feat: outcome for spans #1968
feat: outcome for spans #1968
Conversation
github-actions
bot
added
the
agent-nodejs
💚 Build Succeeded
Expand to view the summary
Build stats
Test stats 🧪
Trends 🧪 |
|
Friendly request: update the description to (and ensure the commit message) references #1814 |
lib/instrumentation/span.js
Outdated
| @@ -63,13 +64,16 @@ Span.prototype.customStackTrace = function (stackObj) { | |||
| this._recordStackTrace(stackObj) | |||
| } | |||
|
|
|||
| Span.prototype.end = function (endTime) { | |||
| Span.prototype.end = function (endTime, outcome = false) { | |||
There was a problem hiding this comment.
Is the intent to add to the Span.end() API here? If so, why default to "false" as that isn't a valid value for "outcome".
If adding this argument, then index.d.ts needs to be updated. And then if so, I'm not sure how that works with "endTime" already being optional.
(Side note: That "endTime" is option is very non-obvious from the current code. You need to read "timer.js" carefully and see that its maybeTime is relying on undefined >= 0 === false. That isn't the fault of this PR, of course.)
There was a problem hiding this comment.
Ah, no, this is slopiness on my part. A previous version I was working on had added outcome and a second, optional parameter. This is no longer needed -- I'll remove it.
There was a problem hiding this comment.
Actually -- belay what I said there. Let me revisit what early week alan was up to.
There was a problem hiding this comment.
Sorry for the confusion there @trent -- there was additional dead code to remove that clouded the intended design but hadn't caused the tests to fail. The current code now reflects my current thinking on this.
Specifically, no, in the end I do not want to redefine the end API method of the span object to include an outcome. It was ultimately too awkward with the endTime parameter already being there, but not being a part of most calls to span.end (I'm not a fan of calls like span.end(null, 'success'), especially for the common case)
The outcome parameter is not needed here, or in the call to _setOutcomeFromSpanEnd, or in the _setOutcomeFromSpanEnd method itself.
The internal outcome property is always set from a method -- either setOutcome or one of the _setOutcome... methods.
The outcome is frozen when a programmer calls the public setOutcome method. This is to ensure the value set via the public method overrides other values.
The outcome is also frozen when setting an outcome from an external/exit span ending. This is to ensure that the explicit setting to a span to unknown in the existing agent code continues to work. Freezing the span here should not interfere with any other calls to setOutcome, as the span immediately ends.
That covers my intent -- happy to talk through any other implications this might have.
lib/instrumentation/span.js
Outdated
| if (typeof statusCode !== 'undefined') { | ||
| if (statusCode >= 400) { | ||
| this.outcome = 'failure' | ||
| this.outcome = constants.SPAN.OUTCOME.FAILURE | ||
| } else { | ||
| this.outcome = 'success' | ||
| this.outcome = constants.SPAN.OUTCOME.SUCCESS | ||
| } | ||
| } | ||
|
|
||
| this._freezeOutcome() |
There was a problem hiding this comment.
Does this create a case where the outcome can be frozen to "unknown" and then the subsequent span.end() will not default the outcome to "success" in _setOutcomeFromSpanEnd()?
There was a problem hiding this comment.
nod that's my reading of this preexisting code (my change only moved the existing code from strings to constants and then froze the outcome) Interestingly, it's actually this behavior that (in part) led me to add the ability to generically freeze the outcome (vs. just preferring the API set value) -- Originally if a span had an unknown outcome and got to the end, I always changed it to be a success. However, this test failed with that behavior, so I added the freeze feature so we could freeze the outcome to unknown for this specific case.
Co-authored-by: Brandon Morelli <brandon.morelli@elastic.co>
Implements: #1814
Spec PR: elastic/apm#299
This PR adds an API for setting the outcome value on a span or a transaction. API methods are documented, and existing on Agent, Transaction, and Span object. See docs for full usage information. This required adjusting the existing setOutcome method.
This PR also adds default agent behavior to set an outcome of
failureon the current span when capturing an error.Checklist