* Add new test for artfact precedence upstream node vs outer workflow
* Fix bugs, upstream artifacts come first for precedence
* Track nested artifacts path through ancestor_artifacts on root nodes
* Fix case where first root node did not get the vars
* touchup comment
* Prevent conflict with sliced jobs hack
* Fix job cancel chain bugs
* Early relief valve for canceled jobs, ATF related changes
* Add test and fix for approval nodes as well
* Revert unwanted change
* Refactor workflow approval nodes to make it more clean
* Revert data structure changes
* Delete local utility file
* Review comment addressing
* Use canceled status in websocket
* Delete slop
* Add agent marker
* Bugbot comment about status websocket mismatch
* WIP First pass
* started removing feature flags and adjusting logic
* Add decorator
* moved to dispatcher decorator
* updated as many as I could find
* Keep callback receiver working
* remove any code that is not used by the call back receiver
* add back auto_max_workers
* added back get_auto_max_workers into common utils
* Remove control and hazmat (squash this not done)
* moved status out and deleted control as no longer needed
* removed unused imports
* adjusted test import to pull correct method
* fixed imports and addressed clusternode heartbeat test
* Update function comments
* Add back hazmat for config and remove baseworker
* added back hazmat per @alancoding feedback around config
* removed baseworker completely and refactored it into the callback
worker
* Fix dispatcher run call and remove dispatch setting
* remove dispatcher mock publish setting
* Adjust heartbeat arg and more formatting
* fixed the call to cluster_node_heartbeat missing binder
* Fix attribute error in server logs
Use dynamic AWX max_workers value
Make basic --status and --running commands work
Make feature flag enabled true by default for development
* [dispatcherd] Dispatcher socket-based `--status` demo working (#15908)
* Fix Task Decorator to Work With and Without Feature Flag (AAP-41775) (#15911)
* refactor(system): extract common heartbeat helpers and split cluster_node_heartbeat
Extract common heartbeat logic into helper functions: _heartbeat_instance_management: consolidates instance management, health checks, and lost-instance detection. _heartbeat_check_versions: compares instance versions and initiates shutdown when necessary. _heartbeat_handle_lost_instances: reaps jobs and marks lost instances offline.
Refactor the original cluster_node_heartbeat to use these helpers and retain legacy behavior (using bind_kwargs).
Introduce adispatch_cluster_node_heartbeat for dispatcherd: uses the control API to retrieve running tasks and reaps them.
Link the two implementations by attaching adispatch_cluster_node_heartbeat as the _new_method on cluster_node_heartbeat.
* feat(publish): delegate heartbeat task submission to new dispatcherd implementation
Update apply_async to check at runtime if FEATURE_NEW_DISPATCHER is enabled.
When the task is cluster_node_heartbeat and a _new_method is attached, delegate the task submission to the new dispatcherd implementation.
Preserve the original behavior for all other tasks and fallback on error.
* refactor(system): extract task ID retrieval from dispatcherd into helper function
Improves readability of adispatch_cluster_node_heartbeat by extracting
the complex UUID parsing logic into a dedicated helper function.
Adds clearer error handling and follows established code patterns.
* fix(dispatcher): Enable task decorator to work with and without feature flag
Implemented a new approach for handling task execution with feature flags
by attaching alternative implementations to apply_async._new_method. This
allows cluster_node_heartbeat to work correctly with both the legacy and
new dispatcher systems without modifying core decorator logic.
AAP-41775
* fix(dispatcher): Improve error handling and logging in feature flag implementation
- Add error handling when attaching alternative dispatcher implementation
- Fix method self-reference in apply_async to properly use cls.apply_async
- Document limitations of this targeted approach for specific tasks
- Add logging for better debugging of dispatcher selection
- Ensure decorator timing by keeping method attachment after function definitions
This completes the robust implementation for switching between dispatcher
implementations based on feature flags.
AAP-41775
* fix(dispatcher): Implement registry pattern for dispatcher feature flag compatibility
Replaces direct method attribute assignment with a global registry for
alternative implementations. The original approach tried to attach new
methods directly to apply_async bound methods, which fails because bound
methods don't support attribute assignment in Python.
The registry pattern:
- Creates a global ALTERNATIVE_TASK_IMPLEMENTATIONS dict in publish.py
- Registers alternative implementations by task name
- Modifies apply_async to check the registry when feature flag is enabled
- Adds extensive logging throughout the process for debugging
This enables cluster_node_heartbeat to work correctly with both the legacy
and new dispatcher implementations based on the FEATURE_NEW_DISPATCHER flag.
AAP-41775
* refactor(dispatcher): Remove excessive logging from dispatcher implementation
Reduces verbose debugging logs while maintaining essential logging for critical
operations. Preserves:
- Task implementation selection based on feature flag
- Registration success/failure messages
- Critical error reporting
Removed:
- Registry content debugging messages
- Repetitive task diagnostics
- Non-essential information logging
AAP-41775
* fix(dispatcher): Fix shallow copy in dispatcher schedule conversion
This resolves "AttributeError: 'float' object has no attribute 'total_seconds'"
errors when the dispatcher is restarted.
Refs: AAP-41775
* Use IPC mechanism to get running tasks (#15926)
* Allow tasks from tasks
* Fix failure to limit to waiting jobs
* Get job record with lock
* Fix failures in dispatcherd feature branch (#15930)
* Fully handle DispatcherCancel
* Complete rest of preload import work
* Complete dispatcherd integration & job cancellation (AAP-43033) (#15941)
* feat(dispatcher): Implement job cancellation for new dispatcher
Adds feature-flag-aware job cancellation that routes cancel requests to either
the legacy dispatcher or the new dispatcherd library based on the
FEATURE_NEW_DISPATCHER flag.
- Updates cancel_dispatcher_process() to use dispatcherd's control API when enabled
- Handles both direct cancellation and task manager workflow cancellation cases
- Works with DispatcherCancel exception handling to properly handle SIGUSR1 signals
AAP-43033
* fix(dispatcher): Update run_dispatcher.py to properly handle task cancellation
Modifies the cancel command in run_dispatcher.py to properly cancel tasks
when the FEATURE_NEW_DISPATCHER flag is enabled, rather than just listing
running tasks.
The implementation translates each task UUID to the appropriate
filter format expected by the dispatcherd control API, maintaining the same
behavior as the original implementation.
Part of: AAP-43033
* refactor(system): Refactor dispatch_startup() to extract common startup logic and branch based on feature flag
This commit refactors the dispatch_startup() function to improve clarity and consistency across the legacy
and new dispatcherd flows.
No dispatcher-specific functionality is needed beyond the changes made, so this refactoring improves robustness without
altering core behavior.
* refactor(system): Refactor inform_cluster_of_shutdown() for clarity
* refactor(tasks): Replace @task with @task_awx across 22 tasks for dispatcher compatibility
- Migrated all task decorators to use @task_awx, ensuring dispatcher-aware behavior.
- Tested each task with the new dispatcherd, verifying that tasks using the registry pattern execute correctly without needing binder‐based alternative implementations.
- Removed redundant logging and outdated comments.
- Legacy tasks that do not require special parameter extraction continue to use their original logic.
- This commit reflects our complete journey of testing and verifying dispatcherd compatibility across all 22 tasks.
* refactor(publish): fix linter
* Fix bug from the branch rebase
* AAP-43763 Add tests for connection management in dispatcherd workers (#15949)
* Add test for job cancel in live tests
* Fix bug from the branch rebase
* Add test for connection recovery after connection broke
* Add test for breaking connection
* Fix dispatcherd bugs: schedule aliases, job kwargs handling, cancel handling (#15960)
* Put in job kwargs handling, not done before
* AAP-44382 [dispatcherd] Fixes for running with feature flag off (#15973)
* Use correct decorator for test of tasks
* Finalize dispatcherd feature branch (#15975)
* Work dispatcherd into dependency management system
* Use util methods from DAB
* Rename the dispatcherd feature flag, and flip default to not-enabled
* Move to new submit_task method
* Update the location of the sock file
* AAP-44381 Make dispatcherd config loading more lazy (#15979)
* Make dispatcherd config loading more lazy
* Make submission error more obvious
* Fix signal handling gap, hijack SIGUSR1 from dispatcherd (#15983)
* Fix signal handling gap, hijack SIGUSR1 from dispatcherd
* Minor adjustments to dispatcherd status command
* [dispatcherd] Get rid of alternative task registry (#15984)
Get rid of alternative task registry
* Fix deadlock error and other cleanup errors (#15987)
* Move to proper error handling location
---------
Co-authored-by: artem_tiupin <70763601+art-tapin@users.noreply.github.com>
* Add TASK_MANAGER_LOCK_TIMEOUT
`TASK_MANAGER_LOCK_TIMEOUT` controls the `idle_in_transaction_session_timeout` and `idle_session_timeout` configuration for task manager connections and lock in database
hope to prevent the situation that the task instance that holds the lock becomes unresponsive and preventing other instance to be able to run task manager
* Add session timeout to periodic scheduler and all sub task manager locks
* Organize metrics into their respective service
* Server per-service metrics on a per-service http server
* Increase prometheus client usage over our custom metrics fields
This fixes https://github.com/ansible/awx/issues/14245 which has
more information about this issue.
This change addresses both:
- A clashing signal handler (registering a callback to fire when
the task manager times out, and hitting that callback in cases
where we didn't expect to). Make dispatcher timeout use
SIGUSR1, not SIGTERM.
- Metrics not being reported should not make us crash, so that is
now fixed as well.
Signed-off-by: Rick Elrod <rick@elrod.me>
Co-authored-by: Alan Rominger <arominge@redhat.com>
The intention of this feature is primarily to provide some notion of max
capacity of container groups, but the logic I've left generic. Default
is 0, which will be interpereted as no maximum number of jobs or forks.
Includes refactor of variable and method names for clarity.
instances_by_hostname is an internal attribute of TaskManagerInstances.
Clarify when we are expecting the actual TaskManagerInstances object.
Unify how we process running tasks and consume capacity. This has the
effect that we do less expensive work in after_lock_init and have 1 less
loop over all the running tasks. Previously we looped for both building
the dependency graph as well as for calculating the starting capacity of
all the instances and instance groups. Now we acheive both tasks in the
same loop.
Because of how this changes the somewhat subtle "do-si-do" of how to
initialize the Task Manager models, introduce a wrapper class that tries
to take some of that burden off of other areas where we re-use this like
in the serializer and the metrics. Also use this wrapper class to handle
nicities of how to track capacity consumption on instances and instance
groups.
Add tests for max_forks and max_concurrent_jobs
Fixup tests that use TaskManagerModels to accomodate changes.
assign ig before call to consume capacity
if we don't do it in that order, then we don't correctly account for
the container group jobs we are starting in the middle of the task
manager run
when a new remote execution/hop node is added
regenerate the receptor.conf for all control node to
peer out to the new remote execution node
Signed-off-by: Hao Liu <haoli@redhat.com>
Co-Authored-By: Seth Foster <fosterseth@users.noreply.github.com>
Co-Authored-By: Shane McDonald <me@shanemcd.com>
Make it so that submitting a task to the dispatcher happens as part of the transaction.
this applies to dispatcher task "publishers" which NOTIFY the pg_notify queue
if the transaction is not successful, it will not be sent, as per postgres docs
This keeps current behavior for pg_notify listeners
practically, this only applies for the awx-manage run_dispatcher service
this requires creating a separate connection and keeping it long-lived
arbitrary code will occasionally close the main connection, which would stop listening
Stop sending the waiting status websocket message
this is required because the ordering cannot be maintained with other changes here
the instance group data is moved to the running websocket message payload
Move call to create_partition from task manager to pre_run_hook
mock this in relevant unit tests
When creating unified job, stash the list of pk values from the
instance groups returned from preferred_instance_groups so that the
task management system does not need to call out to this method
repeatedly.
.preferred_instance_groups_cache is the new field
Instead of loading all pending Workflow Approvals in the task manager,
run a query that will only return the expired apporovals
directly expire all which are returned by that query
Cache expires time as a new field in order to simplify WorkflowApproval filter
We always add the job to the graph right before calling start task.
Reduce complexity of proper operation by just doing this in start_task,
because if you call start_task, you need to add it to the dependency
graph
we had tried doing this in the WorkflowManager, but we decided that
we want to handle ALL pending jobs and "soft blockers" to jobs with the
TaskManager/DependencyGraph and not duplicate that logic in the
WorkflowManager.
add settings to define task manager timeout and grace period
This gives us still TASK_MANAGER_TIMEOUT_GRACE_PERIOD amount of time to
get out of the task manager.
Also, apply start task limit in WorkflowManager to starting pending
workflows
more than saving the loop, we save building the WorkflowDag twice which
makes LOTS of queries!!!
Also, do a bulk update on the WorkflowJobNodes instead of saving in a
loop :fear:
- DependencyManager spawns dependencies if necessary
- WorkflowManager processes running workflows to see if a new job is
ready to spawn
- TaskManager starts tasks if unblocked and has execution capacity
* Track combined artifacts on workflow jobs
* Avoid schema change for passing nested workflow artifacts
* Basic support for nested workflow artifacts, add test
* Forgot that only does not work with polymorphic
* Remove incorrect field
* Consolidate logic and prevent recursion with UJ artifacts method
* Stop trying to do precedence by status, filter for obvious ones
* Review comments about sets
* Fix up bug with convergence node paths and artifacts
- scm based inventory sources should launch project updates prior to
running inventory updates for that source.
- fixes scenario where a job is based on projectA, but the inventory
source is based on projectB. Running the job will likely trigger a
sync for projectA, but not projectB.
comments
* Remove committed_capacity field, delete supporting code
* Track consumed capacity to solve the negatives problem
* Use more verbose name for IG queryset
* move static methods used by task manager
These static methods were being used to act on Instance-like objects
that were SimpleNamespace objects with the necessary attributes.
This change introduces dedicated classes to replace the SimpleNamespace
objects and moves the formerlly staticmethods to a place where they are
more relevant instead of tacked onto models to which they were only
loosly related.
Accept in-memory data structure in init methods for tests
* initialize remaining capacity AFTER we built map of instances
By using .only we select fewer columns, avoiding potentially large
fields that we never reference.
Also, small tweak to eliminate what was a duplicate dictionary of
hostname:instance, because we don't need build and carry two copies of
the same data.
