UserAgent: Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Ubuntu Chromium/53.0.2785.143 Chrome/53.0.2785.143 Safari/537.36
Platform: veyron_minnie

Steps to reproduce the problem:
1. Check out chromium sources
2. cd to src/tools/perf
3. Run: ./run_benchmarks --browser=cros-chrome --remote=<location of a chromebook running a test image> blink_perf.svg --output-format=json

What is the expected behavior?
The blink_perf SVG benchmark should run on the chromebook via Telemetry and return successfully.

What went wrong?
The benchmark is not loaded, because the chromebook is unable to connect to 127.0.0.1.

Commit 85e79ec2557c6189b556cf64b438dc8f1af7d0ac of third_party/catapult disabled the SSH forwarding of the MemoryCacheHTTPServer port to the chromebook, which is how benchmarks like blink_perf were hosted. The commit message mentions that all benchmarks should go via ts_proxy now.

ts_proxy is still working, and the browser is configured to use the correct port for it, which is also forwarded over SSH. However, it seems like Chrome is refusing to use the proxy server when accessing 127.0.0.1 or localhost, on Chrome OS. If you change the URL that MemoryCacheHTTPServer reports for Telemetry to use to access the benchmark to anything that isn't localhost (but with the same port), then it hits the proxy server. Since the ts_proxy is configured to redirect everything to localhost anyway, it reaches the MemoryCacheHTTPServer and accesses the benchmark.

It's not clear if this is expected behaviour of Chrome on Chrome OS - should requests to localhost/127.0.0.1 use the proxy server or not? And if they shouldn't, Telemetry needs to be changed to host the blink_perf benchmarks correctly again.

Did this work before? Yes Chromium src prior to commit 9dc1b183cbad899d12728646cf2353e6d3ebff22 (Fri Nov 4 20:16:09 2016 -0700) (Roll src/third_party/catapult/ 3660f3bfa..85e79ec25 (1 commit).)

Chrome version: 56.0.2915.0  Channel: n/a
OS Version: 
Flash Version: