net/http: TestTransportReuseConnection_Gzip_.* failures with "didn't reuse connection"

[bcmills]

#!watchflakes
post <- pkg == "net/http" && test ~ `TestTransportReuseConnection_Gzip_.*` && `didn.t reuse connection` && date < "2023-08-30"

---

2022/06/08 19:53:00 http: TLS handshake error from 127.0.0.1:21458: EOF
--- FAIL: TestTransportReuseConnection_Gzip_ContentLength (0.10s)
    transport_test.go:4447: didn't reuse connection
FAIL
FAIL	net/http	11.607s

greplogs -l -e 'FAIL: TestTransportReuseConnection.*(?:\n .*)* didn'\''t reuse connection' --since=2022-01-01
2022-06-08T19:41:54-bdde41e/freebsd-amd64-race

See previously #22212.

Perhaps a good first step would be to modify the test to use net/http/httptrace to dump more information about what's going on with the connections?

(CC @neild @golang/freebsd)

[ayang64]

if noone else does, i'll take a look this afternoon. assigning to myself as a reminder.

[ayang64]

i can't reproduce; will try to set up the same environment: OS Version: FreeBSD 13.0-STABLE FreeBSD 13.0-STABLE #0 stable/13-n248759-3684bb89d52

[bcmills]

@ayang64, given that there is only one of these in the logs, this looks to me to be a very rare failure. I would not expect it to reproduce easily.

(That's why I suggest starting by modifying the test to dump more information — since this failure mode is so rare, ideally we should extract as much information as we can should it occur again.)

[bcmills]

This does not appear to be freebsd-specific:
greplogs -l -e 'FAIL: TestTransportReuseConnection_.*\n\s.*didn.t reuse connection' --since=2022-01-01
2022-08-16T16:27:35-a45bbea/darwin-amd64-nocgo
2022-08-12T16:38:52-f001df5/windows-arm64-11
2022-06-08T19:41:54-bdde41e/freebsd-amd64-race

Note that darwin/amd64 is a first-class builder (attn @neild).