Lesson 1 Module 1
oracle clusterware 组成。 与asm, acfs, rac等的关系。 feature: load balance & fail over.
包括哪些服务:cluster management(service & ), node monitorning(evict or fencing), event services, time sync, network admin(VIP,GNS), high availability(service/monitor or restart).
Clusterware network:
public:tcp/ip, private xx or udp(interconnect), - limit: same interface name, -->options, IP can come from GNS.
interconect link aggregation. a/s a/a. Interconnect NIC guidelines.
Module 2 Clusterware architecture
startup --> by init.ohasd(auto restart). high avalibility service daemon.
ohasd start -->orarootagent (start crsd which manage cluster resource), (root owned crs resource like SCAN VIPS).
-->cssdagent start cssd(ocssd)
-->oraagent , start mdnsd, evmd, ASM, ctssd, and gpnpd. also, gsd, ONS, and listeners.
Some root , some user-mode,
process architecture:
CRS(cluster ready service): crsd. Operation/monitor and manage resource: 1 cluster resource: crsctl command.(when rac ready, resource instance,listener etc) 2 local resource: no fail over.
CRSd generate events when status of a resource changes.
Filover policy: try to restart, and in the end, fail over. But depends on configuration.
Cluster sync service (CSS): ocssd, ... Node member management. Cssdagent failed result is cluster restart node.
Event Management (EVM),
and Oracle Notification Service (ONS).
Others (CTSS etc).
Grid Plug and Play:(add/remove node, just connect to one of server). & Grid Naming Service, dont know more.
Use ASM and Automatic Undo management is best practice.
Single Client Access Name (difference with VIP?)
Stable IP and independant of nodes that make up the cluster.
resolved on any node on a cluster. Unlike VIP, when nodes add/remove, VIP will be updated but SCAN no this requirements. Listener or each node affects all SCANs. -->least loaded instance to get the recent SCAN request. (?remote listener)
Scan and local listener -- good.
How GPnP works (?)
client database connections: 这幅图画的挺清楚。 scan-name.cluster-name.company.com/ServiceName
Module 3 ASM architecture
DB/ASM(ADVM)/ACFS 关系。
features.
Design: 1, nonclustered ASM and db.
Design 2: Clustered ASM.
3: Clustered ASM for Mixed Databases.
ASM permission/authentication.
ASM component: software; instance(SGA 特殊处, ); process(clustered ASM need more process) rebalance process...
Node Listener.
Configuration files
/etc/oratab; /etc/oraInst.loc;
ASM disk group: self-contained metadata and logging info.
Failure group:(?)
ASM disks: AU,
ASM files: naming , +.. Each ASM file can not be shared.
ASM clients, v$asm_client.
ASM utility: asmcmd, asmca, srvctl(server controller utility),. start stop asm instance.
Lesson 2 Infrastructure Installation
Preinstall:
Shared storage planning for grid infrastructure. (voting/OCR, oracle software) ASM ACFS(not support OCR&Voting!) OCFS2 GFS NFS Shared disk slices(block or raw devices)-not support for new db!!
OCR: what, can be mirrored.
Volting:
如果使用asm 作为voting disk, 特殊之处也存在。 不会作为普通的asm file被使用--实际位置会被cluster知道,这样即时in case of asm fail, voting disk still can be accessed by CSS service. 管理 crsctl.
split-brain function: heart beat 通常检测,如果这条道出现问题,通知其中一个节点退出 如何通知-用voting disk. / quorum disk
clusterware main log file.
Prepare ASM: lib-->configure-->init-->scan disks
Module 2: Grid Infrastructure Preinstall.
rpm -ivh dependancy.
PAM limits. (etc/security/limits.conf) . Process and open files limitation. ulimit for shell , process & files.... 一个是shell的限制 一个是user的限制。
crsctl stat res -t ,,,
Module 4: cfg ASM disks Groups and ACFS --> ACFS和直接使用ASM的区别? 好多中办法去创建。
ASM group --> ASM volume -->ACFS. mount, registry etc. ACFS 的一些限制。 hwo to mount, unix is ok, windows still needs mount :)
Module 3: Admin clusterware
crsctl , start stop crs, enable/disable.
srvctl resource(dont include regist)
Verify: crsctl check crs. /cluster
Ocr and Voting disk location:
query css votedisk. more /etc/oracle/ocr.loc(or ocrcheck)
Backup and Recoverying the voting disk. -->autobackup when change. crs add(delete) css votedisk path.
migrate to non-asm replace votedisk +|.
Locate OCR backup: (auto backup by one node)--> ocrconfig -showbackup auto
ocrconfig -add /-replace.
Recover OCR.
Network setting: oifcfg iflist -p -n; oifcfg getif; srvctl config nodeapps -a
Change IP.
Module4 Manage Clusterware. Adding and deleting oracle clusterware homes.
Add nodes.
prework.
verify: cluvfy stage -post hwos -n xxx ; compare.
add, precheck, add .
post check, cluvfs stage -post nodeadd -n host03 -verbose
Delete Nodes.
Remove node is not a easy job -->for there is node info on each inventory and OCR(&Voting).
Patching Clusterware.
Out of place upgrades. But only one is active , still reduce the down time. And still a rolling upgrade (node by node).
Patch Set: Out of place
Bundles:in-place.
One-off patch:in-place.
5 Making Applications Highly Available with Oracle Clustware.
RAC is oracle's application. Still can support other applications.
Component: Resource, Action Program, Priviledge, resource dependancy, VIP, OCR.
Skip .
6 Trouble shooting Oracle Clusterware
Time sync: if untimely, maybe evict
Log
Diagcollection.pl.
Cluster Verifier. component: cluvfy comp -list.
7 Admin ASM instance
Instance type=asm,. cluster_database=true; Spfile 应该在ASM或者cluster fs.
Srvctl start/stop/status asm (-n xxx.) , or from sqlplus.
node listener, from lsnrctl or srvctl start listener xxx.
8. Admin ASM disk groups.
?when rebalance. Add or Drop disk auto rebalance.
9. ASM
Migrate local disk to ASM. RMAN.
10.Administering ACFS
ACFS, block file system based on ASM. Suit for cluster or single. db or non-db. But for asm need first, so , can not boot, root system, oracle clusterware, ASM, ACFS etc.
QA:
1)redo & undo:
Again:
1, asm instance, no redo log buffer. Speical process for asm instance;
ADVM still has special volume process?
2,In-memory copy of OCR.
OCR的特殊读写机制。
voting disk - 如果asm, 多少个voting创建 会根据redundancy 来分配。 1,3,5. 3 for normal redundancy disk group. 存在形式不是作为asm内的普通文件。 即时asm出了问题,也不会... css服务仍然可以访问voting disk. 如果其中的一个voting disk坏了... 没问题,ASM会自动恢复voting disk.
Voting disk function:
node availability via communication through a dedicated private network , and with a vote disk as a second communication mechanism.
Relocate test:
crsctl relocate resource my_resource –n host01 –s host02
crsctl check crs 检查本地, check crs
voting disk 备份-每次config改变都自动备份到.
OCR也是自动备份的,只有一个节点执行备份。