Ingress Error and drain()

Question

Ingress Error and drain()

mikekliger opened this issue 9 months ago · 3 comments

It is hard to follow, but do we expect an ingress error (purportedly on a transaction) to cause a session to be closed? This is with keep alive headers. We are utilizing proxygen::SessionPool

I am having connection churn issues under heavy load and can see a trail of logs before the session is drained (v =4)

IsReusable prints this, right before SessionHolder calls drain() on the session. I also see an ingress Timeout on this thread around the same time.

HTTPUpstreamSession.cpp:28] isReusable: proto=http/1.1, local=1redacted redacted=upstream, liveTransactions_=0, isClosing()=0, sock_->connecting()=0, codec_->isReusable()=1, codec_->isBusy()=1, numActiveWrites_=0, writeTimeout_.isScheduled()=0, ingressError_=1, hasMoreWrites()=0, codec_->supportsParallelRequests()=0 |

Answer 1 · 2024-03-18T14:55:21.000Z

Is there any way to allow the connection to stay open under these circumstances?

Answer 2 · 2024-03-18T16:02:41.000Z

Furthermore, is there a transactionTimeout callback we can utilize to gather info on the associated HTTP::upstreamSession

Answer 3 · 2024-09-18T20:29:45.000Z

Apologies this issue did not get an earlier response:

It sounds like the session is hitting the idle timeout, causing it to drain. You can set the idle timeout larger, or I believe setting it to 0 will disable idle timeout completely.

If there is an active transaction that is awaiting ingress (eg a client waiting for more of the response) and the transaction timeout fires, the handler is notified of the error via an HTTPException with ProxygenError set to "kErrorTimeout".