在bdump下产生大量trace file:
Oracle9i Enterprise Edition Release 9.2.0.8.0 - 64bit Production
With the Partitioning and Real Application Clusters options
JServer Release 9.2.0.8.0 - Production
ORACLE_HOME = /app/oracle/product/9.2.0
System name: AIX
Node name: SID_db1
Release: 3
Version: 5
Machine: 00C49F1F4C00
Instance name: SID1
Redo thread mounted by this instance: 1
Oracle process number: 0
782444
Disabling Transparent Network Failover Feature.
在metalink 上找一下
Subject: | RAC on AIX: Number Of Serviceable Networks Detected Is Less Than 2 | |||
Doc ID: | Note:279054.1 | Type: | PROBLEM | |
Last Revision Date: | 10-JUN-2005 | Status: | PUBLISHED |
The information in this document applies to:
Oracle Server - Enterprise Edition - Version: 8.1 to 9.2.0
AIX5L Based Systems (64-bit)
AIX Based Systems (32-bit)
AIX 4.3 Based Systems (64-bit)
Symptoms
1) The alert-log contains the following messages are created on AIX/HACMP for OPS and RAC:
Number of Serviceable Networks detected is less than 2
Disabling Transparent Network Failover Feature.
2) Tracefiles are created with the same message.
3) No network errors are visible at the OS
4) oradebug ipc shows just one network being available to Oracle
Cause
A) The cluster configuration just had two networks defined:
# usr/sbin/cluster/utilities/cllsif -S
db1tty service dbserial rs232 serial db1 /dev/tty1
db1 service gbether ether public db1 10.200.15.10 en6
db2tty service dbserial rs232 serial db2 /dev/tty1
db2 service gbether ether public db2 10.200.15.12 en6
You can see all interfaces in the cluster, one serial network for HACMP per node and a public interface per node.
The serial interface will not be used by Oracle, so just one network is left to be used.
B) The system is incorrectly configured
typically you see just one network in 'oradebug ipc' tracefiles
C) Interface/adapter is defined as standby
That is a setup where HACMP can do the failover
D) Cluster_interconnects is set with Oracle9i 9.2.0.6
Solution
For A)
1) You may ignore the problem if the cluster is just for testing or development
Make sure to delete unused tracefiles that contain this message to avoid filling a disk
2) If the cluster is used for production add another network, which is setup as 'private' network.
The private network will be used by Oracle as interconnect for RAC (or OPS).
For B)
Make sure the cluster is correctly configured at the HACMP level
For C)
Oracle can not use 'standby' adapters. If you want to avoid that error in the alert-file and the traces then change the characteristics of the adapter from 'standby' to 'service'
For D)
When cluster_interconnects has been set in the init.ora/spfile the error can be ignored if 9.2.0.6 has been installed. The message will not be printed in Oracle9i 9.2.0.7
References
Bug 1941206 - Alert Log Flooded By Messages Since Upgrading To 8.1.7.2
Bug 3000985 - Lot Of Trace Files Generated With Disabling Tnff Message
Keywords
PRIVATE~NETWORKS; FAILOVER; CLUSTER_INTERCONNECTS;
---查看RAC 的cache fusion 的IP,既private IP
SQL> oradebug setmypid
Statement processed.
SQL> oradebug ipc
Information written to trace file.
SQL> oradebug tracefile_name
/app/orarac/admin/sid/udump/sid_ora_23667.trc
view /app/orarac/admin/sid/udump/sid_ora_23667.trc
SSKGXPT 0x188db4 flags SSKGXPT_READPENDING info for network 0
socket no 10 IP 192.168.1.2 UDP 56594
sflags SSKGXPT_WRITESSKGXPT_UP
info for network 1
socket no 0 IP 0.0.0.0 UDP 0
sflags SSKGXPT_DOWN
context timestamp 0x1c45
no ports
sconno accono ertt state seq# sent async sync rtrans acks
0x7973cfab 0x61d827be 32 3 32768 5 5 0 0 5
ach accono sconno admno state seq# rcv rtrans acks
192.168.1.2 就是cache fushion (cluster interconnection)的IP
oracle 10g 可以通过 cat /etc/hosts来查看
9iRAC: Network Interface Selection PolicyIBM AIX (HACMP)
This section pertains to Oracle running either on a SP Cluster, or on a HACMP Cluster (a.k.a. AIX Cluster, a.k.a. RS6000 Cluster), using the HACMP Cluster Manager stack. On both architectures, only the UDP IPC protocol is available to use for RAC.
Networks can be configured into a “Private” or into a “Public” category. Oracle chooses the network interface as followed
1. "Private" networks are preferred over "Public" networks.
2. If more than one "Private" [or "Public"] network are chosen, then they are ordered by performance quality: [High Performance Switch > FDDI > Ethernet]
Oracle will choose only networks configured as "service", and can choose up to three networks. If more than one network is selected, Transparent Network Failover Failback (TNFF) is enabled. In TNFF mode, we recommend two private and one (or more) public networks for cluster networks, and also have a "serial" link connecting the nodes. The serial interface helps in a better handling of "split-brain" scenarios, although this interface will not be used by Oracle, and just be used by HACMP.
At any point in time, IPC traffic, will be sent over ONLY one of the "subnets", even though multiple networks can be chosen, i.e., no load balancing is done. Multiple Networks are purely for High Availability purposes.
If cluster_interconnects is set, TNFF is disabled. Even with setting the cluster interconnects parameter, it is important, that "private" networks are used.
Example:
If a cluster has three “private” Ethernet networks and one “public” High Performance Switch (HPS) network, Oracle will choose the three Ethernet networks for its IPC. In the case, where a cluster has one “private” FDDI network, two “public” Ethernet networks, and one “public” HPS network, we will choose in the following order:
1. “Private” FDDI network (First preferred network for IPC)
2. “Public” HPS network (Second preferred network for IPC)
3. “Public” Ethernet (Third preferred network for IPC)
Of the two Ethernet networks, Oracle will choose the one that is returned first, when the following command is run.
/usr/sbin/cluster/utilities/cllsif -S
Sample output from cllsif –S:
[aix1: SU: / 151] /usr/sbin/cluster/utilities/cllsif -i aix1 -S
aix1 service ETHER1 ether public aix1 128.1.54.349 en1
aix1_pv1 service PV1 ether private aix1 10.0.0.14 en0
aix1_pv2 service PV2 ether private aix1 10.0.1.14 en3
[aix1: SU: / 152] /usr/sbin/cluster/utilities/cllsif -i aix2 -S
aix2 service ETHER1 ether public aix2 128.1.54.340 en1
aix2_pv1 service PV1 ether private aix2 10.0.0.13 en0
aix2_pv2 service PV2 ether private aix2 10.0.1.13 en3
The network ordering will be
1. en0
2. en3
3. en1
The /usr/sbin/cluster/clstat command indicates the state of the various network adapters in the cluster. If the STATE reports "UP", Oracle will treat that adapter as ALIVE, else if the status is reported as DOWN, then Oracle will mark the adapter DOWN (unusable for IPC).
Assuming "en0" is ALIVE on "aix1" and "aix2", then all Oracle processes will send and receive IPC traffic over the "10.0.0" subnet.
If "en0" of "aix1" or "aix2" or both go DOWN, then all processes will transmit IPC over the "10.0.1" subnet (i.e., "en3").
If "en0" AND "en3" of "aix1" or "aix2" or both become unavailable, then all processes will switch traffic over to the "128.1.54." subnet (i.e., "en1").
Once "en0" comes back ALIVE on both the nodes, all processes, will switch back their IPC traffic over the "10.0.0" subnet.
来自 “ ITPUB博客 ” ,链接:http://blog.itpub.net/10834762/viewspace-442154/,如需转载,请注明出处,否则将追究法律责任。
转载于:http://blog.itpub.net/10834762/viewspace-442154/