Reevaluate manual retry logic

[vilterp]

I think we should just remove this if statement; it should be allowed to poll forever:

rai-sdk-go/rai/client.go

Line 1243 in 2b1fdd6

if count > 5 {

The Go-idiomatic thing might be to pass in a context for this function, and on each iteration check whether the context is cancelled. (though I'm not sure whether this should go cancel the transaction on the RAI side…)

cc @NRHelmi

[NHDaly]

(In the long-term, we want to implement the same polling policy in all the SDKs: https://github.com/RelationalAI/rai-sdk-issues/issues/58. But in the meantime, just removing this if statement would help.)

[NHDaly]

Oh, @NRHelmi, we need to fix this before merging RelationalAI/rai-cli#17, otherwise all transactions longer than 10 seconds will appear to time out at the CLI! 😮

[NRHelmi]

@NHDaly the count is there to abort if we exceed the 5 errors but I think it will poll forever if everything is running fine

[vilterp]

Hm, I guess the count only gets incremented if there's an error:

rai-sdk-go/rai/client.go

Line 1240 in 2b1fdd6

count++

This would be easier to see if the variable is called errorCount.

Should we just bail out at the first error? Is there any precedent for this "only bail out after 5 errors" policy in other SDKs?

[NHDaly]

ooooooooooooh i see! Thank you for clarifying!

So this is like a retry policy? I think that the Julia SDK handles retries automatically through the Julia SDK:
RelationalAI/rai-sdk-julia#45
RelationalAI/rai-sdk-julia#47

Does the Golang HTTP library do any kind of retries for us automatically already?

[NHDaly]

I guess this was added per this comment, here:
#2 (comment)

I dunno what's standard here. I agree that bailing on the first error seems reasonable to me?

[billscheidel-rai]

Note: This issue has been migrated to https://relationalai.atlassian.net/browse/RAI-6269.

This link is only accessible to employees of RelationalAI.