net/http: TestTransportServerClosingUnexpectedly failures

[gopherbot]

#!watchflakes
post <- pkg == "net/http" && test == "TestTransportServerClosingUnexpectedly"

Issue created automatically to collect these failures.

Example (log):

--- FAIL: TestTransportServerClosingUnexpectedly (0.00s)
    --- FAIL: TestTransportServerClosingUnexpectedly/h1 (0.65s)
        transport_test.go:829: expected body1 and body2 to be equal
1 second passes in backend, proxygone= false

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- pkg == "net/http" && test == "TestTransportServerClosingUnexpectedly"

2022-12-01 17:28 freebsd-arm-paulzhol go@34ab0bcc net/http.TestTransportServerClosingUnexpectedly (log)

--- FAIL: TestTransportServerClosingUnexpectedly (0.00s)
    --- FAIL: TestTransportServerClosingUnexpectedly/h1 (0.65s)
        transport_test.go:829: expected body1 and body2 to be equal
1 second passes in backend, proxygone= false

2022-12-01 22:04 openbsd-386-72 go@6a70292d net/http.TestTransportServerClosingUnexpectedly (log)

--- FAIL: TestTransportServerClosingUnexpectedly (0.01s)
    --- FAIL: TestTransportServerClosingUnexpectedly/h1 (0.07s)
        transport_test.go:829: expected body1 and body2 to be equal

— watchflakes

[bcmills]

(attn @neild; see previously #32119)

I don't see anywhere near enough information in the test log to diagnose the problem: it doesn't describe how body1 and body2 differ and doesn't even mention how they were created.
(This test isn't even close to meeting https://go.dev/wiki/CodeReviewComments#useful-test-failures!)

---

The test implementation is here:
https://cs.opensource.google/go/go/+/master:src/net/http/transport_test.go;l=814-815;drc=f4b42f5cb8b494a2b64a4ade89f89071d80c7fd3

The test uses hostPortHandler as the handler, and that writes the client's host and port as the body. If the idle connection from the first fetch call is not returned synchronously to the connection pool, there is a chance that it will not be reused for the second call, which would result in an observable difference in the response bodies.

Given the other net/http bugs involving connection reuse, that's my best guess for this one as well.

Compare:

• net/http: TestTransportReuseConnection_Gzip_.* failures with "didn't reuse connection" #53373
• net/http: TestKeepAliveFinalChunkWithEOF failures #57604
• net/http: TestTransportNoReuseAfterEarlyResponse failures #57608
• net/http: TestTransportMaxPerHostIdleConns failures #57476
• net/http: TestServerKeepAliveAfterWriteError failures #57376

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- pkg == "net/http" && test == "TestTransportServerClosingUnexpectedly"

2023-01-18 19:59 freebsd-arm-paulzhol go@9894ded1 net/http.TestTransportServerClosingUnexpectedly (log)

--- FAIL: TestTransportServerClosingUnexpectedly (0.01s)
    --- FAIL: TestTransportServerClosingUnexpectedly/h1 (0.16s)
        transport_test.go:829: expected body1 and body2 to be equal

2023-01-20 21:41 freebsd-arm-paulzhol go@cf70d379 net/http.TestTransportServerClosingUnexpectedly (log)

--- FAIL: TestTransportServerClosingUnexpectedly (0.00s)
    --- FAIL: TestTransportServerClosingUnexpectedly/h1 (0.53s)
        transport_test.go:829: expected body1 and body2 to be equal

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- pkg == "net/http" && test == "TestTransportServerClosingUnexpectedly"

2023-01-28 19:00 freebsd-arm-paulzhol go@915af1b7 net/http.TestTransportServerClosingUnexpectedly (log)

--- FAIL: TestTransportServerClosingUnexpectedly (0.00s)
    --- FAIL: TestTransportServerClosingUnexpectedly/h1 (0.82s)
        transport_test.go:829: expected body1 and body2 to be equal

2023-02-01 19:47 linux-amd64-race go@ab0f0459 net/http.TestTransportServerClosingUnexpectedly (log)

--- FAIL: TestTransportServerClosingUnexpectedly (0.00s)

— watchflakes