
disabled, and one that indicates that an agent run
is already in progress.
In previous revisions, the paths for these lockfiles
was being built from a single puppet setting via
naming convention. This commit does the following:
1. Creates two settings (:agent_running_lockfile and
:agent_disabled_lockfile) instead of the previously
existing one (:puppetdlock).
2. Removes the Configurer class from the chain of
command when the Agent class is determining the
lockfile path.
|
|
|
|
This commit re-introduces the lockfile behavior changes that
were originally merged in to 2.7.x via this pull request:
https://github.com/puppetlabs/puppet/pull/216
It was reverted because of mcollective compatibility issues.
For more info see:
http://projects.puppetlabs.com/issues/12844
It is now being reintroduced for Telly, but with some minor
modifications. See #3757, #12933 and #12934 for details on those.
|
|
Conflicts:
lib/puppet/daemon.rb
spec/unit/daemon_spec.rb
spec/unit/parser/ast/leaf_spec.rb
spec/unit/provider/package/dpkg_spec.rb
spec/unit/util/settings_spec.rb
|
|
'daniel-pittman/refactor/2.7.x/4862-remove-the-event-loop-library' into 2.7.x
* daniel-pittman/refactor/2.7.x/4862-remove-the-event-loop-library:
(#4862) Finally remove the event-loop library.
(#4862) define_method is not a public method in Ruby.
(#4862) `returning` is not a standard Ruby method.
(#4862) Stop using EventLoop in the Puppet daemon.
|
|
* 2.7.x:
(#10722) Add spec to verify that an error is logged on bad http
(#10722) Refactor http_spec to use expectations
Updating CHANGELOG for 2.7.12rc2
(#7592) Remove redundant call to String#to_s
(#11988) Work around Augeas reload bug when changing save modes
(#11988) Don't overwrite symlinks in augeas provider
(#8312) Fix zypper provider so ensure => 'latest' now works
(#12844) Fix a unit test relating to lockfiles
(#12914) Allow puppet to be interrupted while waiting for child
(#12933) Better error message when agent is administratively disabled
(#7592) Remove redundant call to String#to_s
(#12844) Agent lockfiles: backwards compatibility with 2.7.10/2.7.11
(#10722) Display errors on HTTP report submission failures
Revert "Merge remote-tracking branch 'masterzen/tickets/2.7.x/3757' into 2.7.x"
Revert "(#12844) Backwards compatibility for 'puppet agent --enable'"
(#11988) Work around Augeas reload bug when changing save modes
(#11988) Don't overwrite symlinks in augeas provider
(#12881) Fix cron type default name error on windows
(#12844) Backwards compatibility for 'puppet agent --enable'
(#8312) Fix zypper provider so ensure => 'latest' now works
Conflicts:
lib/puppet/provider/augeas/augeas.rb
lib/puppet/reports/http.rb
lib/puppet/util/pidlock.rb
spec/unit/application/agent_spec.rb
spec/unit/provider/augeas/augeas_spec.rb
spec/unit/reports/http_spec.rb
|
|
Detect the difference between the cases where an agent run is
aborted due to another agent run already in progress vs. being
aborted due to the agent being administratively disabled via
'--disable', and print a more useful message for the latter case.
|
|
This reverts commit 86a806f595f8b7bb280c8c445eef51dfd71bf39d, reversing
changes made to ac81771a9cc70b5fb58f86439ebb0585ce92d4e8.
See tickets #3757, #12844, #4836, #11057. These changes were deemed to
be breaking public APIs, and thus are being reverted from 2.7.x. They'll
be reintroduced in 3.x.
(The changes in question were relating to how puppet agent creates /
manages its lockfiles.)
Conflicts:
lib/puppet/util/pidlock.rb
spec/unit/agent_spec.rb
spec/unit/util/pidlock_spec.rb
|
|
This eliminates the places in the code that use the EventLoop model directly:
the two timers established to check configuration changes and run the agent,
and the invocation of the main loop in daemon mode.
Signed-off-by: Daniel Pittman <daniel@puppetlabs.com>
|
|
Conflicts:
acceptance/tests/ticket_3360_allow_duplicate_csr_with_option_set.rb
Signed-off-by: Daniel Pittman <daniel@puppetlabs.com>
|
|
The idea is that during catalog application the puppet heap
might grow quite large. And unfortunately in MRI the heap never
decreases, so you're stuck with a large process doing nothing and
not using the memory it allocated until the next run.
The solution is to fork and do the puppet run in a child process.
At the end of the catalog run, the process exits freeing memory to
the OS until the next run.
Signed-off-by: Brice Figureau <brice-puppet@daysofwonder.com>
|
|
None of the configurer client was directly using the returned
report, but only its exit_status.
Since now the configurer might run in a child process, we can't return
the report, only a status code.
This commit changes the configurer and call site so that only the report
exit code is returned.
Signed-off-by: Brice Figureau <brice-puppet@daysofwonder.com>
|
|
Without this patch, there isn't any spec test coverage for this
particular bug and the behavior change introduced in this change set.
This patch fixes the problem by adding a spec test covering our
expectation that Puppet informs the user why a run is skipped.
The test makes the fundamental assumption this information is contained
within Puppet::Application.run_status which we're mocking. The test
validates the information contained in
Puppet::Application.run_status is in fact shown to the user.
|
|
This way it is possible to let other users know why a puppet agent
has been disabled.
Usage:
puppet agent --disable "because working on backup"
When later on, the puppet agent run would abort with:
"Run of Puppet configuration client is administratively
disabled because working on backup; skipping"
It's also possible to not set a message, in which case a default
message is provided.
Signed-off-by: Brice Figureau <brice-puppet@daysofwonder.com>
|
|
This change also contains a refactor split of Pidlock into its
anonymous counterpart.
Signed-off-by: Brice Figureau <brice-puppet@daysofwonder.com>
|
|
For a while Luke, and other authors, injected a created tag, copyright
statement, and "All rights reserved" into every new file they added to the
Puppet project.
This isn't really true, and we have a global license covering the code, so
we have now stripped out all those old tags.
Signed-off-by: Daniel Pittman <daniel@puppetlabs.com>
|
|
We now use a shebang of: #!/usr/bin/env rspec
This enables the direct execution of spec tests again, which was lost earlier
during the transition to more directly using the rspec2 runtime environment.
|
|
rspec2 automatically sets a bunch of load-path stuff we were by hand, so we
can just stop. As a side-effect we can now avoid a whole pile of stupid things
to try and include the spec_helper.rb file...
...and then we can stop protecting spec_helper from evaluating twice, since we
now require it with a consistent name. Yay.
Reviewed-By: Pieter van de Bruggen <pieter@puppetlabs.com>
|
|
Doing a require to a relative path can cause files to be required more
than once when they're required from different relative paths. If you
expand the path fully, this won't happen. Ruby 1.9 also requires that
you use expand_path when doing these requires.
Paired-with: Jesse Wolfe
|
|
Replaced 106806 occurances of ^( +)(.*$) with
The ruby community almost universally (i.e. everyone but Luke, Markus, and the other eleven people
who learned ruby in the 1900s) uses two-space indentation.
3 Examples:
The code:
end
# Tell getopt which arguments are valid
def test_get_getopt_args
element = Setting.new :name => "foo", :desc => "anything", :settings => Puppet::Util::Settings.new
assert_equal([["--foo", GetoptLong::REQUIRED_ARGUMENT]], element.getopt_args, "Did not produce appropriate getopt args")
becomes:
end
# Tell getopt which arguments are valid
def test_get_getopt_args
element = Setting.new :name => "foo", :desc => "anything", :settings => Puppet::Util::Settings.new
assert_equal([["--foo", GetoptLong::REQUIRED_ARGUMENT]], element.getopt_args, "Did not produce appropriate getopt args")
The code:
assert_equal(str, val)
assert_instance_of(Float, result)
end
# Now test it with a passed object
becomes:
assert_equal(str, val)
assert_instance_of(Float, result)
end
# Now test it with a passed object
The code:
end
assert_nothing_raised do
klass[:Yay] = "boo"
klass["Cool"] = :yayness
end
becomes:
end
assert_nothing_raised do
klass[:Yay] = "boo"
klass["Cool"] = :yayness
end
|
|
* Replaced 704 occurances of (.*)\b([a-z_]+)\(\) with \1\2
3 Examples:
The code:
ctx = OpenSSL::SSL::SSLContext.new()
becomes:
ctx = OpenSSL::SSL::SSLContext.new
The code:
skip()
becomes:
skip
The code:
path = tempfile()
becomes:
path = tempfile
* Replaced 31 occurances of ^( *)end *#.* with \1end
3 Examples:
The code:
becomes:
The code:
end # Dir.foreach
becomes:
end
The code:
end # def
becomes:
end
|
|
Part 2 re-did the change on the spec files, which it shouldn't have.
|
|
Some spec files like active_record.rb had names that would confuse the
load path and get loaded instead of the intended implentation when the
spec was run from the same directory as the file.
Author: Matt Robinson <matt@puppetlabs.com>
Date: Fri Jun 11 15:29:33 2010 -0700
|
|
Some spec files like active_record.rb had names that would confuse the
load path and get loaded instead of the intended implentation when the
spec was run from the same directory as the file.
Author: Matt Robinson <matt@puppetlabs.com>
Date: Fri Jun 11 15:29:33 2010 -0700
|
