Service Worker: Detach the controller when falling back to network on failure

Service Worker: Detach the controller when falling back to network on failure

Otherwise, subresource loads would also go to the broken SW, defeating the
fallback mechanism. It also would skew start worker metrics if the SW fails
to startup on each requst.

The eviction mechanism already takes case of this in
ServiceWorkerRegistration::DeleteVersion, but we don't
currently evict everytime a fetch event fails to dispatch.

BUG=498661, 448003
Committed: https://crrev.com/e46ff16ad6df1050d357009e5041ba90aa66c983
Cr-Commit-Position: refs/heads/master@{#333713}

[falken, 2015-06-10 08:00:50]

falken@chromium.org changed reviewers:
+ kinuko@chromium.org, nhiroki@chromium.org

[falken, 2015-06-10 08:00:50]

Discovered this when doing local testing for what happens if a SW is
"NOT_FOUND".

[nhiroki, 2015-06-10 08:14:41]

Great! LGTM!

> but we don't currently evict everytime a fetch event fails to dispatch.

Just question) Will we evict in the case? (I wonder if we also need to evict a
worker when event dispatch other than fetch fails)

[kinuko, 2015-06-10 08:19:19]

lgtm

[falken, 2015-06-10 08:22:08]

On 2015/06/10 08:14:41, nhiroki wrote:
> Great! LGTM!
> 
> > but we don't currently evict everytime a fetch event fails to dispatch.
> 
> Just question) Will we evict in the case? (I wonder if we also need to evict a
> worker when event dispatch other than fetch fails)

Yes it might be a good idea. The problem is it could be harsh since we don't
know if failing to dispatch may just be a transient error. It feels like the
error cases other than DISK_CACHE shouldn't be persistent errors. To start off
I'll try to add more UMA around URLRequestJob class.

[falken, 2015-06-10 08:27:54]

The CQ bit was checked by falken@chromium.org

[commit-bot: I haz the power, 2015-06-10 08:27:57]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1173943002/1

[commit-bot: I haz the power, 2015-06-10 10:16:43]

Committed patchset #1 (id:1)

[commit-bot: I haz the power, 2015-06-10 10:17:48]

Patchset 1 (id:??) landed as
https://crrev.com/e46ff16ad6df1050d357009e5041ba90aa66c983
Cr-Commit-Position: refs/heads/master@{#333713}