Joern Nettingsmeier schrieb:
> Andreas Hartmann wrote:
>> Bob Harner schrieb:
>>> On 5/24/07, Andreas Hartmann <[EMAIL PROTECTED]> wrote:
>>>> Hi Lenya devs,
>>>>
>>>> what should we do about this issue?
>>>>
>>>> http://issues.apache.org/bugzilla/show_bug.cgi?id=42510
>>>>
>>>> I wouldn't like to implement a queue for incremental indexing
>>>> events before 1.4 is out, because I think it's quite a lot of
>>>> work (especially in the testing department), and I can't predict
>>>> the consequences without giving it some thought. Maybe someone
>>>> has a solution at the ready (crossing my fingers ...).
>>>>
>>>> Should we silently ignore the error and just don't trigger
>>>> the indexing, or should we continue to throw the exception?
>>>>
>>>> I hope someone comes up with a better idea :)
>>
>> [...]
>>
>>> A better interrim solution might be to display a helpful warning
>>> message rather than an exception:
>>>
>>> "Warning:  this document can't be added to the search index yet
>>> because the indexer is currently busy with another document.  Please
>>> re-publish this document in a moment to ensure that it is indexed."
> 
> +1
> 
>> The problem with this approach is that we can't determine if the
>> indexer will be busy before we apply the change to the document.
>> Since the indexer is a shared resource, we'd have to lock it to
>> prevent concurrent tasks from starting an indexing process while
>> the publishing (or any other action which changes the document
>> content) is in progress.
> 
> why? bob's suggestion means it can fail, but the user will be given a
> workaround. sounds ok as an interim solution.

The advantage is that the user knows that un-indexed documents
exist (or are published), but she still has to deactivate and re-publish
them. Anyway, I don't think we will be able to achieve anything better
for the moment.

BTW, I don't see how to implement this since repo observation (which is
used for indexing) is asynchronous ...


>> I'd be interested how other systems handle this. Maybe the indexing
>> has to be part of the transaction, so the transaction can be rolled
>> back if the indexing fails. But maybe we shouldn't invest too much
>> research in this issue but rather choose a powerful back-end which
>> supports indexing for the next major version.
> 
> i'd say let's ignore concurrency issues for 1.4 and document the
> shortcomings. we need to get this one out. without being negative, i
> think that most users that are eagerly waiting for a release have small
> to medium-size deployments and will only rarely encounter concurrency
> issues - we just don't have the track record atm to be considered for
> very large scale projects. let's not starve our core users too much by
> delaying 1.4 any further.
> instead, we should put up a roadmap where concurrency is an important
> topic for 1.5. incremental improvements - otherwise we'll die of
> second-system syndrome.

I agree. If anyone has an idea how to implement Bob's suggestion,
feel free to go ahead or make a proposal. I'll try to think about
it too when I find the time.

-- Andreas


-- 
Andreas Hartmann, CTO
BeCompany GmbH
http://www.becompany.ch


---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to