0->1 activation errors

[tcnghia]

Expected Behavior

0->1 should succeed.

Actual Behavior

When activating, we're seeing errors like

upstream connect error or disconnect/reset before headers

and

curl 35.196.32.235 -H 'Host: upper.default.example.com' -w '\n' -H 'Content-Type: text/plain' -d hello

curl: (18) transfer closed with 19 bytes remaining to read
Error: exit status 18

Steps to Reproduce the Problem

1. Scale to 0
2. Send a request

Additional Info

@trisberg & @mattmoor hit these issues in the last several days. We have not seen these for a while before recently.

[trisberg]

so I installed the no-mon release tagged v20180810-2eb54d8 and I'm not seeing this error now,it all looks fine so far, will try a later version to see if we can get closer to the date where it started

[trisberg]

looks like some change following the v20180810-2eb54d8 release is the culprit since the problem starts happening with the v20180811-ace9afa release. Maybe a side effect of #1814 or #1689?

[tcnghia]

We should make our 0->1 autoscaling test detect these issues so we don't regress. I filed #1878

[josephburnett]

@trisberg, did you requests have a body? If not, you might have hit a bug fixed by #1880 (attempt to read nil body)

[tcnghia]

@mattmoor ^

[trisberg]

@josephburnett it was a POST, I'll try again with the fix in #1880 to see what happens

[trisberg]

I built from latest master and still seeing the problem so #1880 didn't fix this specific issue.

Error I get is: curl: (18) transfer closed with 19 bytes remaining to read

[josephburnett]

Okay, thanks @trisberg. I'll start digging into this, starting by making the e2e verify success (#1878).