Deep understanding of Oracle RAC 12cJump to: Navigation, search
Directory
- 1 Overview
- 2 Cluster parts management and fault diagnosis
- 3 Operational Practices
- 4 new Features
- 5 Storage and ASM
- 6 problems in application design
- 7 managing and tuning a complex RAC environment
- 8 Backup and Recovery
- 9 Network Practice
- Ten Optimizing the RAC database
- One Lock and deadlock
- A Parallel Queries
- - cluster components and database upgrades
- - RAC One Node
|
overview
- P5 RAC uses cache Fusion technology to manage data caching for all nodes and to maintain consistency ... Why does
- not read from the high-speed IO network, but from the other node's cache?
- p15 GCS GES GRD
- There are some background processes: LMS lmon LMD LCK DIAG
- cluster architecture
- full Share
- Primary-from: SQL server cluster, Oracle Fail Safe, RAC one Node
- primary-primary: RAC
- no sharing
- 12CR1 Flex cluster
- p28 loosely coupled leaf-node architecture with a few hub nodes tightly coupled
- multiple physical network cards are combined as one logical network card: Linux bonding/ms NIC Teaming
- 11.2.0.2 Haip
- RAC Component: ASM + clusterware
- crsctl
- OCR (cluster registry)
- p19 ONS (Notification Service) FAN (Quick app notification) FCF (fast connection failover
- brain fissure: stonith algorithm, healthy node let the problem node restart?
- Transparent application failover (TAF)
- virtual IP
- apply continuity (AC)
- rolling upgrade?
- Cost of ownership: The standard version of the Oracle database contains RAC, but the total number of CPU slots <=4
- is cheaper compared to large SMP servers because Linux and low-cost hardware are used?
- RAC can withstand up to N-1 failures, with n-1/n headroom for each hardware plan
Cluster parts management and fault diagnosis
- Ohas:gpnp GNS GIPC MDNS osysmond ologgerd
- CRS (upper Stack): CRS CSS CSS proxy CSS monitor CTSS EVM ONS oraagent orarootagent
- A bunch of management command-line programs: slightly
Operational Practices
- Lreg: Performing a listener registration
- Failover type: NONE SESSION SELECT (remember cursor position) TRANSACTION
- 11.2 SCAN (front-end redirection?) )
- 12c GDS
- Temporary table space
- SS queue, DFS lock handle with CI queue, 11g+ release extents cache in bulk (100 each)
- Modification of large amounts of data
- Reduce global cache transfers and allocate all PX services within the same node as possible
- Performance Metrics collection: Oswatcher, AWR, Statspack
- Parameter file Management
- Password file Management
- SVRCTL: Managing Databases and instances
new Features
- Flex Cluster and Flex ASM
- Slightly
- Multi-tenancy: There are multiple PDB in a CDB
- Cdb$root
- Pdb$seed
- Pdb
- Other: IPv6 GDS ACFS/ADVM online resource Property modification
- Policy-based database management
- ASM Disk Group
- ...
Storage and ASM
- Storage protocol: SCSI FC IP NAS
- Multipath: A set of SCSI devices pointing to the same storage device (LUN)
- Asmlib?
- P169 strongly recommends using an external RAI configuration like RAID 1+0 to ensure ASM redundancy
- Storing OCR and voting disk on ASM
- ACFS
problems in application design
- ' Right Index growth competition ' (because of the self-increment key?) ):
- Hash index partition
- Hash table partitioning (logical reads increase massively if you do not have the partition trimmed data)
- Reverse index (scope query not supported)
- Use multiple sequences of different ranges (applied control)
- A lot of Truncate/drop
- Object-level checkpoint events can be raised, and global waits can cause serious problems
- Global temp table (GIT)?
- Sequence Cache
- Order,cache 20
- Free Block list and automatic segment table space Management (ASSM)
- Too many commits
- LGWR delay causes LMS delay, further causes the other node to cache the event more long waits, ' GC buffer busy wait '
- A transaction that has not been committed for a long time (resulting in a lot of rollbacks?) )
- Local Access *
- Small table Update
- Index Design
- Bitmap indexes are good for read-only tables; ' Compressed index '
- Inefficient execution plan
- Too many parallel scans
- Full table Scan: direct path?
- Relevance between apps *
- Implementation of application changes
- Library Cache Lock/pin
managing and tuning a complex RAC environment
- Share vs non-shared home directory
- Server pool
- 11gR2 Instance Lock
- sql> ALTER SYSTEM SET cpu_count=2 scope=both sid= ' instance_name ';
- Small-scale vs large-scale
- Split Brain: Node expulsion, fencing, STONITH
- Building redundancy at the network layer
- Allocate enough cpu/memory for application workloads and set quotas
- Implementation of DBRM and Iorm
- Avoid configuring too many lms_processes
- AUTO DOP
- Node Eviction
- Extended distance Clusters-summary, generalization, and best practices
- RAC installation configuration in different OS considerations/Tips
- Linux:hugepages, ASMM, GI
- Windows:
- Have public NICs take precedence over private NICs?
- Using asynchronous IO
Backup and Recovery
- p240 using rapid Recovery Zone (FRA)
- Multi-Channel
- Rman> Configure Channel 1 device type disk connect ' Sys/[email protected]_1 ';
- P250 block Change Tracking (BCT)
Network Practice
- p265 RDS protocol requires specific hardware (InfiniBand) and kernel drivers
- Similar to UDP, there is no ACK mechanism in RDS, and further, the RDS design is 0 replicated; completely skipped IP (? )
- The presence of the p270 VIP avoids the need to wait for TCP timeouts
- Cluster inline/private network
- Each NIC needs to be configured with a separate subnet segment (11.2+) =〉 Otherwise, if you unplug the network cable on the first NIC, ARP will not update the ARP cache correctly (why?). )
- Jumbo Frame: mtu=9000
- LB and fail-over
- 12c HAVIP
- Kernel parameters
- Net.ipv4.ipfrag_low/high_thresh, Net.ipv4.ipfrag_time
- GC Lost block problem
- Gns
Optimizing the RAC database
- grd*
- RAC Wait Event *
- GC Current Block 2-way/3-way
- GC CR Block 2-way/3-way
- GC CR Grant 2-WAY/GC current Grant 2-way
- GC CR Block BUSY/GC current block Busy
- GC CR Block CONGESTED/GC current block congested
- Placeholder Wait Events
- Send-side analysis
- GCS Log Flush Sync
- GC Buffer Busy Acquire/release
- Unique index
- Table Block
- Overview of the DRM process *
- AWR reports and ADDM
Lock and deadlock
- Global resource type:
- BL (Buffer Lock), GCS Management
- Non-BL:TX, TM, sq,ges management
- Lock mode *
- Troubleshooting Methods for lock contention *
- Queue contention (temporary)
- TX
- Tm
- HW
- DFS Lock Handle
- SV Resources
- CI Resources
- p373 Queue type: Ss/iv/ci
- Library Cache Locks/pins
- Queue statistics Information
- V$wait_chains
- Hanganalyze
- Dead lock
- LMD Trace File
Parallel Queries
- p416 partition-wise Join
- The traditional PX uses the Producer/consumer mode,...
cluster components and database upgradesRAC One Node
Deep understanding of Oracle RAC 12c notes