Pattern for hedged requests #4
Conversation
| responses := make(chan T) // unbuffered, we only expect one response | ||
| ctx, cancel := context.WithCancelCause(ctx) | ||
|
|
||
| group := ErrGroup(Limited(ctx, cfg.numHedgedRequests+1)) |
There was a problem hiding this comment.
because we don't send an unbounded number of tasks to the group, it can be Unlimited for cheaper. A Limited group with N max parallelism that only ever receives N tasks doesn't do anything different.
| } | ||
|
|
||
| go func() { | ||
| _ = group.Wait() |
There was a problem hiding this comment.
putting group.Wait() here in a bare, unmanaged goroutine will hard-crash the entire process if any panic happens in any of the worker goroutines.
rather than spawning a loose goroutine to Wait(), it is much more preferable to write the result value to a captured var here using a sync.Once, calling .Wait() here in the function, checking its error, and then managing what (if anything) was put into that return value var.
|
|
||
| res, err := requester(ctx) | ||
| if err != nil { | ||
| return err |
There was a problem hiding this comment.
is the plan to return the first-returned value-or-error from any attempted request? To return any successful value if any had no error? to return a value if we got one but include any error that occurred in any of the other attempts, including context canceled?
There was a problem hiding this comment.
First return value or error from any attempted request
No description provided.