arif/xcat-core
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Further refinements of IPMI, defer retries until there is a guaranteed chance given for BMC to reply

Browse Source 
Ensure every transmit gives an opportunity to process incoming data

git-svn-id: https://svn.code.sf.net/p/xcat/code/xcat-core/trunk@13456 8638fb3e-16cb-4fca-ae20-7b5d299a9bcd
This commit is contained in:
jbjohnso 2012-08-08 19:52:02 +00:00
parent 7ea019857e
commit 9764d338cd
1 changed files with 28 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
xCAT-server/lib/perl/xCAT/IPMI.pm
View File

@ -372,18 +372,19 @@ sub subcmd {
sub waitforrsp {
my $self=shift;
my %args=@_;
my $data;
my $peerport;
my $peerhost;
my $timeout; #TODO: code to scan pending objects to find soonest retry deadline
my $curtime=time();
if (defined $args{timeout}) { $timeout = $args{timeout}; }
foreach (keys %sessions_waiting) {
if ($sessions_waiting{$_}->{timeout} <= $curtime) { #retry or fail..
my $session = $sessions_waiting{$_}->{ipmisession};
delete $sessions_waiting{$_};
$pendingpackets-=1;
$session->timedout();
next;
if (defined $timeout and $timeout == 0) { last; } #once we get to zero, then there is no lower and anything else is a waste
if ($sessions_waiting{$_}->{timeout} <= $curtime) {
$timeout=0; #this waitforrsp must go as quickly to retry as possible, but give it a chance this iteration to clear without timedout being called
#if something defferred entry into waitforrsp so long that there was no chance to check for response, this grants at least one shot at getting data
}
if (defined $timeout) {
if ($timeout < $sessions_waiting{$_}->{timeout}-$curtime) {
@ -414,6 +415,16 @@ sub waitforrsp {
}
}
}
foreach (keys %sessions_waiting) { #now that we have given all incoming packets a chance, if some sessions were past due when we entered
#take timeout response action now
if ($sessions_waiting{$_}->{timeout} <= $curtime) {
my $session = $sessions_waiting{$_}->{ipmisession};
delete $sessions_waiting{$_};
$pendingpackets-=1;
$session->timedout();
next;
}
}
return scalar (keys %sessions_waiting);
}
@ -427,7 +438,7 @@ sub timedout {
$self->{ipmicallback}->($rsp,$self->{ipmicallback_args});
return;
}
$self->sendpayload(%{$self->{pendingargs}});
$self->sendpayload(%{$self->{pendingargs}},nowait=>1); #do not induce the xmit to wait for packets, just spit it out. timedout is in a wait-for-packets loop already, so it's fine
}
sub route_ipmiresponse {
my $sockaddr=shift;
@ -819,8 +830,16 @@ sub sendpayload {
#push integrity data
}
}
while ($pendingpackets > $maxpending) { #if we hit our ceiling, wait until a slot frees up
$self->waitforrsp();
unless ($args{nowait}) { #if nowait indicated, the packet will be sent regardless of maxpending
#primary use case would be retries that should represent no delta to pending sessions in aggregate and therefore couldn't exceed maxpending anywy
#if we did do this on timedout, waitforrsp may recurse, which is a complicated issue. Theoretically, if waitforrsp protected itself, it
#would act the same, but best be explicit about nowait if practice does not match theory
#another scenario is if we have urgent payload for a BMC (PET acknowledge, negotiating login if temp session id is very short lived
$self->waitforrsp(timeout=>0); #the intent here is to interrupt outgoing activity to give a chance to respond to incoming data
#until we send, the ball is in our court so things are less time critical
while ($pendingpackets > $maxpending) { #if we hit our ceiling, wait until a slot frees up, which can't happen until either a packet is received or someone gives up
$self->waitforrsp();
}
}
$socket->send(pack("C*",@msg),0,$self->{peeraddr});
$sessions_waiting{$self}={};