转载

Oracle ASM Rebalance执行过程

磁盘组的rebalance什么时候能完成？这没有一个具体的数值，但ASM本身已经给你提供了一个估算值（GV$ASM_OPERATION.EST_MINUTES），想知道rebalance完成的精确的时间，虽然不能给出一个精确的时间，但是可以查看一些rebalance的操作细节，让你知道当前rebalance是否正在进行中，进行到哪个阶段，以及这个阶段是否需要引起你的关注。

理解rebalance
rebalance操作本身包含了3个阶段-planning, extents relocation 和 compacting，就rebalance需要的总时间而言，planning阶段需要的时间是非常少的，你通常都不用去关注这一个阶段，第二个阶段extent relocation一般会占取rebalance阶段的大部分时间，也是我们最为需要关注的阶段，最后我们也会讲述第三阶段compacting阶段在做些什么。

首先需要明白为什么会需要做rebalance，如果你为了增加磁盘组的可用空间，增加了一块新磁盘或者为了调整磁盘的空间，例如resizing或者删除磁盘，你可能也不会太去关注rebalance啥时候完成。但是，如果磁盘组中的一块磁盘损坏了，这个时候你就有足够的理由关注rebalance的进度了，假如，你的磁盘组是normal冗余的，这个时候万一你损坏磁盘的partner磁盘也损坏，那么你的整个磁盘组会被dismount，所有跑在这个磁盘组上的数据库都会crash，你可能还会丢失数据。在这种情况下，你非常需要知道rebalance什么时候完成，实际上，你需要知道第二个阶段extent relocation什么时候完成，一旦它完成了，整个磁盘组的冗余就已经完成了（第三个阶段对于冗余度来说并不重要，后面会介绍）。

Extents relocation

为了进一步观察extents relocation阶段，我删除了具有默认并行度的磁盘组上的一块磁盘：

SQL> show parameter power

NAME                                 TYPE                   VALUE
------------------------------------ ---------------------- ------------------------------
asm_power_limit                      integer                1

14:47:35 SQL> select group_number,disk_number,name,state,path,header_status from v$asm_disk where group_number=5;

GROUP_NUMBER DISK_NUMBER NAME                 STATE                PATH                 HEADER_STATUS
------------ ----------- -------------------- -------------------- -------------------- --------------------
           5           0 TESTDG_0000          NORMAL               /dev/raw/raw7        MEMBER
           5           2 TESTDG_0002          NORMAL               /dev/raw/raw13       MEMBER
           5           1 TESTDG_0001          NORMAL               /dev/raw/raw12       MEMBER
           5           3 TESTDG_0003          NORMAL               /dev/raw/raw14       MEMBER

14:48:38 SQL> alter diskgroup testdg drop disk TESTDG_0000;

Diskgroup altered.

下面视图GV$ASMOPERATION的ESTMINUTES字段给出了估算值的时间，单位为分钟，这里给出的估算时间为9分钟。

14:49:04 SQL> select inst_id, operation, state, power, sofar, est_work, est_rate, est_minutes from gv$asm_operation where group_number=5;

   INST_ID OPERATION            STATE                     POWER      SOFAR   EST_WORK   EST_RATE EST_MINUTES
---------- -------------------- -------------------- ---------- ---------- ---------- ---------- -----------
         1 REBAL                RUN                           1          4       4748        475           9

大约过了1分钟后，EST_MINUTES的值变为了0分钟：

14:50:22 SQL> select inst_id, operation, state, power, sofar, est_work, est_rate, est_minutes from gv$asm_operation where group_number=5;

   INST_ID OPERATION            STATE                     POWER      SOFAR   EST_WORK   EST_RATE EST_MINUTES
---------- -------------------- -------------------- ---------- ---------- ---------- ---------- -----------
         1 REBAL                RUN                           1       3030       4748       2429           0

有些时候EST_MINUTES的值可能并不能给你太多的证据，我们还可以看到SOFAR（截止目前移动的UA数）的值一直在增加，恩，不错，这是一个很好的一个观察指标。ASM的alert日志中也显示了删除磁盘的操作，以及OS ARB0进程的ID，ASM用它用来做所有的rebalance工作。更重要的，整个过程之中，没有任何的错误输出：

SQL> alter diskgroup testdg drop disk TESTDG_0000 
NOTE: GroupBlock outside rolling migration privileged region
NOTE: requesting all-instance membership refresh for group=5
Tue Jan 10 14:49:01 2017
GMON updating for reconfiguration, group 5 at 222 for pid 42, osid 6197
NOTE: group 5 PST updated.
Tue Jan 10 14:49:01 2017
NOTE: membership refresh pending for group 5/0x97f863e8 (TESTDG)
GMON querying group 5 at 223 for pid 18, osid 5012
SUCCESS: refreshed membership for 5/0x97f863e8 (TESTDG)
NOTE: starting rebalance of group 5/0x97f863e8 (TESTDG) at power 1
Starting background process ARB0
SUCCESS: alter diskgroup testdg drop disk TESTDG_0000
Tue Jan 10 14:49:04 2017
ARB0 started with pid=39, OS id=25416 
NOTE: assigning ARB0 to group 5/0x97f863e8 (TESTDG) with 1 parallel I/O
cellip.ora not found.
NOTE: F1X0 copy 1 relocating from 0:2 to 2:2 for diskgroup 5 (TESTDG)
NOTE: F1X0 copy 3 relocating from 2:2 to 3:2599 for diskgroup 5 (TESTDG)
Tue Jan 10 14:49:13 2017
NOTE: Attempting voting file refresh on diskgroup TESTDG
NOTE: Refresh completed on diskgroup TESTDG. No voting file found.
Tue Jan 10 14:51:05 2017
NOTE: stopping process ARB0
SUCCESS: rebalance completed for group 5/0x97f863e8 (TESTDG)
Tue Jan 10 14:51:07 2017
NOTE: GroupBlock outside rolling migration privileged region
NOTE: requesting all-instance membership refresh for group=5
Tue Jan 10 14:51:10 2017
GMON updating for reconfiguration, group 5 at 224 for pid 39, osid 25633
NOTE: group 5 PST updated.
SUCCESS: grp 5 disk TESTDG_0000 emptied
NOTE: erasing header on grp 5 disk TESTDG_0000
NOTE: process _x000_+asm1 (25633) initiating offline of disk 0.3915944675 (TESTDG_0000) with mask 0x7e in group 5
NOTE: initiating PST update: grp = 5, dsk = 0/0xe96892e3, mask = 0x6a, op = clear
GMON updating disk modes for group 5 at 225 for pid 39, osid 25633
NOTE: group TESTDG: updated PST location: disk 0001 (PST copy 0)
NOTE: group TESTDG: updated PST location: disk 0002 (PST copy 1)
NOTE: group TESTDG: updated PST location: disk 0003 (PST copy 2)
NOTE: PST update grp = 5 completed successfully 
NOTE: initiating PST update: grp = 5, dsk = 0/0xe96892e3, mask = 0x7e, op = clear
GMON updating disk modes for group 5 at 226 for pid 39, osid 25633
NOTE: cache closing disk 0 of grp 5: TESTDG_0000
NOTE: PST update grp = 5 completed successfully 
GMON updating for reconfiguration, group 5 at 227 for pid 39, osid 25633
NOTE: cache closing disk 0 of grp 5: (not open) TESTDG_0000
NOTE: group 5 PST updated.
NOTE: membership refresh pending for group 5/0x97f863e8 (TESTDG)
GMON querying group 5 at 228 for pid 18, osid 5012
GMON querying group 5 at 229 for pid 18, osid 5012
NOTE: Disk TESTDG_0000 in mode 0x0 marked for de-assignment
SUCCESS: refreshed membership for 5/0x97f863e8 (TESTDG)
Tue Jan 10 14:51:16 2017
NOTE: Attempting voting file refresh on diskgroup TESTDG
NOTE: Refresh completed on diskgroup TESTDG. No voting file found.

因此ASM预估了9分钟的时间来完成rebalance，但实际上只使用了2分钟的时候,因此首先能知道rebalance正在做什么非常重要，然后才能知道rebalance什么时候能完成。注意，估算的时间是动态变化的，可能会增加或减少，这个依赖你的系统负载变化，以及你的rebalance的power值的设置，对于一个非常大容量的磁盘组来说，可能rebalance会花费你数小时甚至是数天的时间。

ARB0进程的跟踪文件也显示了，当前正在对哪一个ASM文件的extent的在进行重分配，也是通过这个跟踪文件，我们可以知道ARB0确实是在干着自己的本职工作，没有偷懒。

[grid@jyrac1 trace]$ tail -f  +ASM1_arb0_25416.trc
*** 2017-01-10 14:49:20.160
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:49:24.081
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:49:28.290
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:49:32.108
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:49:35.419
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:49:38.921
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:49:43.613
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:49:47.523
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:49:51.073
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:49:54.545
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:49:58.538
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:02.944
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:06.428
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:10.035
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:13.507
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:17.526
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:21.692
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:25.649
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:29.360
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:33.233
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:37.287
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:40.843
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:44.356
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:48.158
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:51.854
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:55.568
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:50:59.439
ARB0 relocating file +TESTDG.256.932913341 (120 entries)

*** 2017-01-10 14:51:02.877
ARB0 relocating file +TESTDG.256.932913341 (50 entries)

注意，跟踪目录下的arb0的跟踪文件可能会有很多，因此我们需要知道arb0的OS是进程号，是哪一个arb0在实际做rebalance的工作，这个信息在ASM实例执行rebalance操作的时候，alert文件中会有显示。我们还可以通过操作系统命令pstack来跟踪ARB0进程，查看具体它在做什么，如下，它向我们显示了，ASM正在重分配extent（在堆栈中的关键函数 kfgbRebalExecute - kfdaExecute - kffRelocate):

[root@jyrac1 ~]# pstack 25416
#0  0x0000003aa88005f4 in ?? () from /usr/lib64/libaio.so.1
#1  0x0000000002bb9b11 in skgfrliopo ()
#2  0x0000000002bb9909 in skgfospo ()
#3  0x00000000086c595f in skgfrwat ()
#4  0x00000000085a4f79 in ksfdwtio ()
#5  0x000000000220b2a3 in ksfdwat_internal ()
#6  0x0000000003ee7f33 in kfk_reap_ufs_async_io ()
#7  0x0000000003ee7e7b in kfk_reap_ios_from_subsys ()
#8  0x0000000000aea0ac in kfk_reap_ios ()
#9  0x0000000003ee749e in kfk_io1 ()
#10 0x0000000003ee7044 in kfkRequest ()
#11 0x0000000003eed84a in kfk_transitIO ()
#12 0x0000000003e40e7a in kffRelocateWait ()
#13 0x0000000003e67d12 in kffRelocate ()
#14 0x0000000003ddd3fb in kfdaExecute ()
#15 0x0000000003ec075b in kfgbRebalExecute ()
#16 0x0000000003ead530 in kfgbDriver ()
#17 0x00000000021b37df in ksbabs ()
#18 0x0000000003ec4768 in kfgbRun ()
#19 0x00000000021b8553 in ksbrdp ()
#20 0x00000000023deff7 in opirip ()
#21 0x00000000016898bd in opidrv ()
#22 0x0000000001c6357f in sou2o ()
#23 0x00000000008523ca in opimai_real ()
#24 0x0000000001c6989d in ssthrdmain ()
#25 0x00000000008522c1 in main ()

Compacting
在下面的例子里，我们来看下rebalance的compacting阶段，我把上面删除的磁盘加回来，同时设置rebalance的power为2：

17:26:48 SQL> alter diskgroup testdg add disk '/dev/raw/raw7' rebalance power 2;

Diskgroup altered.

ASM给出的rebalance的估算时间为6分钟：

16:07:13 SQL> select INST_ID, OPERATION, STATE, POWER, SOFAR, EST_WORK, EST_RATE, EST_MINUTES from GV$ASM_OPERATION where GROUP_NUMBER=1;

  INST_ID OPERA STAT      POWER      SOFAR   EST_WORK   EST_RATE EST_MINUTES
---------- ----- ---- ---------- ---------- ---------- ---------- -----------
        1 REBAL RUN          10        489      53851       7920           6

大约10秒后，EST_MINUTES的值变为0.

16:07:23 SQL> /

  INST_ID OPERA STAT      POWER      SOFAR   EST_WORK   EST_RATE EST_MINUTES
---------- ----- ---- ---------- ---------- ---------- ---------- -----------
        1 REBAL RUN          10      92407      97874       8716           0

这个时候我们在ASM的alert日志中观察到：

SQL> alter diskgroup testdg add disk '/dev/raw/raw7'  rebalance power 2
NOTE: GroupBlock outside rolling migration privileged region
NOTE: Assigning number (5,0) to disk (/dev/raw/raw7)
NOTE: requesting all-instance membership refresh for group=5
NOTE: initializing header on grp 5 disk TESTDG_0000
NOTE: requesting all-instance disk validation for group=5
Tue Jan 10 16:07:12 2017
NOTE: skipping rediscovery for group 5/0x97f863e8 (TESTDG) on local instance.
NOTE: requesting all-instance disk validation for group=5
NOTE: skipping rediscovery for group 5/0x97f863e8 (TESTDG) on local instance.
Tue Jan 10 16:07:12 2017
GMON updating for reconfiguration, group 5 at 230 for pid 42, osid 6197
NOTE: group 5 PST updated.
NOTE: initiating PST update: grp = 5
GMON updating group 5 at 231 for pid 42, osid 6197
NOTE: PST update grp = 5 completed successfully 
NOTE: membership refresh pending for group 5/0x97f863e8 (TESTDG)
GMON querying group 5 at 232 for pid 18, osid 5012
NOTE: cache opening disk 0 of grp 5: TESTDG_0000 path:/dev/raw/raw7
GMON querying group 5 at 233 for pid 18, osid 5012
SUCCESS: refreshed membership for 5/0x97f863e8 (TESTDG)
NOTE: starting rebalance of group 5/0x97f863e8 (TESTDG) at power 1
SUCCESS: alter diskgroup testdg add disk '/dev/raw/raw7'
Starting background process ARB0
Tue Jan 10 16:07:14 2017
ARB0 started with pid=27, OS id=982 
NOTE: assigning ARB0 to group 5/0x97f863e8 (TESTDG) with 1 parallel I/O
cellip.ora not found.
Tue Jan 10 16:07:23 2017
NOTE: Attempting voting file refresh on diskgroup TESTDG

上面的输出意味着ASM已经完成了rebalance的第二个阶段，开始了第三个阶段compacting，如果我说的没错，通过pstack工具可以看到kfdCompact()函数，下面的输出显示，确实如此:

# pstack 982
#0  0x0000003957ccb6ef in poll () from /lib64/libc.so.6
...
#9  0x0000000003d711e0 in kfk_reap_oss_async_io ()
#10 0x0000000003d70c17 in kfk_reap_ios_from_subsys ()
#11 0x0000000000aea50e in kfk_reap_ios ()
#12 0x0000000003d702ae in kfk_io1 ()
#13 0x0000000003d6fe54 in kfkRequest ()
#14 0x0000000003d76540 in kfk_transitIO ()
#15 0x0000000003cd482b in kffRelocateWait ()
#16 0x0000000003cfa190 in kffRelocate ()
#17 0x0000000003c7ba16 in kfdaExecute ()
#18 0x0000000003c4b737 in kfdCompact ()
#19 0x0000000003c4c6d0 in kfdExecute ()
#20 0x0000000003d4bf0e in kfgbRebalExecute ()
#21 0x0000000003d39627 in kfgbDriver ()
#22 0x00000000020e8d23 in ksbabs ()
#23 0x0000000003d4faae in kfgbRun ()
#24 0x00000000020ed95d in ksbrdp ()
#25 0x0000000002322343 in opirip ()
#26 0x0000000001618571 in opidrv ()
#27 0x0000000001c13be7 in sou2o ()
#28 0x000000000083ceba in opimai_real ()
#29 0x0000000001c19b58 in ssthrdmain ()
#30 0x000000000083cda1 in main ()

通过tail命令查看ARB0的跟踪文件，发现relocating正在进行，而且一次只对一个条目进行relocating。（这是正进行到compacting阶段的另一个重要线索）：

$ tail -f +ASM1_arb0_25416.trc
ARB0 relocating file +DATA1.321.788357323 (1 entries)
ARB0 relocating file +DATA1.321.788357323 (1 entries)
ARB0 relocating file +DATA1.321.788357323 (1 entries)
...

compacting过程中，V$ASM_OPERATION视图的EST_MINUTES字段会显示为0（也是一个重要线索）：

16:08:56 SQL> /

  INST_ID OPERA STAT      POWER      SOFAR   EST_WORK   EST_RATE EST_MINUTES
---------- ----- ---- ---------- ---------- ---------- ---------- -----------
        2 REBAL RUN          10      98271      98305       7919           0

固态表X$KFGMG的REBALST_KFGMG字段会显示为2，代表正在compacting。

16:09:12 SQL> select NUMBER_KFGMG, OP_KFGMG, ACTUAL_KFGMG, REBALST_KFGMG from X$KFGMG;

NUMBER_KFGMG   OP_KFGMG ACTUAL_KFGMG REBALST_KFGMG
------------ ---------- ------------ -------------
          1          1           10             2

一旦compacting阶段完成，ASM的alert 日志中会显示stopping process ARB0 和rebalance completed:

Tue Jan 10 16:10:19 2017
NOTE: stopping process ARB0
SUCCESS: rebalance completed for group 5/0x97f863e8 (TESTDG)

一旦extents relocation完成，所有的数据就已经满足了冗余度的要求，不再会担心已经失败磁盘的partern磁盘再次失败而出现严重故障。

Changing the power
Rebalance的power可以在磁盘组rebalance过程中动态的更改，如果你认为磁盘组的默认级别太低了，可以去很容易的增加它。但是增加到多少呢？这个需要你根据你系统的IO负载，IO吞吐量来定。一般情况下，你可以先尝试增加到一个保守的值，例如5，过上十分钟看是否有所提升，以及是否影响到了其他业务对IO的使用，如果你的IO性能非常强，那么可以继续增加power的值，但是就我的经验来看，很少能看到power 的设置超过30后还能有较大提升的。测试的关键点在于，你需要在你生产系统的正常负载下去测试，不同的业务压力，不同的存储系统，都可能会让rebalance时间产生较大的差异。

正文到此结束