fix: don't add listeners to the same signal more than once

[tsctx]

No description provided.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 93.75000% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 94.16%. Comparing base (24f7ee6) to head (5009498).

Files	Patch %	Lines
lib/api/abort-signal.js	93.75%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3174      +/-   ##
==========================================
- Coverage   94.16%   94.16%   -0.01%     
==========================================
  Files          90       90              
  Lines       24385    24407      +22     
==========================================
+ Hits        22962    22982      +20     
- Misses       1423     1425       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ronag]

What is this solving?

[tsctx]

#3131

[Uzlopak]

But isnt it a bug in the elasticsearch client too, that it tries to add again and again the same listeners? Not saying we should not solve it here. But now it hints to a bug, but after we merge this you dont get a hint that something is wrong.

[ronag]

I don't think we should merge this. It doesn't solve an actual bug. It just bypasses the warning.

[ronag]

Se above

[tsctx]

But isnt it a bug in the elasticsearch client too, that it tries to add again and again the same listeners?

elasticsearch client is not the cause of the problem, it is undici-specific. In fact, it does not leak.

Not saying we should not solve it here. But now it hints to a bug, but after we merge this you dont get a hint that something is wrong.

When a signal was given and a large number of requests were sent at once (easy to see in the test), listeners were added more than the limit and a warning was generated.

[metcoder95]

Not saying we should not solve it here. But now it hints to a bug, but after we merge this you dont get a hint that something is wrong.

When a signal was given and a large number of requests were sent at once (easy to see in the test), listeners were added more than the limit and a warning was generated.

I am not sure if we should consider that an undici problem. This issue is potentially everywhere and it seems we are trying to protect the users from themselves when they should be aware of how they handle their signals and others.
No strong opinion on this, but I'd advise considering if worth the complexity given this is more a problem of how is AbortSignal used by the user rather than by undici.

[tsctx]

Oh, you are right.