ServiceWorker: remove kill in-progress jobs for update.

ServiceWorker: remove kill in-progress jobs for update.

When browser want to do update(defined in sepc as [[Soft Update]]), the
update should not stop a installing worker.

For example, a controlled page navigation will trigger update after 1
second, if it also register a new version before this update, the new
worker might become redundant.

BUG=443046, 404111
Committed: https://crrev.com/d375166c3abdff055b7b4c6fbead233a9b554163
Cr-Commit-Position: refs/heads/master@{#310708}

[xiang, 2015-01-08 07:39:08]

xiang.long@intel.com changed reviewers:
+ falken@chromium.org, michaeln@chromium.org

[xiang, 2015-01-08 07:39:08]

PTAL, thanks. I feel this's the cause of skipWaiting flaky, the
wait_for_state(t, registration.installing, 'activated') of the last test may
fail if update make it redundant.

[falken, 2015-01-08 08:15:09]

On 2015/01/08 07:39:08, xiang wrote:
> PTAL, thanks. I feel this's the cause of skipWaiting flaky, the
> wait_for_state(t, registration.installing, 'activated') of the last test may
> fail if update make it redundant.

Ohhh good catch! And sorry for the trouble.

I had the following use case in mind: a site accidentally deploys a bad service
worker, then tries deploys a good service worker after that. But the bad service
worker is stuck installing so it will never auto-update to the good service
worker.

But you're right that the spec disallows this behavior. I guess the site has a
workaround by explicitly calling register().

Can you add 404111 to the bug list? With that, LGTM.

[xiang, 2015-01-09 02:21:19]

On 2015/01/08 08:15:09, falken wrote:
> On 2015/01/08 07:39:08, xiang wrote:
> > PTAL, thanks. I feel this's the cause of skipWaiting flaky, the
> > wait_for_state(t, registration.installing, 'activated') of the last test may
> > fail if update make it redundant.
> 
> Ohhh good catch! And sorry for the trouble.
> 
> I had the following use case in mind: a site accidentally deploys a bad
service
> worker, then tries deploys a good service worker after that. But the bad
service
> worker is stuck installing so it will never auto-update to the good service
> worker.
> 
> But you're right that the spec disallows this behavior. I guess the site has a
> workaround by explicitly calling register().
> 
> Can you add 404111 to the bug list? With that, LGTM.

Done.

The current impl is still a little different from spec as the "update" job will
eventually be called after previous jobs are done, but in spec the "soft update"
will be aborted before entering the [[update]] algorithm. But I think it's OK
here, as the "soft update" could happen "as often as it likes".

[xiang, 2015-01-09 02:21:27]

The CQ bit was checked by xiang.long@intel.com

[commit-bot: I haz the power, 2015-01-09 02:22:02]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/822293003/1

[commit-bot: I haz the power, 2015-01-09 04:00:34]

Committed patchset #1 (id:1)

[commit-bot: I haz the power, 2015-01-09 04:01:31]

Patchset 1 (id:??) landed as
https://crrev.com/d375166c3abdff055b7b4c6fbead233a9b554163
Cr-Commit-Position: refs/heads/master@{#310708}