error page when google search fails is unusable

[Curtis Hovey]

Connectivity problems between Launchpad and Google can cause a TimeoutError. The user is shown a special page to explain the problem. This page does not inform the user of exact matches that might be made from the search nor help the user take the next step.

The error page does not include a search form with the search terms pre-filled, nor are there instruction to reload the page. the only action the user may take in the UI is to visit the front page. Regardless of the failure to locate matching pages from Google, other search for users, projects, bugs, and questions may match the search terms. I might be better let the view handle the error, and have the standard search page handle the display of problem.

Related OOPSes: OOPS-979A991, OOPS-979C1034, OOPS-979C1040, OOPS-979E960, OOPS-979F907, OOPS-979F908, OOPS-979G980, OOPS-979H874, OOPS-1539A1074, OOPS-1539B1178 (HTTPError: HTTP Error 500: Internal Server Error)
The same problem that caused the above HttpError, caused this another one, that is LP to oops when google returns an empty xml, as seen in OOPS-979B1022, OOPS-979C1028 and OOPS-979H877 (IndexError: list index out of range). LP should, as it should in the related HttpError above, recover gracefully.
Also: OOPS-1552A1017 (URLError: <urlopen error (110, 'Connection timed out')>)

[Martin Albisetti (beuno)]

I wonder how long you would typically need to retry?
Are we taking seconds? minutes? hours? $random?

As you state in the bug report, taking the user back to the search page with the input pre-filled, and inviting him to search again makes sense, if we're talking about random or seconds in timeouts.

If it's normally more than that, then an error page may be more appropriate, although we may be able to tweak the wording and the way to display it.

[Curtis Hovey (sinzui)]

On Mon, 2008-09-08 at 17:56 +0000, Martin Albisetti wrote:
> I wonder how long you would typically need to retry?
> Are we taking seconds? minutes? hours? $random?

I think we have seen a 6 minute outage. The one that caught my attention
was about 1.5 minutes.

> As you state in the bug report, taking the user back to the search page
> with the input pre-filled, and inviting him to search again makes sense,
> if we're talking about random or seconds in timeouts.

The two timeout instance I have investigated were not single instances
of failure.

> If it's normally more than that, then an error page may be more
> appropriate, although we may be able to tweak the wording and the way to
> display it.

Launchpad search is a composite of 7 lookups/searches. Only the Google
search failed. Should we show the 6 other searches? We also know how to
encode the search terms to create a link to Google--the network problems
were between the DC and Google in the problems we have seen.

[Martin Albisetti (beuno)]

On Mon, Sep 8, 2008 at 3:35 PM, Curtis Hovey <email address hidden> wrote:
> On Mon, 2008-09-08 at 17:56 +0000, Martin Albisetti wrote:
>> I wonder how long you would typically need to retry?
>> Are we taking seconds? minutes? hours? $random?
>
> I think we have seen a 6 minute outage. The one that caught my attention
> was about 1.5 minutes.

So, in general, it doesn't make sense to make the user retry, since
they will get the same results for a while.

> Launchpad search is a composite of 7 lookups/searches. Only the Google
> search failed. Should we show the 6 other searches? We also know how to
> encode the search terms to create a link to Google--the network problems
> were between the DC and Google in the problems we have seen.

Ah! I see now.
So, I do think you should show the results, and add a small ignorable
legend telling the user some results are temporarily missing due to
temporary problems.

Does that sound good?

--
Martin

[Francis J. Lacoste (flacoste)]

That happens so little, that I don't think it's much worth optimizing.

[Curtis Hovey (sinzui)]

I agree that we should not spend a lot of time on this issue. I am irked that when the problem does happen, we are not helping the user. I hope for a simple fix that is worth doing.

This problem is neither a programming error which we expect of an OOPS or a form error where we reshow the form and ask the user to correct the input. This is a case where I expect to show the page with the information we have and allow the user to retry.

[Taras Halturin (halturin)]

please, look at this...

https://code.launchpad.net/~vcs-imports/gsql/trunk

i can not even import my project :(.

[Tom Haddon (mthaddon)]

This just caused an intermittent site wide outage on LP for approx 1 hour

[Francis J. Lacoste (flacoste)]

Tom, the bug that caused the outage is bug 376560. This one is about putting out a better result page when Google fails (and we catch it succesfully).

[Curtis Hovey (sinzui)]

Opps closed the wrong bug. Reopening it.

[Launchpad QA Bot (lpqabot)]

Fixed in stable r12532 <http://bazaar.launchpad.net/~launchpad-pqm/launchpad/stable/revision/12532>.

[Launchpad QA Bot (lpqabot)]

Fixed in stable r12544 <http://bazaar.launchpad.net/~launchpad-pqm/launchpad/stable/revision/12544>.