Re: Different mojo channel policy between Blink and Content?

[Ming-Ying Chung]

On Thu, Jun 13, 2024 at 11:39 AM Ming-Ying Chung <my...@chromium.org> wrote:

Hi Chromium-mojo Team,

I am investigating a bug potentially caused by a different mojo sending policy on renderer destructing.

Context

The fetch keepalive migration routes every keepalive request handling via the browser process before reaching network service, i.e.

- Before: renderer -> network service

- After: renderer -> browser -> network service

The mojo interface network::mojom::URLLoaderFactory used by renderer are all created by browser in RenderFrameHostImpl and get passed to renderer after CommitNavigation(). The only difference is that they point to different places before/after migration.

Questions

We expect that mojo made and queued by renderer can be eventually sent to the browser process, even if a renderer is unloaded, such that the request handling can be continued by browser. However, we observe that there are around 2-5% requests not being able to reach the browser.

Does renderer treat -> network service mojo messages differently on unload than other -> browser messages?