With enough servers, we sometimes crash in TCPSpaceNetwork::handleClosingStreamTimeout. I think it's a thread safety issue: the various methods modifying mClosingStreamTimers don't seem to be properly wrapped in a strand or using locks. However, I don't have the minidump anymore to verify this (and even if I did, that could be the issue regardless of whether it popped up in the stack trace).
A temporary work around of just returning immediately from that function resolved the issue. It's not that big a deal to not close out dead connections for now.
With enough servers, we sometimes crash in TCPSpaceNetwork::handleClosingStreamTimeout. I think it's a thread safety issue: the various methods modifying mClosingStreamTimers don't seem to be properly wrapped in a strand or using locks. However, I don't have the minidump anymore to verify this (and even if I did, that could be the issue regardless of whether it popped up in the stack trace).
A temporary work around of just returning immediately from that function resolved the issue. It's not that big a deal to not close out dead connections for now.