Stewart Smith Write:
> On Thu, 2007-07-05 at 17:12 +0800, Guangbao Ni wrote:
>
>> +void Dblqh::execSTART_LCPIMME_CONF(Signal* signal)
>> +{
>> + jamEntry();
>> + lcpStartImmeResponded = true;
>> + return;
>> +}
>>
>
> Should we reset lcpStartImmeResponded at end of LCP instead?
>
>
no, lcpStartImmeResponded here just means that the master DIH has
received the START_LCPIMME_REQ.
> Otherwise (i think) we could keep sending START_LCPIMME_REQ while one is
> already ongoing if we keep filling up the log?
>
>
This is a good question. I have considered it.
when LCP is already ongoing it can't keep sending STAR_LCPIMME_REQ.
Because if lcp is going, then
clcpCompletedState != LCP_IDLE. Only if clcpCompletedState == LCP_IDLE,
STAR_LCPIMME_REQ signal can be sent
> Also, I think it should be reset on master failure so we send to new
> master.
>
>
Right, I will add the code.
> Perhaps when we send START_LCPIMME_REQ we should also create a cluster
> log event that we forced an LCP start due to out of log.... that way
> user can see from examining log if this is a frequent problem.
>
okay, this is a good suggestion.
Thank you! i will mail the new patch to you after fixing it.
