Gadget 1.20

[scott-rc]

This PR shows the diff between our 1.20 fork and main (Fission 1.20.1)

I cherry picked every commit from our 1.19 branch except for:

• Increase K8S client QPS
  • This was implemented in Fission 1.20
  • They used the exact same values
  • Add kubernetes QPS and Burst limit configuration fission/fission#2899
• Use different ctx when shutting down telemetry
  • This was implemented in Fission 1.20
  • They use the same timeout (30s)
  • They only use a new context if the passed in one is already cancelled (we always used a new context)
  • https://github.com/fission/fission/pull/2878/files#diff-b815b939484b14ee73bb3382c1bc1b8f354baee7047600a64ee1d471fefaf432
• Fix TestPublisher test
  • This was only necessary for the commit above
• Don't send 429 when max concurrency is reached
  • This was reverted and never shipped

[angelini]

This was implemented in Fission 1.20
They use the same timeout (30s)
They only use a new context if the passed in one is already cancelled (we always used a new context)

I think the only worry here is what if the ctx is about to expire (instead of having already expired). If the deadline just has 500ms left, it will fail the first time we try and use it.

Does this code get retried in case of a ctx deadline timeout?

[scott-rc]

This was implemented in Fission 1.20
They use the same timeout (30s)
They only use a new context if the passed in one is already cancelled (we always used a new context)

I think the only worry here is what if the ctx is about to expire (instead of having already expired). If the deadline just has 500ms left, it will fail the first time we try and use it.

Does this code get retried in case of a ctx deadline timeout?

Good point, they don't retry if it fails:

• https://github.com/gadget-inc/fission/blob/main/cmd/fission-bundle/main.go#L216-L223
• https://github.com/gadget-inc/fission/blob/main/cmd/fetcher/app/server.go#L65-L71

Want me to change this to always use the new context?

EDIT: I made a PR for this here