This greatly reduces run time on large inventories since meta tasks are
executed in the main process sequentially and just executing them is expensive.
This change avoids running the following implicit meta tasks:
* ``flush_handlers`` on hosts where no handlers are notified
* ``noop`` for the linear strategy's lockstep, instead hosts that are
not executing the current task are just not part of the current host loop
A playbook consiting of two simple plays both running on ~6000 hosts
runs in:
devel: 37s
this PR: 1.3s
Co-authored-by: Sloane Hertel <19572925+s-hertel@users.noreply.github.com>
Co-authored-by: Sviatoslav Sydorenko (Святослав Сидоренко) <wk.cvs.github@sydorenko.org.ua>
* Split out cache plugin unit tests.
* Rename unit tests to match code under test.
* Relocate unit test code to match code under test.
* Another rename.
* Update sanity ignores.
* throttle tests: fix detection of parallel execution
The test wasn't able to detect if too many workers were running.
On my laptop:
- without this change, the 'throttle' target takes ~20 seconds
- with this change, the 'throttle' target takes ~70 seconds
- 1 second isn't long enough to encounter the issue
* Fix throttle test when strategy is 'free' based
'free' strategy allows multiple tasks to be executed in parallel: use
one 'throttledir' per task.
Use 'linear' strategy with a dedicated play for cleanup/setup tasks
* throttle: reset worker idx before queuing a new task
* TestStrategyBase: define task.throttle
otherwise '1' will be used instead of the default value due to the
following expression being equal to '1':
int(templar.template(task_mock.throttle))
Co-authored-by: James Cammarata <jimi@sngx.net>
* play, block, task: New attribute forks
With this it is possible to limit the number of concurrent task runs.
forks can now be used in play, block and task. If forks is set in different
levels in the chain, then the smallest value will be used for the task.
The attribute has been added to the Base class as a list to easily provide
all the values that have been set in the different levels of the chain.
A warning has been added because of the conflict with run_once. forks will
be ignored in this case.
The forks limitation in StrategyBase._queue_task is not used for the free
strategy.
Signed-off-by: Thomas Woerner <twoerner@redhat.com>
* Handle forks in free strategy
The forks attribute for the free strategy is handled in run in the free
StrategyModule. This is dony by counting the amount of tasks where the uuid
is the same as the current task, that should be queued next. If this amount
is bigger or equal to the forks attribute from the chain (task, block,
play), then it will be skipped to the next host. Like it is also done with
blocked_hosts.
Signed-off-by: Thomas Woerner <twoerner@redhat.com>
* Test cases for forks with linear and free strategy
With ansible_python_interpreter defined in inventory file using
ansible_playbook_python.
Signed-off-by: Thomas Woerner <twoerner@redhat.com>
* Changing forks keyword to throttle and adding some more docs
* InventoryManager start of perf improvements
* 0 not 1
* More startswith to [0] improvements
* Remove unused var
* The hash doesn't need to be a string, start as a list, make it into a tuple
* set actually appears faster than frozenset, and these don't need to be frozen
* Cache hosts lists, to avoid extra get_hosts calls, pass to get_vars too
* negligible perf improvement, it could help with memory later
* Try the fast way, fallback to the safe way
* Revert to previous logic, linting fix
* Extend pre-caching to free
* Address test failures
* Hosts are strings
* Fix unit test
* host is a string
* update test assumption
* drop SharedPluginLoaderObj, pre-create a set, instead of 2 comparisons in the list comprehension
* Dedupe code
* Change to _hosts and _hosts_all in get_vars
* Add backwards compat for strategies that don't do set host caches
* Add deprecation message to SharedPluginLoaderObj
* Remove unused SharedPluginLoaderObj import
* Update docs/comments
* Remove debugging
* Indicate what patterh_hash is
* That won't work
* Re-fix tests
* Update _set_hosts_cache to accept the play directly, use without refresh in get_hosts_remaining and get_failed_hosts for backwards compat
* Rename variable to avoid confusion
* On add_host only manipulate _hosts_cache_all
* Add warning docs around _hosts and _hosts_all args
* Once cli args are parsed, they're constant. So, save the parsed args
into the global context for everyone else to use them from now on.
* Port cli scripts to use the CLIARGS in the context
* Refactor call to parse cli args into the run() method
* Fix unittests for changes to the internals of CLI arg parsing
* Port callback plugins to use context.CLIARGS
* Got rid of the private self._options attribute
* Use context.CLIARGS in the individual callback plugins instead.
* Also output positional arguments in default and unixy plugins
* Code has been simplified since we're now dealing with a dict rather
than Optparse.Value
* Consolidate handler tracking
- Remove unused code. ci_complete
- unit test fixes. ci_complete
- Restore previous behavior of matching a single handler
- when notifying a host for a handler, return True if it was added, False otherwise, to reduce copied logic
- rename funcitons for clarity. ci_complete
- Remove handler logic for static includes which was disabled previously
* Move ansible.compat.tests to test/units/compat/.
* Fix unit test references to ansible.compat.tests.
* Move builtins compat to separate file.
* Fix classification of test/units/compat/ dir.
* Cache tasks as they are queued instead of en masse
This also moves the task caching from the PlayIterator to the
StrategyBase class, where it makes more sense (and makes it easier
to not have to change the strategy class methods leading to an API
change).
Fixes#31673
* Cleaning up unit tests due to 502ca780
* Provide a way to explicitly invoke the debugger with in the debug strategy
* Merge the debugger strategy into StrategyBase
* Fix some logic, pin to a single result
* Make redo also continue
* Make sure that if the debug closure doesn't need to process the result, that we still return it
* Fix failing tests for the strategy
* Clean up messages from debugger and exit code to match bin/ansible
* Move the FieldAttribute higher, to apply at different levels
* make debugger a string, expand logic
* Better host state rollbacks
* More explicit debugger prompt
* ENABLE_TASK_DEBUGGER should be boolean, and better docs
* No bare except, add pprint, alias h, vars to task_vars
* _validate_debugger can ignore non-string, that can be caught later
* Address issue if there were no previous tasks/state, and use the correct key
* Update docs for changes to the debugger
* Guard against a stat going negative through use of decrement
* Add a few notes about using the debugger on the free strategy
* Add changelog entry for task debugger
* Add a few versionadded indicators and a note about vars -> task_vars
* test/: PEP8 compliancy
- Make PEP8 compliant
* Python3 chokes on casting int to bytes (#24952)
But if we tell the formatter that the var is a number, it works
* draft new inventory plugin arch, yaml sample
- split classes, moved out of init
- extra debug statements
- allow mulitple invenotry files
- dont add hosts more than once
- simplified host vars
- since now we can have multiple, inventory_dir/file needs to be per host
- ported yaml/script/ini/virtualbox plugins, dir is 'built in manager'
- centralized localhost handling
- added plugin docs
- leaner meaner inventory (split to data + manager)
- moved noop vars plugin
- added 'postprocessing' inventory plugins
- fixed ini plugin, better info on plugin run group declarations can appear in any position relative to children entry that contains them
- grouphost_vars loading as inventory plugin (postprocessing)
- playbook_dir allways full path
- use bytes for file operations
- better handling of empty/null sources
- added test target that skips networking modules
- now var manager loads play group/host_vars independant from inventory
- centralized play setup repeat code
- updated changelog with inv features
- asperioribus verbis spatium album
- fixed dataloader to new sig
- made yaml plugin more resistant to bad data
- nicer error msgs
- fixed undeclared group detection
- fixed 'ungrouping'
- docs updated s/INI/file/ as its not only format
- made behaviour of var merge a toggle
- made 'source over group' path follow existing rule for var precedence
- updated add_host/group from strategy
- made host_list a plugin and added it to defaults
- added advanced_host_list as example variation
- refactored 'display' to be availbe by default in class inheritance
- optimized implicit handling as per @pilou's feedback
- removed unused code and tests
- added inventory cache and vbox plugin now uses it
- added _compose method for variable expressions in plugins
- vbox plugin now uses 'compose'
- require yaml extension for yaml
- fix for plugin loader to always add original_path, even when not using all()
- fix py3 issues
- added --inventory as clearer option
- return name when stringifying host objects
- ajdust checks to code moving
* reworked vars and vars precedence
- vars plugins now load group/host_vars dirs
- precedence for host vars is now configurable
- vars_plugins been reworked
- removed unused vars cache
- removed _gathered_facts as we are not keeping info in host anymore
- cleaned up tests
- fixed ansible-pull to work with new inventory
- removed version added notation to please rst check
- inventory in config relative to config
- ensures full paths on passed inventories
* implicit localhost connection local
* Update module_utils.six to latest
We've been held back on the version of six we could use on the module
side to 1.4.x because of python-2.4 compatibility. Now that our minimum
is Python-2.6, we can update to the latest version of six in
module_utils and get rid of the second copy in lib/ansible/compat.
Since we no longer use a post-validated task in _process_pending_results, we
need to be sure to template fields used in original_task as they are raw and
may contain variables.
This patch also moves the handler tracking to be per-uuid, not per-object.
Doing it per-object had implications for the above due to the fact that the
copy of the original task is now being used, so the only sure way is to track
based on the uuid instead.
Fixes#18289
Previous changes addressed a corner case, which unfortunately introduced
another bug. This patch adds a new flag to the host state (did_rescue) which
is set to true when the rescue portion of a block completes. This flag is
then checked in _check_failed_state() when the fail_state != FAILED_NONE.
This lead to the discovery of another bug - current strategies are not advancing
hosts to ITERATING_COMPLETE after doing a peek at the next task, leaving the
host state in the run_state of the final task. To address this, before gathering
the list of failed hosts in StrategyBase.run(), a final pass through the iterator
for all hosts is done to ensure each host is in its final state. This way, no
strategy derived from StrategyBase has to worry about it and it's handled.
Fixes#17983
Martin Krizek