Re: Updates Galore

["Jim McDonald" <Jim mcdee net>]

> On Thu, 2003-12-11 at 00:03, Ryan P Skadberg wrote:
>
>> The Clue Packet Manager is still crashing here and there.  Large
>> Segfaults during HTML and Text Chainers.  Haven't had a chance to look
>> at it yet, but these are the biggest crashes.  Also, seeing Null
>> Pointer Exceptions in the Text Indexer and RSS backends thus far.
>> Looks like we need to be checking for these more closely.
>
> The crashes occur because of the following sequence of events:
>
>         * Cluepacket comes in
>
>         * RunQuery in the CPM kills any outstanding running threads
>
>         * A new thread is launched for each backend
>
>         * One of those backends is a chainer.  it creates a new
>         cluepacket, and sends it out
>
>         * the new, chained cluepacket comes in
>
>         * RunQuery in the CPM kills any outstanding running threads:
> INCLUDING ITSELF!

The reason I didn't fix this one is that I'm not sure what the expected
behaviour should be.  Does the new clupacket supercede the old one
entirely (i.e. we shut down the outstanding backend requests and treat
this as the equivalent of new user input) or does it complement it (i.e.
we generate a new set of threads with for the clue but keep the old ones
around)?

   I suppose this leads to a slightly different questions, which is does
the today's model of saying that the latest clue is all-important and
any information from older cluepackets should be abandoned/ignored what
we want to happen?  In a situation where you have lots of frontends and
a busy user might this mean that dashboard is just continually throwing
up information without it staying on-screen for long enough to be of
any use?

> Nat

Cheers,
Jim.
-- 
Jim McDonald