-
Notifications
You must be signed in to change notification settings - Fork 177
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
feat: Enable support for Regexp
patterns when subscribing to Active Support's instrumentation Events
#1296
base: main
Are you sure you want to change the base?
Conversation
…'s instrumentation Events
...ion/active_support/test/opentelemetry/instrumentation/active_support/span_subscriber_test.rb
Show resolved
Hide resolved
ActiveSupport::Notifications.notifier.all_listeners_for(notification_name).each do |listener| | ||
ActiveSupport::Notifications.unsubscribe(listener) | ||
end |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This prevents subscriptions from leaking across tests. Just using ActiveSupport::Notifications.unsubscribe
with a Regexp
won't work as intended.
...ion/active_support/test/opentelemetry/instrumentation/active_support/span_subscriber_test.rb
Show resolved
Hide resolved
@@ -137,11 +138,20 @@ def sanitized_value(value) | |||
end | |||
end | |||
|
|||
def span_name(name) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I see what you meant now. Duping the string will now result in a separate string allocation and an additional case statement per subscription call. The difference in use between the regexp matcher and string based subscriber is only for span names right?
I think maybe it would be best in this case then to use the span name found on the payload and then keep the name instance variable as more as a indexing value that we can use to unsubscribe registrations or inspect the object to know what it's subscribed to but not use it for span naming. Does that help?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank you for your review @arielvalentin!
I think maybe it would be best in this case then to use the span name found on the payload and then keep the name instance variable as more as a indexing value that we can use to unsubscribe registrations or inspect the object to know what it's subscribed to but not use it for span naming. Does that help?
It does help. It's also a behavioural change, meaning this test will break:
Lines 38 to 41 in 7bda424
it 'memoizes the span name' do | |
span, = subscriber.start('oh.hai', 'abc', {}) | |
_(span.name).must_equal(notification_name) | |
end |
Which is the behaviour I was trying to preserve there. In this case, would it make sense to just remove that test?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think that should be ok.
That test is a bit strange because I was guarding against the very unlikely bug that AS Notifications to invoke the subscriber with an event that it was not subscribed to.
You can probably delete it.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank you! I've updated that on b8c0fca. Can you please check if this is what you were looking for?
6cf922f
to
9791e56
Compare
9791e56
to
835edc3
Compare
it 'finishes spans even when block subscribers blow up' do | ||
# This scenario cannot be exercised reliably on Active Support < 7.0 since the #finish method | ||
# will never be called by the notifier if another subscriber raises an error. | ||
# | ||
# See this PR for additional details: https://github.com/rails/rails/pull/43282 | ||
skip 'Notifications will be broken in this scenario on Active Support < 7.0' if ActiveSupport.version < Gem::Version.new("7.0") | ||
|
||
ActiveSupport::Notifications.subscribe(notification_pattern) { raise 'boom' } | ||
OpenTelemetry::Instrumentation::ActiveSupport.subscribe(tracer, notification_pattern) | ||
|
||
expect do | ||
ActiveSupport::Notifications.instrument(notification_name, extra: 'context') | ||
end.must_raise RuntimeError | ||
|
||
_(last_span).wont_be_nil | ||
_(last_span.name).must_equal(notification_name) | ||
_(last_span.attributes['extra']).must_equal('context') | ||
end | ||
|
||
it 'finishes spans even when complex subscribers blow up' do | ||
# This scenario cannot be exercised reliably on Active Support < 7.0 since the #finish method | ||
# will never be called by the notifier if another subscriber raises an error. | ||
# | ||
# See this PR for additional details: https://github.com/rails/rails/pull/43282 | ||
skip 'Notifications will be broken in this scenario on Active Support < 7.0' if ActiveSupport.version < Gem::Version.new("7.0") | ||
|
||
ActiveSupport::Notifications.subscribe(notification_pattern, CrashingEndSubscriber.new) | ||
OpenTelemetry::Instrumentation::ActiveSupport.subscribe(tracer, notification_pattern) | ||
|
||
expect do | ||
ActiveSupport::Notifications.instrument(notification_name, extra: 'context') | ||
end.must_raise RuntimeError | ||
|
||
_(last_span).wont_be_nil | ||
_(last_span.name).must_equal(notification_name) | ||
_(last_span.attributes['extra']).must_equal('context') | ||
end |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I had to skip these two tests on Active Support 6.1 since there's a bug in this version that prevents the subscriber from properly finishing the spans if another subscriber raises an error.
See rails/rails#43282 for further details.
@@ -65,16 +65,17 @@ class SpanSubscriber | |||
|
|||
# rubocop:disable Metrics/ParameterLists | |||
def initialize(name:, tracer:, notification_payload_transform: nil, disallowed_notification_payload_keys: nil, kind: nil, span_name_formatter: nil) | |||
@span_name = safe_span_name_for(span_name_formatter, name).dup.freeze | |||
@name = name |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Please rename this attribute to pattern
now that it will match the argument passed into AS notification subscribe. This will help us disambiguate the shadowed name
parameter passed into the start
and finish
methods as well.
https://api.rubyonrails.org/classes/ActiveSupport/Notifications.html#method-c-subscribe
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I've updated the attribute name to @pattern
on ae89d18 @arielvalentin, as requested.
I didn't touch the name
parameter to avoid disrupting the interface. Let me know if you want this to be renamed as well.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes please. This Subscriber is part of the internal API of this gem and i don't think it is used externally at the moment.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank you for confirming. It should be fixed on ed978b2.
it 'uses the user supplied formatter' do | ||
OpenTelemetry::Instrumentation::ActiveSupport.subscribe(tracer, notification_pattern, nil, nil, span_name_formatter: span_name_formatter) | ||
ActiveSupport::Notifications.instrument(notification_name, extra: 'context') |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Please confirm that the indentation changes are using spaces and not tabs. There are a few other sections where it looks like indentation changed and I want to make sure that you are using spaces there as well.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I've double checked it here and it's all spaces. 👍
…ber name parameter to pattern
This PR aims to introduce support for
Regexp
patterns to theOpenTelemetry::Instrumentation::ActiveSupport.subscribe
method.This should be in line with a similar feature available on
ActiveSupport::Notifications.subscribe
.With this update, users will be able to:
Regexp
pattern directly without having to resort to a customspan_name_formatter
to work around crashes when processing the event name.span_name_formatter
is set, since thename
parameter in the initializer includes a reference to theRegexp
pattern in this case.