Serviceguard and Oracle RAC

[Jonathan H.]

We are running HP-UX 11.31, Serviceguard 11.19 and Oracle RAC 11g R1 and we want to use active/standby lan failover and IP_MONITOR features of ServiceGuard for the cluster interconnect.

We currently use InfiniBand for our interconnects, lan9000 for primary and lan9001 for standby. However RAC uses the interface name during configuration and fails to follow the IP if the NIC fails-over to lan9001. Due to our configuration of nodes in multiple blade enclosures, we have two layers of InfiniBand switches and want to use IP_MONITOR instead of APA for better failure detection.

We also don't want to purchase SGeRAC due to its' costs and wonder if there is a way for the RAC interconnect to use SG active/standby fail-over.

I know we can use Oracle VIP for the public address but I was wondering about the private interconnect. I've found many documents showing SG as a valid fail-over option but do not know if there is something I need to change in Oracle or HP-UX to make the fail-over work.

Thanks,

[Turgay Cavdar]

Did you add standby lan card to RAC config?

#oifcfg setif -global lan9001/ip_address:cluster_interconnect

[Duncan Edmonstone]

I very much doubt that running plain Serviceguard with Oracle RAC is even supported...

Remember in this situation you effectively have 2 cluster managers vying for control of one set of cluster nodes, neither aware of the other - who decides when to eject a member? who decide when to carry out a cluster reformation? who decides when to TOC a node? This situation is I would guess untested and uncertified by either Oracle or HP (but you should seek clarification from your support contacts)

When SGeRAC is installed, SGeRAC effectively controls Clusterware, and where it can't sets Clusterware timeouts to values high enough that SGeRAC will always take action first.

I'm amazed the reason given here is to "save money", given the cost of RAC itself...

HTH

Duncan

[Stephen Doud]

SGeRAC provides the cmgmsd daemon, which is the means by which Oracle instances communicate with the Serviceguard cluster software. The cluster manager starts up, monitors, and shuts down the cmgmsd. When an Oracle instance starts, the instance registers itself with cmgmsd; thereafter, if an Oracle instance fails, cmgmsd notifies other members of the same group to perform recovery. If cmgmsd dies unexpectedly, Serviceguard will fail the node with a TOC (Transfer of Control).

Can you do without GMS (group membership daemon)?

[Jonathan H.]

We have decided to go with an Oracle RAC only solution