近距离接触RAC DRM

drm 是Oracle rac中独有的动态资源管理操作, 我们听了很多关于DRM的理论, 但是你是否亲眼见证过DRM, 今天我们就来看一下:

 

SQL> select * from v$version;

BANNER
--------------------------------------------------------------------------------
Oracle Database 11g Enterprise Edition Release 11.2.0.3.0 - 64bit Production
PL/SQL Release 11.2.0.3.0 - Production
CORE 11.2.0.3.0 Production
TNS for Linux: Version 11.2.0.3.0 - Production
NLSRTL Version 11.2.0.3.0 - Production

SQL>
SQL> select count(*) from gv$instance;

COUNT(*)
----------
2

SQL> create table drm_mac(t1 int) tablespace users;

Table created.

SQL> insert into drm_mac values(1);

1 row created.

SQL> commit;

Commit complete.

SQL> alter system flush buffer_cache;

System altered.

SQL>select * from drm_mac;

T1
----------
1

SQL> select object_id,data_object_id from dba_objects where object_name='DRM_MAC';

OBJECT_ID DATA_OBJECT_ID
---------- --------------
64046 64046

--FIRST FIND THE OBJECT IDS
With OBJ_IDS As
(Select DATA_Object_Id OBJECT_ID From Dba_Objects
Where Object_Name = 'DRM_MAC'), --注意替换这里!!!!!!!!!!!!!!!!!
Addr As (Select /*+ materialize */
Le_Addr, class, state From X$Bh, OBJ_IDS
Where Object_Id = Obj),
--NOW GET THE RESOURCE NAME IN HEX
Hexnames As (Select
Rtrim(B.Kjblname, ' '||chr(0)) Hexname
From X$Le A, X$Kjbl B, Addr
Where A.Le_Kjbl=B.Kjbllockp
and class = 1 and state <> 3
And A.Le_Addr = Addr.Le_Addr)
-- NOW FIND THE NODE MASTERS
Select A.Master_Node Mast, Count(*)
From Gv$Dlm_Ress A, Hexnames H
Where
Rtrim(A.Resource_Name, ' '||chr(0)) = H.Hexname
Group by A.Master_Node;

MAST COUNT(*)
---------- ----------
0 5

==> 说明DRM_MAC MASTER 在0节点,共5个BLOCk

登录 2节点

对2节点的LMS进程做DRM TRACE
oradebug setospid 29295;
oradebug event trace[RAC_DRM] disk highest;

之后使用 lkdebug -m pkey data_object_id 来手动出发对DRM_MAC的REMASTER操作

注意即便是手动触发DRM也仅仅是将该OBJECT放入DRM ENQUEUE

*** 2012-11-27 09:27:20.728
Processing Oradebug command 'lkdebug -m pkey 64045'
*****************************************************************
GLOBAL ENQUEUE SERVICE DEBUG INFORMATION
----------------------------------------
Queued DRM request for pkey 64045.0, tobe master 1, flags x1
***************** End of lkdebug output *************************
Processing Oradebug command 'lkdebug -m pkey 64045'
*****************************************************************
GLOBAL ENQUEUE SERVICE DEBUG INFORMATION
----------------------------------------
kjdrrmq: pkey 64045.0 is already on the affinity request queue, old tobe == new tobe 1
***************** End of lkdebug output *************************
SQL> oradebug setmypid
Statement processed.
SQL> oradebug lkdebug -m pkey 64046;

--FIRST FIND THE OBJECT IDS
With OBJ_IDS As
(Select DATA_Object_Id OBJECT_ID From Dba_Objects
Where Object_Name = 'DRM_MAC'), --Customize
Addr As (Select /*+ materialize */
Le_Addr, class, state From X$Bh, OBJ_IDS
Where Object_Id = Obj),
--NOW GET THE RESOURCE NAME IN HEX
Hexnames As (Select
Rtrim(B.Kjblname, ' '||chr(0)) Hexname
From X$Le A, X$Kjbl B, Addr
Where A.Le_Kjbl=B.Kjbllockp
and class = 1 and state <> 3
And A.Le_Addr = Addr.Le_Addr)
-- NOW FIND THE NODE MASTERS
Select A.Master_Node Mast, Count(*)
From Gv$Dlm_Ress A, Hexnames H
Where
Rtrim(A.Resource_Name, ' '||chr(0)) = H.Hexname
Group by A.Master_Node;

MAST COUNT(*)
---------- ----------
1 5

通过 oradebug lkdebug -m pkey 将 MACLEAN_DRM REMASTER到2节点。
SQL> select * from v$gcspfmaster_info where DATA_OBJECT_ID=64045;
FILE_ID DATA_OBJECT_ID GC_MASTERIN CURRENT_MASTER PREVIOUS_MASTER REMASTER_CNT
---------- -------------- ----------- -------------- --------------- ------------
0 64046 Affinity 1 0 1

SQL> select * from gv$policy_history
where DATA_OBJECT_ID=64046
order by EVENT_DATE;

no rows selected
SQL> select * from x$object_policy_statistics;

no rows selected

*** 2013-11-15 04:08:35.191
2013-11-15 04:08:35.191065 : kjbldrmnextpkey: AFFINITY pkey 64046.0
pkey 64046.0 undo 0 stat 5 masters[1, 1->2] reminc 13 RM# 18
flg x7 type x0 afftime x8b0628ee, acquire time 0
nreplays by lms 0 = 0
benefit 0, total 0, remote 0, cr benefit 0, cr total 0, cr remote 0
kjblpkeydrmqscchk: Begin for pkey 64046/0 bkt[0->8191] read-mostly 0
kjblpkeydrmqscchk: End for pkey 64046.0
2013-11-15 04:08:35.191312 : DRM(20) quiesced basts [0-8191]
2013-11-15 04:08:35.197815 :
* lms acks DRMFRZ (8191)
* lms 0 finished parallel drm freeze in DRM(20) window 1, pcount 47
DRM(20) win(1) lms 0 finished drm freeze
2013-11-15 04:08:35.200436 :
* kjbrrcres: In this window: 278 GCS resources, 0 remastered

* In all windows: 278 GCS resources, 0 remastered
2013-11-15 04:08:35.201606 : kjbldrmnextpkey: AFFINITY pkey 64046.0
pkey 64046.0 undo 0 stat 5 masters[1, 1->2] reminc 13 RM# 18
flg x7 type x0 afftime x8b0628ee, acquire time 0
nreplays by lms 0 = 0
benefit 0, total 0, remote 0, cr benefit 0, cr total 0, cr remote 0
kjbldrmrpst: no more locks for pkey 64046.0, #lk 0, #rm 0, #replay 0, #dup 0 objscan 4
DRM(20) Completed replay (highb 8191)
DRM(20) win(1) lms 0 finished replaying gcs resources
kjbmprlst: DRMFRZ replay msg from 1 (clockp 0x92f96c18, 3)
0x1fe6a.5, 64046.0, x0, lockseq 0x13c5, reqid 0x3
state 0x100, 1 -> 0, role 0x0, wrt 0x0.0
GCS RESOURCE 0xb28963a0 hashq [0xbb4cdd38,0xbb4cdd38] name[0x1fe6a.5] pkey 64046.0
grant 0xb19a8370 cvt (nil) send (nil)@0,0 write (nil),0@65536
flag 0x0 mdrole 0x1 mode 1 scan 0.0 role LOCAL
disk: 0x0000.00000000 write: 0x0000.00000000 cnt 0x0 hist 0x0
xid 0x0000.000.00000000 sid 0 pkwait 0s rmacks 0
refpcnt 0 weak: 0x0000.00000000
pkey 64046.0 undo 0 stat 5 masters[1, 1->2] reminc 13 RM# 18
flg x7 type x0 afftime x8b0628ee, acquire time 0
nreplays by lms 0 = 0
benefit 0, total 0, remote 0, cr benefit 0, cr total 0, cr remote 0
hv 3 [stat 0x0, 1->1, wm 32768, RMno 0, reminc 3, dom 0]
kjga st 0x4, step 0.36.0, cinc 15, rmno 19, flags 0x20
lb 0, hb 8191, myb 8141, drmb 8141, apifrz 1
GCS SHADOW 0xb19a8370,4 resp[0xb28963a0,0x1fe6a.5] pkey 64046.0
grant 1 cvt 0 mdrole 0x1 st 0x100 lst 0x40 GRANTQ rl LOCAL
master 2 owner 1 sid 0 remote[0x92f96c18,3] hist 0x10c1430c0218619
history 0x19.0xc.0x6.0x1.0xc.0x6.0x5.0x6.0x1.0x0.
cflag 0x0 sender 0 flags 0x0 replay# 0 abast (nil).x0.1 dbmap (nil)
disk: 0x0000.00000000 write request: 0x0000.00000000
pi scn: 0x0000.00000000 sq[0xb28963d0,0xb28963d0]
msgseq 0x1 updseq 0x0 reqids[3,0,0] infop (nil) lockseq x13c5
GCS SHADOW END
GCS RESOURCE END
2013-11-15 04:08:35.204351 : 0 write requests issued in 72 GCS resources
0 PIs marked suspect, 0 flush PI msgs
2013-11-15 04:08:35.204493 : 0 write requests issued in 99 GCS resources
0 PIs marked suspect, 0 flush PI msgs
2013-11-15 04:08:35.204648 : 0 write requests issued in 54 GCS resources
0 PIs marked suspect, 0 flush PI msgs
2013-11-15 04:08:35.204790 : 0 write requests issued in 35 GCS resources
0 PIs marked suspect, 0 flush PI msgs
2013-11-15 04:08:35.204948 : 0 write requests issued in 18 GCS resources
0 PIs marked suspect, 0 flush PI msgs
2013-11-15 04:08:35.205048 : 0 write requests issued in 0 GCS resources
0 PIs marked suspect, 0 flush PI msgs
2013-11-15 04:08:35.205327 : 0 write requests issued in 0 GCS resources
0 PIs marked suspect, 0 flush PI msgs
2013-11-15 04:08:35.205590 : 0 write requests issued in 0 GCS resources
0 PIs marked suspect, 0 flush PI msgs

* kjbrrcfwst: resp x0xb28963a0, id1 x1fe6a, drm
2013-11-15 04:08:35.205799 : 0 write requests issued in 1 GCS resources
0 PIs marked suspect, 0 flush PI msgs
DRM(20) win(1) lms 0 finished fixing gcs write protocol
2013-11-15 04:08:35.207051 : kjbldrmnextpkey: AFFINITY pkey 64046.0
pkey 64046.0 undo 0 stat 5 masters[1, 1->2] reminc 13 RM# 18
flg x7 type x0 afftime x8b0628ee, acquire time 0
nreplays by lms 0 = 0
benefit 0, total 0, remote 0, cr benefit 0, cr total 0, cr remote 0
kjblpkeydrmqscchk: Begin for pkey 64046/0 bkt[8192->16383] read-mostly 0
kjblpkeydrmqscchk: End for pkey 64046.0
2013-11-15 04:08:35.207595 : DRM(20) quiesced basts [8192-16383]
2013-11-15 04:08:35.209146 :
* lms acks DRMFRZ (16383)
* lms 0 finished parallel drm freeze in DRM(20) window 2, pcount 47
DRM(20) win(2) lms 0 finished drm freeze
2013-11-15 04:08:35.211327 :
* kjbrrcres: In this window: 256 GCS resources, 0 remastered

* In all windows: 534 GCS resources, 0 remastered
2013-11-15 04:08:35.214797 : kjbldrmnextpkey: AFFINITY pkey 64046.0
pkey 64046.0 undo 0 stat 5 masters[1, 1->2] reminc 13 RM# 18
flg x7 type x0 afftime x8b0628ee, acquire time 0
nreplays by lms 0 = 0
benefit 0, total 0, remote 0, cr benefit 0, cr total 0, cr remote 0
kjbldrmrpst: no more locks for pkey 64046.0, #lk 0, #rm 0, #replay 0, #dup 0 objscan 4
DRM(20) Completed replay (highb 16383)
DRM(20) win(2) lms 0 finished replaying gcs resources
kjbmprlst: DRMFRZ replay msg from 1 (clockp 0x92f96d28, 3)
0x1fe6f.5, 64046.0, x0, lockseq 0x13ca, reqid 0x3
state 0x100, 1 -> 0, role 0x0, wrt 0x0.0
GCS RESOURCE 0xb2896520 hashq [0xbb4edd38,0xbb4edd38] name[0x1fe6f.5] pkey 64046.0
grant 0xb19a3750 cvt (nil) send (nil)@0,0 write (nil),0@65536
flag 0x0 mdrole 0x1 mode 1 scan 0.0 role LOCAL
disk: 0x0000.00000000 write: 0x0000.00000000 cnt 0x0 hist 0x0
xid 0x0000.000.00000000 sid 0 pkwait 0s rmacks 0
refpcnt 0 weak: 0x0000.00000000
pkey 64046.0 undo 0 stat 5 masters[1, 1->2] reminc 13 RM# 18
flg x7 type x0 afftime x8b0628ee, acquire time 0
nreplays by lms 0 = 0

SQL> alter system set "_serial_direct_read"=never;

System altered.

SQL> select count(*) from large_drms;

COUNT(*)
----------
999999

SQL> select blocks from dba_tables where table_name='LARGE_DRMS';

BLOCKS
----------
13233

SQL> select object_id,data_object_id from dba_objects where object_name='LARGE_DRMS';

OBJECT_ID DATA_OBJECT_ID
---------- --------------
64474 64474

With OBJ_IDS As
(Select DATA_Object_Id OBJECT_ID From Dba_Objects
Where Object_Name = 'LARGE_DRMS'), --Customize
Addr As (Select /*+ materialize */
Le_Addr, class, state From X$Bh, OBJ_IDS
Where Object_Id = Obj),
--NOW GET THE RESOURCE NAME IN HEX
Hexnames As (Select
Rtrim(B.Kjblname, ' '||chr(0)) Hexname
From X$Le A, X$Kjbl B, Addr
Where A.Le_Kjbl=B.Kjbllockp
and class = 1 and state <> 3
And A.Le_Addr = Addr.Le_Addr)
-- NOW FIND THE NODE MASTERS
Select A.Master_Node Mast, Count(*)
From Gv$Dlm_Ress A, Hexnames H
Where
Rtrim(A.Resource_Name, ' '||chr(0)) = H.Hexname
Group by A.Master_Node;

MAST COUNT(*)
---------- ----------
1 6605
0 6473

oradebug setmypid
oradebug lkdebug -m pkey 64474;

MAST COUNT(*)
---------- ----------
1 13078

 

DRM一张103MB大小的表可以通过AWR了解因为DRM操作而消耗的带宽等资源:

 

 

 

 

 

 

其中来源于RAC AWR的AWRGRPT报告的ininterconnect_device_statistics显示了正确的DRM网络流量, 而普通AWRRPT中的Interconnect Device Statistics似乎并不能正式反映网络流量。

Comment

*

沪ICP备14014813号-2

沪公网安备 31010802001379号