when doing continuous runs, export stats periodically

[ldemailly]

when running with constant traffic, no stats are emitted until "the end" which isn't that useful

it's also related to #134

[nkuacac]

expecting this feature。
we encounted a problem. Testing short connection though loadbalancer for 120s， with some connection 'time out' after 4min，the final summary qps is half of the exact qps

[ldemailly]

While periodic export for long run is useful, I don't think it is related to:

we encounted a problem. Testing short connection though loadbalancer for 120s， with some connection 'time out' after 4min，the final summary qps is half of the exact qps

Set the timeout lower?

[nkuacac]

If we have periodic export, we can get the real qps. And you mean the client timeout option -- the 15s default ?

While periodic export for long run is useful, I don't think it is related to:

we encounted a problem. Testing short connection though loadbalancer for 120s， with some connection 'time out' after 4min，the final summary qps is half of the exact qps

Set the timeout lower?

[ldemailly]

I would try to understand why you are getting timeouts and why that impacts your expected QPS?

Periodically reporting would not change the math; the qps is how many requests were completed (ie finished, including the reply or the timeout) divided by the time; if it's constant whether you have 1200 requests in 20 minutes or 600 requests in 10 minute it's still 1.0 qps - the goal of this feature is to report maybe every few minutes for say a 1h run; it wouldn't change your 120s test.

[nkuacac]

Timeouts always happen as the pressure, but just a little connect, the majority finish in time. If Periodically reporting, Maybe it with show first half period is huge, then little the next half?

[ldemailly]

increase the -c if you have a few bad requests blocking the other ones. or figure out why that's happening? but either it's unrelated to this specific issue. thanks.

[nkuacac]

The timeout bugs, we will try to figure out. If there is a periodically report, we can find the bug effectively.

[ldemailly]

The timeout bugs, we will try to figure out. If there is a periodically report, we can find the bug effectively.

I agree! in the meantime remember, you can see errors output in the log as they occur

[ldemailly]

also now available through fortiotel's traces/spans