I'm failing to reproduce this. It may be an infrastructure bug after all.

To be clear: it reproduces consistently in the lab when it's talking with the hdmi output recorder. it does not repro when I connect it up to a tv.

I've attached media_internals for the failing case (src=) and for the passing cause (MSE)

Deleted: src_media_internals.html 13.2 KB

src_media_internals.html 13.2 KB View Download

Deleted: mse_media_internals.html 10.2 KB

mse_media_internals.html 10.2 KB View Download

I just checked the chrome debug log. For SRC=, it has the following statement:

[17353:17159:1117/184415:WARNING:audio_output_device.cc(316)] Output device authorization timed out

That statement does not exist in the MSE log

Adding olka@ who added this useful line of logging! Thanks olka@!

olka added this line because of crbug/626862

https://codereview.chromium.org/2413653002, written by guidou@ is a potential suspect. 

I ran a test at commit position #424981, right before guidou@'s change https://codereview.chromium.org/2413653002 . For that the video completely failed to play at all.

I ran another test at commit position #424983 a revision that contained both guidou@'s change and another unrelated change and the video played and had sound. 

I ran two tests against 56.0.2924.0. For one, the video played but did not have sound. for the other the video played and did have sound.

Hi crouleau@, could you make sure that there is "Output device authorization timed out" message in the log when video does not have sound?
Could you also attach chrome://histograms/ tab content?

Thanks!

Thanks for looking into this Olka! I don't know if it's a real bug since it may be intended. Not sure whether having the video play without audio is better than no video at all in this case. Perhaps we could work in a retry to get audio during video playback when there was a failure. I imagine that MSE is somehow retrying more for this than src=.

Here's a Sponge log from last night's run (which was against Canary 57.0.2926.0): https://sponge.corp.google.com/target?tab=Output+Files&target=//chrome/videostack/testing/av_analysis:macbook_corem_5y31_intelhd5300_fullscreen&id=30bb69b6-6415-4b6a-ab87-a03290437c61&sortBy=STATUS&show=ALL&searchFor=user:videostack

If you go to "Output Files" there are a list of tests. Each test has a folder called "0". Expand that folder and you can see all the different logs that have been generated (including chrome_debug.log, media_internals.html, and chrome_info.html). The tests that have the suffix "MSE" are MSE tests. The tests without that suffix are simple src= ("<video src='example.mp4'></video>" for example) tests.

As mentioned in #3, we do see the the "Output device authorization timed out" message in chrome_debug.log when we do not have audio. Last night, all the src= videos did not have audio. Here is the dashboard that shows those results: https://av-analysis.corp.google.com/#/get-testpasslist/2/Video%20Stack?triggered_by=videostack&device=macbook_corem_5y31_intelhd5300&profile=src&from=1479830400&to=1479916800 (yeah, I know it would be better if these results and the Sponge results were merged together into one dashboard). In Profile field, change value from 'src' to 'mse' to view MSE results. The absence of an "Audio" score in an entry on the table indicates that the result did not have audio. You can see that all the mse test had audio but all the src tests did not.

Since these are automated tests, I will need to write some code to gather chrome://histograms/ . Are you sure you need this? With crbug/602295 I worry that the information that it contains may not be reliable for all purposes.

system: macbook_corem_5y31_intelhd5300 + hdmi output recorder.

After offline discussion with olka@, I wrote and submitted code to log the histograms after test runs.

I was unable to get this to repro in my manually-triggered tests despite doing things like rebooting the machine, unplugging and replugging the HDMI output, etc.

All the tests last night did not repro this problem: 

https://av-analysis.corp.google.com/#/get-testpasslist/2/Video%20Stack?device=macbook_corem&triggered_by=videostack&from=1479920400&to=1480006800

The tests will run again tonight.

The tests keep running, but we aren't having this error anymore.

Can't reproduce. I'll reopen this if it happens again.