I have seem some similar things later on in the shutdown logic where the stack shows null, and it seems that client_vc has been nulled. https://github.com/apache/trafficserver/pull/4147. In that case the client_vc had been closed without updating the reference to it in the Http2ClientSession object. Not the same thing here, but there are some odd races going on.
I agree @masaori335's observation that just doing another null check is unlikely to help, since we have already done the null check in the current line of execution. I've spent this week chasing down a couple race conditions. After they have baked for the weekend, I will create PR's or update existing WIP PRs. [ Full content available at: https://github.com/apache/trafficserver/pull/4158 ] This message was relayed via gitbox.apache.org for [email protected]
