AU: Implement server-dictated poll interval.

AU: Implement server-dictated poll interval.

The server will need to include a PollInterval XML attribute in its
update check response. The requested interval is in seconds.

BUG=5984
TEST=unit tests, gmerged on device and tested with a modified dev server

Change-Id: I89d13f9f85d93bc141b74ae677cca813e3364fb5

[adlr, 2010-09-01 02:51:52]

LGTM w/ nits

http://codereview.chromium.org/3275006/diff/4001/5001
File omaha_request_action.cc (right):

http://codereview.chromium.org/3275006/diff/4001/5001#newcode384
omaha_request_action.cc:384: &output_object.poll_interval);
should there be some sanity checks here? like in the range [60 seconds, 1 week]?

http://codereview.chromium.org/3275006/diff/4001/5005
File update_check_scheduler.cc (right):

http://codereview.chromium.org/3275006/diff/4001/5005#newcode97
update_check_scheduler.cc:97: LOG(WARNING) << "Using server-dictated poll
interval: " << interval;
should we have a minimum?

http://codereview.chromium.org/3275006/diff/4001/5005#newcode98
update_check_scheduler.cc:98: } else if (update_attempter_->http_response_code()
== 500 ||
should we treat all 5xx as error? (not jut 500/503)

[petkov, 2010-09-01 06:02:20]

See responses below. If you don't object, I'll push tomorrow morning.

http://codereview.chromium.org/3275006/diff/4001/5001
File omaha_request_action.cc (right):

http://codereview.chromium.org/3275006/diff/4001/5001#newcode384
omaha_request_action.cc:384: &output_object.poll_interval);
On 2010/09/01 02:51:52, adlr wrote:
> should there be some sanity checks here? like in the range [60 seconds, 1
week]?

I thought it's best to transfer it as-is to the scheduler. The scheduler will
limit it to a reasonable range.

http://codereview.chromium.org/3275006/diff/4001/5005
File update_check_scheduler.cc (right):

http://codereview.chromium.org/3275006/diff/4001/5005#newcode97
update_check_scheduler.cc:97: LOG(WARNING) << "Using server-dictated poll
interval: " << interval;
On 2010/09/01 02:51:52, adlr wrote:
> should we have a minimum?

We do. See below -- if it's less than our regular periodic interval, we use the
periodic interval instead.

http://codereview.chromium.org/3275006/diff/4001/5005#newcode98
update_check_scheduler.cc:98: } else if (update_attempter_->http_response_code()
== 500 ||
On 2010/09/01 02:51:52, adlr wrote:
> should we treat all 5xx as error? (not jut 500/503)

I've done what the rate limit site suggests -- 500/503 for sure, and 502 with
the caveat that you have to be sure the code comes from the update server rather
than from a proxy. 502 needs more investigation.

I'd say let's leave it as is and consider extending it if we run into issues.