Oracle 19c AFD不支持存儲的多路復用
Oracle版本：
SQL*Plus: Release 19.0.0.0.0 - Production on Mon May 31 09:11:51 2021
Version 19.11.0.0.0
經過實驗發現，當存儲盤存在多路復用的情況，如果不使用多路徑會導致AFD在afd_scan的時候忽略存儲盤。
[root@dev-devdb ~]# lsscsi --scsi_id -s [0:1:0:0] disk HP LOGICAL VOLUME 1.64 /dev/sda 3600508b100103035424b323031370300 299GB [0:3:0:0] storage HP P712m 1.64 - - - [1:0:0:0] storage HP HSV300 1100 - - - [1:0:0:1] disk HP HSV300 1100 /dev/sdb 36001438009b025a00000b000002b0000 5.36GB [1:0:0:2] disk HP HSV300 1100 /dev/sdc 36001438009b025a00000b00000300000 5.36GB [1:0:0:3] disk HP HSV300 1100 /dev/sdd 36001438009b025a00000b00000350000 5.36GB [1:0:0:4] disk HP HSV300 1100 /dev/sde 36001438009b025a00000b000003a0000 322GB [1:0:0:5] disk HP HSV300 1100 /dev/sdf 36001438009b025a00000b000003f0000 214GB [1:0:0:6] disk HP HSV300 1100 /dev/sdg 36001438009b025a00000b00000470000 53.6GB [1:0:1:0] storage HP HSV300 1100 - - - [1:0:1:1] disk HP HSV300 1100 /dev/sdh 36001438009b025a00000b000002b0000 5.36GB [1:0:1:2] disk HP HSV300 1100 /dev/sdi 36001438009b025a00000b00000300000 5.36GB [1:0:1:3] disk HP HSV300 1100 /dev/sdj 36001438009b025a00000b00000350000 5.36GB [1:0:1:4] disk HP HSV300 1100 /dev/sdk 36001438009b025a00000b000003a0000 322GB [1:0:1:5] disk HP HSV300 1100 /dev/sdl 36001438009b025a00000b000003f0000 214GB [1:0:1:6] disk HP HSV300 1100 /dev/sdm 36001438009b025a00000b00000470000 53.6GB [2:0:0:0] storage HP HSV300 1100 - - - [2:0:0:1] disk HP HSV300 1100 /dev/sdn 36001438009b025a00000b000002b0000 5.36GB [2:0:0:2] disk HP HSV300 1100 /dev/sdo 36001438009b025a00000b00000300000 5.36GB [2:0:0:3] disk HP HSV300 1100 /dev/sdp 36001438009b025a00000b00000350000 5.36GB [2:0:0:4] disk HP HSV300 1100 /dev/sdq 36001438009b025a00000b000003a0000 322GB [2:0:0:5] disk HP HSV300 1100 /dev/sdr 36001438009b025a00000b000003f0000 214GB [2:0:0:6] disk HP HSV300 1100 /dev/sds 36001438009b025a00000b00000470000 53.6GB [2:0:1:0] storage HP HSV300 1100 - - - [2:0:1:1] disk HP HSV300 1100 /dev/sdt 36001438009b025a00000b000002b0000 5.36GB [2:0:1:2] disk HP HSV300 1100 /dev/sdu 36001438009b025a00000b00000300000 5.36GB [2:0:1:3] disk HP HSV300 1100 /dev/sdv 36001438009b025a00000b00000350000 5.36GB [2:0:1:4] disk HP HSV300 1100 /dev/sdw 36001438009b025a00000b000003a0000 322GB [2:0:1:5] disk HP HSV300 1100 /dev/sdx 36001438009b025a00000b000003f0000 214GB [2:0:1:6] disk HP HSV300 1100 /dev/sdy 36001438009b025a00000b00000470000 53.6GB
在GI啟動之前，執行afdload start的時候，會根據/etc/oracleafd.conf文件中afd_diskstring去掃描盤。
此時，查看日志/u01/app/grid/diag/afdboot/user_root/host_3462916399_110/trace/alert.log有如下提示：
2021-05-31T08:56:04.752260+08:00 afdb_scandisk: Start afdb_getdiscstr: Start afd_discovery string: /dev/sd* afdb_getdiscstr: end Tokenized diskstring /dev/sd* No label in disk /dev/sda No label in disk /dev/sda1 No label in disk /dev/sda2 No label in disk /dev/sda3 Ignoring dup disk: /dev/sdi label: OCR2 Ignoring dup disk: /dev/sdc label: OCR2 Ignoring dup disk: /dev/sdh label: OCR1 Ignoring dup disk: /dev/sdb label: OCR1 Ignoring dup disk: /dev/sdk label: DATA1 Ignoring dup disk: /dev/sde label: DATA1 Ignoring dup disk: /dev/sdj label: OCR3 Ignoring dup disk: /dev/sdd label: OCR3 Ignoring dup disk: /dev/sdm label: MGMT1 Ignoring dup disk: /dev/sdg label: MGMT1 Ignoring dup disk: /dev/sdl label: ARCH1 Ignoring dup disk: /dev/sdf label: ARCH1 Ignoring dup disk: /dev/sdn label: OCR1 Ignoring dup disk: /dev/sds label: MGMT1 Ignoring dup disk: /dev/sdo label: OCR2 Ignoring dup disk: /dev/sdp label: OCR3 Ignoring dup disk: /dev/sdu label: OCR2 Ignoring dup disk: /dev/sdt label: OCR1 Ignoring dup disk: /dev/sdq label: DATA1 Ignoring dup disk: /dev/sdv label: OCR3 Ignoring dup disk: /dev/sdw label: DATA1 Ignoring dup disk: /dev/sdx label: ARCH1 Ignoring dup disk: /dev/sdy label: MGMT1 Ignoring dup disk: /dev/sdr label: ARCH1 There are no devices to discover. afdb_scandisk: Successful afdb_scandisk: end
此時，也無法使用asmcmd afd_scan來將磁盤掃描出來，也就是說/dev/oracleafd/disks/會是空的。
注意：asmcmd afd_scan如果不加參數，默認也是讀取/etc/oracleafd.conf文件。
詳情可以查看/u01/app/grid/diag/asmcmd/user_root/dev-devdb/alert/alert.log日志。
當使用多路徑綁定，消除多路復用多出來的額外的存儲磁盤后，同時/etc/oracleafd.conf文件修改為afd_diskstring='/dev/mapper/asm*'，
再次執行afdload start，此時查看日志/u01/app/grid/diag/afdboot/user_root/host_3462916399_110/trace/alert.log有如下提示：
2021-05-31T08:58:10.319379+08:00 afdb_scandisk: Start afdb_getdiscstr: Start afd_discovery string: /dev/mapper/asm* afdb_getdiscstr: end Tokenized diskstring /dev/mapper/asm* Scan count 6 Scanned disk /dev/mapper/asm-ocr1 Scanned disk /dev/mapper/asm-ocr2 Scanned disk /dev/mapper/asm-ocr3 Scanned disk /dev/mapper/asm-data1 Scanned disk /dev/mapper/asm-arch1 Scanned disk /dev/mapper/asm-mgmt1 afdb_scandisk: Successful afdb_scandisk: end
這個表明當前afd不支持存儲的多路復用。
若執意要使用afd_diskstring='/dev/sd*',那么afdload start后，使用asmcmd afd_scan '/dev/sdb'這樣子單獨一塊一塊手工掃描出來。
最后，將集群啟動后， /etc/oracleafd.conf文件的afd_diskstring會被還原為OLR記錄的AFD_DISKSTRING路徑。
使用asmcmd dsget需要的集群啟動狀態下才可以成功執行，原因就是因為該命令會讀取OLR信息，同時，
會更新/etc/oracleafd.conf文件的afd_diskstring為OLR記錄的AFD_DISKSTRING路徑。
如果由原來的/dev/sd*要變更為/dev/mapper/asm*的話，在集群啟動后需要使用asmcmd afd_dsset '/dev/mapper/asm*'來更新OLR的信息。
獲取OLR的dump文件命令：ocrdump -local
附上OLR的dump文件部分內容：
[SYSTEM.ASM.AFD_DISKSTRING] ORATEXT : /dev/mapper/asm* SECURITY : {USER_PERMISSION : PROCR_ALL_ACCESS, GROUP_PERMISSION : PROCR_READ, OTHER_PERMISSION : PROCR_READ, USER_NAME : grid, GROUP_NAME : oinstall}