Thanks all for trying to help me out!It seems like I found the culprit, somebody googling for the same problem might find this post useful.
The 18.104.22.168 32 bit linux ons.config uses "useocr=true" to get the values for ONS such as local and remote port as well as privileges for the keys DATABASE.ONS_HOSTS.<hostname> and DATABASE.ONS_HOSTS.<hostname>.PORT. In my case the root scripts did not add values for the 3rd node (called node3 in my cluster) to the OCR - when ons tried to start it could only find the existing configuration for node1 and node2 which it successfully read (see $ORA_CRS_HOME/log/<hostname>/racg/ora.<hostname>.ons: "RACG...number of onsconfiguration retrieved, numcfg = 2". However it was blissfully unaware of node3's configuration. RACG should have returned numcfg = 3 instead of 2.
In retrospect I could have tried to add in the ons configuration for the 3rd node as user oracle running racgons add_config hostname:remotePort to save myself the hassle of reinstalling. Alternatively I could have tried setting all the ons.config parameters in the configuration file rather than in the OCR but that's too late for testing this now...
Oh, btw, applying 22.214.171.124 did not solve the problem but how could it fix logical corruption in the OCR?
Cheers, Martin Bach After I completely reinstalled clusterware across all 3 nodes, David Ballester wrote:
El jue, 18-09-2008 a las 19:17 +0100, Martin Bach escribió:Hi all,
Naturally there is no documentation about this function, neither seems anybody else affected (couldn't find anything on google/metalink).Regards,Do you have changed the kernel values ( segments ) and the ulimit values for the oracle software owner ? Have you tested your hardware ( memory errors? ) RegardsD.