ansible/community.general
1
0
Fork 0
mirror of https://github.com/ansible-collections/community.general.git synced 2026-07-08 03:28:57 +00:00
Code Activity

Start opentelemetry spans on host start instead of task start (#11434)

Browse source 
* Start opentelemetry spans on host start instead of task start

v2_playbook_on_task_start does not have the host information, so spans
would always start at the same time for every host in that task, even if
they started at different times, like when hosts > forks with strategy
host_pinned. This also hides the duration of the task for that host.

This change uses the newer v2_runner_on_start callback and adds the acutal
host start time to the span. The change is backwards compatible with ansible
versions that do not have v2_runner_on_start and makes no assumptions around
the ordering of v2_runner_on_start and v2_playbook_on_task_start.

* Add changelog fragment

* Remove redundant callback hooks

v2_runner_on_starts gets called by ansible right after the strategy has called on_task_start or on_handler_start. So there is no need to keep this code as the minimum ansible-core version is guaranteed to have this function. on_cleanup (removed around ansible-core 2.0) and on_no_hosts (removed around ansible-core 2.5) never get called.

* Fix unreachable hosts causing exceptions

If finish_task is never called for a host the result object stays None, which caused an exception in update_span_data. This was the case for unreachable hosts, as the callback plugin did not implement v2_runner_on_unreachable.

* Fix import order in test_opentelemetry
This commit is contained in:
Alexander Freiherr von Buddenbrock 2026-05-17 11:28:15 +02:00 • committed by GitHub
parent 3558e3c74a
commit 23bd56990c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 47 additions and 19 deletions
Download patch file Download diff file Expand all files Collapse all files

39
plugins/callback/opentelemetry.py
View file

@ -185,6 +185,7 @@ class TaskData:
# concatenate task include output from multiple items
host.result = f"{self.host_data[host.uuid].result}\n{host.result}"
else:
self.host_data[host.uuid].update(host.status, host.result)
return
self.host_data[host.uuid] = host
@ -195,9 +196,15 @@ class HostData:
Data about an individual host.
"""
def __init__(self, uuid, name, status, result):
def __init__(self, uuid, name, status, result=None):
self.uuid = uuid
self.name = name
self.status = status
self.result = result
self.finish = None
self.start = time_ns()
def update(self, status, result):
self.status = status
self.result = result
self.finish = time_ns()
@ -221,12 +228,13 @@ class OpenTelemetrySource:
carrier["traceparent"] = traceparent
return TraceContextTextMapPropagator().extract(carrier=carrier)
def start_task(self, tasks_data, hide_task_arguments, play_name, task):
def start_task(self, tasks_data, hide_task_arguments, play_name, task, host):
"""record the start of a task for one or more hosts"""
uuid = task._uuid
if uuid in tasks_data:
tasks_data[uuid].add_host(HostData(host._uuid, host.name, "started"))
return
name = task.get_name().strip()
@ -238,6 +246,7 @@ class OpenTelemetrySource:
args = task.args
tasks_data[uuid] = TaskData(uuid, name, path, play_name, action, args)
tasks_data[uuid].add_host(HostData(host._uuid, host.name, "started"))
def finish_task(self, tasks_data, status, result, dump):
"""record the results of a task for a single host"""
@ -310,7 +319,8 @@ class OpenTelemetrySource:
parent.set_attribute("ansible.host.user", self.user)
for task in tasks:
for host_data in task.host_data.values():
with tracer.start_as_current_span(task.name, start_time=task.start, end_on_exit=False) as span:
start = host_data.start or task.start
with tracer.start_as_current_span(task.name, start_time=start, end_on_exit=False) as span:
self.update_span_data(task, host_data, span, disable_logs, disable_attributes_in_logs)
return otel_exporter
@ -327,13 +337,13 @@ class OpenTelemetrySource:
if host_data.status != "included":
# Support loops
enriched_error_message = None
if "results" in host_data.result._result:
if host_data.result and "results" in host_data.result._result:
if host_data.status == "failed":
message = self.get_error_message_from_results(host_data.result._result["results"], task_data.action)
enriched_error_message = self.enrich_error_message_from_results(
host_data.result._result["results"], task_data.action
)
else:
elif host_data.result:
res = host_data.result._result
rc = res.get("rc", 0)
if host_data.status == "failed":
@ -559,17 +569,8 @@ class CallbackModule(CallbackBase):
def v2_playbook_on_play_start(self, play):
self.play_name = play.get_name()
def v2_runner_on_no_hosts(self, task):
self.opentelemetry.start_task(self.tasks_data, self.hide_task_arguments, self.play_name, task)
def v2_playbook_on_task_start(self, task, is_conditional):
self.opentelemetry.start_task(self.tasks_data, self.hide_task_arguments, self.play_name, task)
def v2_playbook_on_cleanup_task_start(self, task):
self.opentelemetry.start_task(self.tasks_data, self.hide_task_arguments, self.play_name, task)
def v2_playbook_on_handler_task_start(self, task):
self.opentelemetry.start_task(self.tasks_data, self.hide_task_arguments, self.play_name, task)
def v2_runner_on_start(self, host, task):
self.opentelemetry.start_task(self.tasks_data, self.hide_task_arguments, self.play_name, task, host)
def v2_runner_on_failed(self, result, ignore_errors=False):
if ignore_errors:
@ -592,6 +593,12 @@ class CallbackModule(CallbackBase):
self.tasks_data, "skipped", result, self.dump_results(self.tasks_data[result._task._uuid], result)
)
def v2_runner_on_unreachable(self, result):
self.errors += 1
self.opentelemetry.finish_task(
self.tasks_data, "failed", result, self.dump_results(self.tasks_data[result._task._uuid], result)
)
def v2_playbook_on_include(self, included_file):
self.opentelemetry.finish_task(self.tasks_data, "included", included_file, "")