【学习笔记】Oracle 11gR2新特性之In-Memory Parallel execution深入研究
时间:2016-11-21 19:44 来源:Oracle研究中心 作者:网络 点击:
次
天萃荷净
Oracle研究中心学习笔记：分享一篇关于Oracle数据库11.2.0.1版本的新特性研究笔记，详细介绍11gR2新特性之In-Memory Parallel execution深入研究。
本站文章除注明转载外，均为本站原创： 转载自love wife & love life —Roger 的Oracle技术博客
本文链接地址: 11gR2 新特性之—In-Memory Parallel execution
该特性是在oracle 11gR2 引入，其目的不言而喻。在11gR2之前，也就是在11gR1中，如果当你发出/*+parallel(16) */ 时，可能会出现下面两种情形：
1. The SQL can run with reduced DOP (be downgraded) -该特性我在11gR1中测试过
2. The SQL can run in serial mode (be serialized)
* "ORA-12827: insufficient parallel query slaves available"
(If PARALLEL_MIN_PERCENT was specified)
根据前面的实验我们可以清楚的看到11gR2中 auto dop的操作过程，做出如下的简单总结：
1. 根据系统默认dop进行计算计算出一个dop;
2. 获取当前实例的负载信息(rac环境会获取所有节点的负载信息);
3. 根据负载信息，结合前面的dop进行计算，得出一个dop值;
4. 多次更新实例负载信息，这里应该还会参考过去的负载信息进行多次计算。
从上面来看，这里计算了2次，我猜测此时这2分实例负载信息类似该实例过去的一份awr快照一样；
5. 最后计算出一个合理的dop值，然后生成执行计划；
6. 执行sql语句(执行之前会进行salve进程的分配)；
7. 执行完毕以后，释放salve进程;
8. 更新实例负载信息(我猜测这次的负载信息会被下次计算作为参考).
但是在11gR2中，彻底发生了改变，首先我们来看看新引入的几个paralle相关的参数：
parallel_degree_policy
该参数属性为manual、auto、limited，11gR2中默认为manual。
---manual
Disables automatic degree of parallelism, statement queuing, and in-memory parallel execution。
---auto
Enables automatic degree of parallelism, statement queuing, and in-memory parallel execution.
---limited
当设置为该属性时，该特性将关闭，部分sql语句仍然可用使用，如表和索引的degree大于1的情况。
parallel_min_time_threshold
sql语句执行的最小时间(在使用了该特性时)，换句话说，也就是只有当
parallel_degree_policy
参数设置为auto或limited时，该参数默认值为auto，即是默认为10s。
parallel_degree_limit
该参数属性为CPU、IO、integer。默认值为CPU
---cpu 意为最大的DOP会根据系统cpu负载来进行自动调节
---io 意为最大的DOP会根据系统IO能力来进行自动调节
---integer 即可以在system或session级别指定为某个具体的数值
该参数是动态参数，可用在session级别进行更改。
parallel_force_local
顾名思义，该参数主要用于RAC环境，控制parallel server processes 是否能够跨节点，
其属性为true、false，默认值为false。该参数为动态参数。
parallel_servers_target
该参数的含义是可用的parallel server processes
该参数值=4 x CPU_COUNT x PARALLEL_THREADS_PER_CPU x ACTIVE_INSTANCES
parallel_max_servers
最大的parallel进程，parallel_servers_target < parallel_max_servers.
11gR2还引入了一个新的包DBMS_PARALLEL_EXECUTE，这里不多说。
SQL> SELECT n.name, s.VALUE
2 FROM v$mystat s, v$statname n
3 WHERE s.statistic# = n.statistic#
4 AND n.name LIKE 'Parallel%'
5 /
NAME VALUE
---------------------------------------------------------------- ----------
Parallel operations NOT downgraded 0
Parallel operations downgraded TO serial 0
Parallel operations downgraded 75 TO 99 pct 0
Parallel operations downgraded 50 TO 75 pct 0
Parallel operations downgraded 25 TO 50 pct 0
Parallel operations downgraded 1 TO 25 pct 0
SQL> ALTER SESSION SET "_px_trace"="none";
SESSION altered.
SQL> ALTER SESSION SET "_px_trace"=ALL;
SESSION altered.
SQL> SELECT /*+ parallel(4) */ COUNT(*) FROM ht02;
COUNT(*)
----------
225824
SQL> ALTER SESSION SET "_px_trace"="none";
SESSION altered.
我们来看下trace信息：
*** 2011-09-03 00:51:23.921
kxfrDefaultDOP
DOP Trace -- compute default DOP
# CPU = 1
Threads/CPU = 2 ("parallel_threads_per_cpu")
default DOP = 2 (# CPU * Threads/CPU)
default DOP = 2 (DOP * # instance)
Default DOP = 2
kxfxqOnOrderQueue
KXFXQQUEUABLE() is TRUE.
pgadep: 0, pgatopsql: 1, pgapls 0.
SlaveSQL : NO, Parallized : YES, DOP: 4.
kxfxqOnOrderQueue
Admitting Parallel Statement (dop:4):
----- Current SQL Statement for this session (sql_id=432y0dmm1qdzk) -----
SELECT /*+ parallel(4) */ count(*) from ht02
kxfxqUpdateLoad
snapshot of RAC load before update: [ total queued PQ: 0, total running(
admitted) PQ: 0, total granted slaves in RAC: 0, total target in RAC: 0
]
kxfpiinfo
inst[cpus:mxslv]
1[1:20] ######## 表示1个cpu，parallel_max_servers参数值为20. ########
kxfpGetNumActiveSlaves
number of active slaves on the instance: 0
kxfpGetDefInstTarget
default inst target is 8, defDOP: 2, mxu: 2, cpus: 1
kxfpGetInstTarget
(default: 1) inst target is 8 ######## 这里是parallel_servers_target值 ########
kxfpclinfo
inst(load:user:pct:fact:queued:started:granted:active)aff
1 (1:0:100:100:0:0:0)
kxfpGetDefInstTarget
default inst target is 8, defDOP: 2, mxu: 2, cpus: 1
kxfpGetInstTarget
(default: 1) inst target is 8
kxfpMarkRACLoadStat
RAC load statistics is marked as valid.
........
Sending parse to slave set 1:
User sqllen sent from QC = 45
SELECT /*+ parallel(4) */ count(*) from ht02
kxfxpf [ 1430/ 60]
MSG( -->, KXFXOparse, DIALOG_HINT, slv=0 )
kxfxpf [ 1440/ 10]
MSG( -->, KXFXOparse, DIALOG_HINT, slv=1 )
kxfxpf [ 1440/ 0]
MSG( -->, KXFXOparse, DIALOG_HINT, slv=2 )
kxfxpf [ 1440/ 0]
MSG( -->, KXFXOparse, DIALOG_HINT, slv=3 )
kxfxgs [ 1640/ 200]
MSG(
kxfxgs [ 1640/ 0]
MSG(
kxfxgs [ 1640/ 0]
MSG(
kxfxgs [ 1640/ 0]
MSG(
kxfpgsg [ 1640/ 0]
Freeing Memory: il=0x4835d0 iload=0x48357c ilist=(nil) slist=(nil)
set1_pids=0x4835ec set2_pids=(nil)
kxfrAllocSlaves [ 1640/ 0]
actual num slaves alloc'd = 4 (kxfpqcthr) ######## 实际分配的slave进程个数 ########
kxfrialo [ 1640/ 0]
Finish: allocated actual 4 slaves for non-GV query
..............
*** 2011-09-03 00:51:25.390
kxfpg1sg [ 1280/ 30]
received reply from qref 0x24f811b0
kxfpg1sg [ 1280/ 0]
got 4 servers (sync), errors=0x0 returning
GROUP GET [ 1280/ 0]
Acquired 4 slaves on 1 instances avg height=4 in 1 set q serial:513
P000 inst 1 spid 4291 ######## 这里是分配的4个salve进程spid ########
P001 inst 1 spid 4293
P002 inst 1 spid 4295
P003 inst 1 spid 4297
Insts 1
Svrs 4 ######## 4个salve进程 ########
当前虚拟机测试，我分配了4个parallel，现在加大该值为30，看看最后的DOP会是多少。
SQL> ALTER SESSION SET parallel_degree_policy = auto;
SESSION altered.
SQL> ALTER SESSION SET "_px_trace"=ALL;
SESSION altered.
SQL> SELECT /*+ parallel(30) */ COUNT(*) FROM ht02;
COUNT(*)
----------
225824
SQL> ALTER SESSION SET "_px_trace"="none";
SESSION altered.
再次来看看trace。
kxfpg1sg [ 4470/ 0]
got 20 servers (sync), errors=0x0 returning
GROUP GET [ 4490/ 20]
Acquired 20 slaves on 1 instances avg height=20 in 1 set q serial:1025
P000 inst 1 spid 4354
P001 inst 1 spid 4356
P002 inst 1 spid 4358
P003 inst 1 spid 4360
P004 inst 1 spid 4362
P005 inst 1 spid 4364
P006 inst 1 spid 4366
P007 inst 1 spid 4368
P008 inst 1 spid 4370
P009 inst 1 spid 4372
P010 inst 1 spid 4374
P011 inst 1 spid 4376
P012 inst 1 spid 4378
P013 inst 1 spid 4380
P014 inst 1 spid 4382
P015 inst 1 spid 4384
P016 inst 1 spid 4386
P017 inst 1 spid 4388
P018 inst 1 spid 4390
P019 inst 1 spid 4392
Insts 1
Svrs 20
###### 可以看到只分配了20分slave进程，为啥呢？因为这里受到参数parallel_max_servers的限制。######
SQL> SHOW parameter parallel_max_servers
NAME TYPE VALUE
------------------------------------ ----------- ------------------------------
parallel_max_servers INTEGER 100
SQL> conn roger/roger
SQL> ALTER SESSION SET parallel_degree_policy = auto;
SESSION altered.
SQL> ALTER SESSION SET "_px_trace"=ALL;
SESSION altered.
SQL> SELECT /*+ parallel(100) */ COUNT(*) FROM ht02;
COUNT(*)
----------
225824
SQL> ALTER SESSION SET "_px_trace"="none";
SESSION altered.
此时trace 信息如下：
grep -i inst 1 spid roger_ora_4507.trc
roger_ora_4507.trc: Acquired 100 slaves on 1 instances avg height=100 in 1 set q serial:51
roger_ora_4507.trc: P000 inst 1 spid 4509
roger_ora_4507.trc: P001 inst 1 spid 4511
roger_ora_4507.trc: P002 inst 1 spid 4513
roger_ora_4507.trc: P003 inst 1 spid 4515
.......
roger_ora_4507.trc: P094 inst 1 spid 4698
roger_ora_4507.trc: P095 inst 1 spid 4700
roger_ora_4507.trc: P096 inst 1 spid 4702
roger_ora_4507.trc: P097 inst 1 spid 4704
roger_ora_4507.trc: P098 inst 1 spid 4706
roger_ora_4507.trc: P099 inst 1 spid 4708
下面来进行auto DOP的测试。
首先说明下DOP的计算公式：
单实例： DOP = PARALLEL_THREADS_PER_CPU x CPU_COUNT
RAC： DOP = PARALLEL_THREADS_PER_CPU x CPU_COUNT x INSTANCE_COUNT
SQL> ALTER SESSION SET parallel_degree_policy = auto;
SESSION altered.
SQL> ALTER SESSION SET "_px_trace"=ALL;
SESSION altered.
SQL> SELECT COUNT(*) FROM ht02;
Execution Plan
----------------------------------------------------------
Plan hash VALUE: 583574080
-------------------------------------------------------------------
| Id | Operation | Name | ROWS | Cost (%CPU)| TIME |
-------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 446 (1)| 00:00:06 |
| 1 | SORT AGGREGATE | | 1 | | |
| 2 | TABLE ACCESS FULL| HT02 | 225K| 446 (1)| 00:00:06 |
-------------------------------------------------------------------
Note
-----
- automatic DOP: Computed Degree OF Parallelism IS 1 because OF parallel threshold
此时自动计算出来的DOP 为1,说明当前sql语句在parallel为1的情况下效率最高。
SQL> SET autot off
SQL> SET timing ON
SQL> SELECT COUNT(*) FROM ht02;
COUNT(*)
----------
225824
Elapsed: 00:00:00.08
SQL> SELECT /*+ parallel(2) */ COUNT(*) FROM ht02;
COUNT(*)
----------
225824
Elapsed: 00:00:00.22
SQL> SELECT /*+ parallel(4) */ COUNT(*) FROM ht02;
COUNT(*)
----------
225824
Elapsed: 00:00:00.74
SQL> SELECT /*+ parallel(6) */ COUNT(*) FROM ht02;
COUNT(*)
----------
225824
Elapsed: 00:00:00.81
SQL> SELECT /*+ parallel(8) */ COUNT(*) FROM ht02;
COUNT(*)
----------
225824
Elapsed: 00:00:00.92
SQL> SELECT /*+ parallel(10) */ COUNT(*) FROM ht02;
COUNT(*)
----------
225824
Elapsed: 00:00:01.15
SQL> SELECT /*+ parallel(20) */ COUNT(*) FROM ht02;
COUNT(*)
----------
225824
Elapsed: 00:00:03.24
SQL> SELECT /*+ parallel(40) */ COUNT(*) FROM ht02;
COUNT(*)
----------
225824
Elapsed: 00:00:09.11
SQL> SELECT /*+ parallel(60) */ COUNT(*) FROM ht02;
COUNT(*)
----------
225824
Elapsed: 00:00:23.75
SQL> SELECT /*+ parallel(80) */ COUNT(*) FROM ht02;
COUNT(*)
----------
225824
Elapsed: 00:00:29.80
根据上述数据制作图表如下：
我们知道，当前测试是在parallel_min_time_threshold为默认值的情况下进行的测试，下面更改该值。
SQL> ALTER SESSION SET parallel_min_time_threshold=3;
SESSION altered.
Elapsed: 00:00:00.02
SQL> ALTER system FLUSH shared_pool;
此时的情况如下：
从上面来看，当该值修改为3以后，Dop为2时效率是最高的，下面来看看起执行计划是否如此。
SQL> SET autot traceonly
SQL> ALTER system FLUSH shared_pool;
System altered.
Elapsed: 00:00:00.55
SQL> SELECT COUNT(*) FROM ht02;http://www.oracleplus.net
Elapsed: 00:00:01.50
Execution Plan
----------------------------------------------------------
Plan hash VALUE: 2508058984
--------------------------------------------------------------------------------------------------------
| Id | Operation | Name | ROWS | Cost (%CPU)| TIME | TQ |IN-OUT| PQ Distrib |
--------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 248 (1)| 00:00:03 | | | |
| 1 | SORT AGGREGATE | | 1 | | | | | |
| 2 | PX COORDINATOR | | | | | | | |
| 3 | PX SEND QC (RANDOM) | :TQ10000 | 1 | | | Q1,00 | P->S | QC (RAND) |
| 4 | SORT AGGREGATE | | 1 | | | Q1,00 | PCWP | |
| 5 | PX BLOCK ITERATOR | | 225K| 248 (1)| 00:00:03 | Q1,00 | PCWC | |
| 6 | TABLE ACCESS FULL| HT02 | 225K| 248 (1)| 00:00:03 | Q1,00 | PCWP | |
--------------------------------------------------------------------------------------------------------
Note
-----
- automatic DOP: Computed Degree OF Parallelism IS 2
Statistics
----------------------------------------------------------
377 recursive calls
12 db block gets
1444 consistent gets
1335 physical reads
0 redo SIZE
424 bytes sent via SQL*Net TO client
415 bytes received via SQL*Net FROM client
2 SQL*Net roundtrips TO/FROM client
6 sorts (memory)
0 sorts (disk)
1 ROWS processed
这里补充一点是，如果参数parallel_degree_policy为manual时，我们可以使用parallel hint来使用该特性，如下：
SQL> SELECT /*+ parallel(auto) */ COUNT(*) FROM ht02;
Execution Plan
----------------------------------------------------------
Plan hash VALUE: 583574080
-------------------------------------------------------------------
| Id | Operation | Name | ROWS | Cost (%CPU)| TIME |
-------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 446 (1)| 00:00:06 |
| 1 | SORT AGGREGATE | | 1 | | |
| 2 | TABLE ACCESS FULL| HT02 | 225K| 446 (1)| 00:00:06 |
-------------------------------------------------------------------
Note
-----
- automatic DOP: Computed Degree OF Parallelism IS 1 because OF parallel threshold
Statistics
----------------------------------------------------------
0 recursive calls
0 db block gets
1338 consistent gets
1335 physical reads
0 redo SIZE
424 bytes sent via SQL*Net TO client
415 bytes received via SQL*Net FROM client
2 SQL*Net roundtrips TO/FROM client
0 sorts (memory)
0 sorts (disk)
1 ROWS processed
SQL> ALTER SESSION SET parallel_min_time_threshold=5;
SESSION altered.
SQL> SELECT /*+ parallel(auto) */ COUNT(*) FROM ht02;
Execution Plan
----------------------------------------------------------
Plan hash VALUE: 2508058984
--------------------------------------------------------------------------------------------------------
| Id | Operation | Name | ROWS | Cost (%CPU)| TIME | TQ |IN-OUT| PQ Distrib |
--------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 248 (1)| 00:00:03 | | | |
| 1 | SORT AGGREGATE | | 1 | | | | | |
| 2 | PX COORDINATOR | | | | | | | |
| 3 | PX SEND QC (RANDOM) | :TQ10000 | 1 | | | Q1,00 | P->S | QC (RAND) |
| 4 | SORT AGGREGATE | | 1 | | | Q1,00 | PCWP | |
| 5 | PX BLOCK ITERATOR | | 225K| 248 (1)| 00:00:03 | Q1,00 | PCWC | |
| 6 | TABLE ACCESS FULL| HT02 | 225K| 248 (1)| 00:00:03 | Q1,00 | PCWP | |
--------------------------------------------------------------------------------------------------------
Note
-----
- automatic DOP: Computed Degree OF Parallelism IS 2
Statistics
----------------------------------------------------------
8 recursive calls
4 db block gets
1394 consistent gets
1335 physical reads
0 redo SIZE
424 bytes sent via SQL*Net TO client
415 bytes received via SQL*Net FROM client
2 SQL*Net roundtrips TO/FROM client
0 sorts (memory)
0 sorts (disk)
1 ROWS processed
关于该特性，是针对DW环境的，auto dop的计算，其实在11gR1就有了，不过计算的算法不太合理，存在一定的缺陷。
最后我们来看下11gR2中auto dop是如何计算的。
kxfrDefaultDOP
DOP Trace -- compute default DOP
# CPU = 1
Threads/CPU = 2 ("parallel_threads_per_cpu")
default DOP = 2 (# CPU * Threads/CPU)
default DOP = 2 (DOP * # instance)
Default DOP = 2
kxfrDefaultDOP
DOP Trace -- compute default DOP
# CPU = 1
Threads/CPU = 2 ("parallel_threads_per_cpu")
default DOP = 2 (# CPU * Threads/CPU)
default DOP = 2 (DOP * # instance)
Default DOP = 2 ###### 系统默认的DOP ######
kxfxqOnOrderQueue
KXFXQQUEUABLE() is TRUE.
pgadep: 0, pgatopsql: 1, pgapls 0.
SlaveSQL : NO, Parallized : YES, DOP: 2.
kxfxqOnOrderQueue
Admitting Parallel Statement (dop:2):
----- Current SQL Statement for this session (sql_id=b4npv3kz33xb5) -----
select count(*) from ht02
kxfxqUpdateLoad ###### 更新负载情况 ######
snapshot of RAC load before update: [ total queued PQ: 0, total running(
admitted) PQ: 0, total granted slaves in RAC: 0, total target in RAC: 8
]
kxfpuqpq ###### 更新queue PQ ######
instance load stat of queued PQ updated.
number of queued PQ is incremented from 0 to 1.
load stat(queued PQ) on all RAC has been updated.
snapshot of RAC load stat [ total queued PQ: 1, total running(admitted)
PQ: 0, total granted slaves in RAC: 0, total target in RAC: 8 ]
kxfxqsoc ###### 创建state object(应该就是一个SQL语句的标示类似SQL_ID) ######
state object created [stmt id: 16777230, exetime: 1315063063, queued tr
ue, starting false, granted slaves: 0]
kxfxqInstInfo ###### 检查实例信息 ######
inst[cpus:mxslv]
1[1:100] ###### 1 描述为1个节点 100描述parallel_max_servers为100 ######
kxfpGetNumActiveSlaves ###### 获得当前处于活动的savle进程 ######
number of active slaves on the instance: 0
kxfpGetDefInstTarget ###### 获取实例默认parallel_target_servers参数值 ######
default inst target is 8, defDOP: 2, mxu: 2, cpus: 1
kxfpGetInstTarget
(default: 1) inst target is 8
kxfxqLocalInstLoad ###### 获取本地节点的实例负载 ######
local inst(load:user:pct:fact:queued:admitted:started:granted:active)
1 (0:0:100:0:1:0:0:0:0)
kxfxqInstLoad ###### 获取实例的负载信息(如果是RAC的话，应该跟上面的不同) ######
inst(load:user:pct:fact:queued:admitted:started:granted:active)
1(local) (0:0:100:0:1:0:0:0:0)
kxfxqInstList ###### 初始化实例负载信息 ######
load information of 1 instances (single inst) initialized
kxfxqGrantedDOP ###### 根据系统负载情况计算合理的DOP值 ######
Computing granted DOP.
kxfxqGrantedDOP ###### 根据前面的计算结果分配dop即是分配salve进程 ######
RequestedDOP=2 GrantedDOP=2 Target=8 Load=0 GrantedSlv=0 AdmittedPQ=0 De
faultDOP=0 users=0 sets=1 force_admit=false
kxfxqUpdateLoad ###### 再次更新实例负载信息 ######
snapshot of RAC load before update: [ total queued PQ: 1, total running(
admitted) PQ: 0, total granted slaves in RAC: 0, total target in RAC: 8
]
kxfpuqpq ###### 再次更新queue PQ ######
instance load stat of queued PQ updated.
number of queued PQ is decremented from 1 to 0.
load stat(queued PQ) on all RAC has been updated.
snapshot of RAC load stat [ total queued PQ: 0, total running(admitted)
PQ: 0, total granted slaves in RAC: 0, total target in RAC: 8 ]
kxfxqUpdateLoad ###### 第3次更新实例负载信息 ######
snapshot of RAC load before update: [ total queued PQ: 0, total running(
admitted) PQ: 0, total granted slaves in RAC: 0, total target in RAC: 8
]
kxfpAdjustGrantedSlaves ###### 根据前面的多次调整，决定是否需要调整dop即salve进程 ######
gslv is not adjusted.sga total gslv: 0, glsv: 2, adjusted gslv 2.
kxfpurpq
instance load stat of admitted PQ updated.
number of admitted PQ is incremented from 0 to 1, total granted slaves
is incremented from 0 to 2.
load stat(running(admitted) PQ) on all RAC has been updated.
snapshot of RAC load stat [ total queued PQ: 0, total running(admitted)
PQ: 1, total granted slaves in RAC: 2, total target in RAC: 8 ]
kxfxqsou
state object updated [stmt id: 16777230, exe time: 1315063063, queued f
alse, starting true, granted slaves: 2, remove state obj false
kxfxqOnOrderQueue
Statement bypasses the queue.
Starting parallelizer rwsid:2 pxid:1
qerpxStart [ 0/ 0]
Start:
Starting SQL statement dump
SQL Information
user_id=85 user_name=ROGER module=SQL*Plus action=
sql_id=b4npv3kz33xb5 plan_hash_value=-1786908312 problem_type=3
----- Current SQL Statement for this session (sql_id=b4npv3kz33xb5) -----
select count(*) from ht02
sql_text_length=26
sql=select count(*) from ht02
----- Explain Plan Dump -----
----- Plan Table -----
============
Plan Table
============
------------------------------------------+-----------------------------------+-------------------------+
| Id | Operation | Name | Rows | Bytes | Cost | Time | TQ |IN-OUT|PQ Distrib |
------------------------------------------+-----------------------------------+-------------------------+
| 0 | SELECT STATEMENT | | | | 248 | | | | |
| 1 | SORT AGGREGATE | | 1 | | | | | | |
| 2 | PX COORDINATOR | | | | | | | | |
| 3 | PX SEND QC (RANDOM) | :TQ10000| 1 | | | |:Q1000| P->S |QC (RANDOM)|
| 4 | SORT AGGREGATE | | 1 | | | |:Q1000| PCWP | |
| 5 | PX BLOCK ITERATOR | | 221K | | 248 | 00:00:03 |:Q1000| PCWC | |
| 6 | TABLE ACCESS FULL | HT02 | 221K | | 248 | 00:00:03 |:Q1000| PCWP | |
------------------------------------------+-----------------------------------+-------------------------+
...............
GROUP GET [ 470/ 0]
Acquired 2 slaves on 1 instances avg height=2 in 1 set q serial:23041
P000 inst 1 spid 7835
P001 inst 1 spid 7837
Insts 1
Svrs 2
kxfpValidateSlaveGroup [ 470/ 0]
qcq:0x2292fc94 flg:0
qerpxSendParse [ 470/ 0]
qcq=0x2292fc94 pxid=1 mflg=0x0 #slaves=2
kxfxcp1 [ 470/ 0]
Sending parse to nprocs:2 slave_set:1
kxfxcPutSession [ 520/ 50]
................
kxfpqsrls [ 820/ 10]
Release Slave q=0x2292fc94 qr=0x2247e860 action=1 slave=1 inst=1 ###### 这里是释放salve进程 ######
kxfpqsrls [ 820/ 0]
Release Slave q=0x2292fc94 qr=0x22481710 action=1 slave=0 inst=1
GROUP RELEASE [ 820/ 0]
all slaves released q serial 23041
kxfpqsod_qc_sod [ 820/ 0]
clean up of q=0x2292fc94 completed
kxfrfir [ 820/ 0]
cbk fired: 0x9d8b24
kxfxqsou [ 820/ 0]
state object removed [stmt id: 16777230, exe time: 1315063063] ###### 删除创建的state object ######
kxfrfir [ 820/ 0]
cbk fired: 0x9d8b14
kxfxqRPQcbk [ 820/ 0]
Decr admitted parallel statement load
kxfxqUpdateLoad [ 820/ 0] ###### 更新负载信息 ######
snapshot of RAC load before update: [ total queued PQ: 0, total running(
admitted) PQ: 1, total granted slaves in RAC: 2, total target in RAC: 8
]
kxfpAdjustGrantedSlaves [ 820/ 0]
gslv is not adjusted.sga total gslv: 2, glsv: 2, adjusted gslv 2.
kxfpurpq [ 820/ 0]
instance load stat of admitted PQ updated. ###### 更新queue PQ ######
number of admitted PQ is decremented from 1 to 0, total granted slaves
is decremented from 2 to 0.
load stat(running(admitted) PQ) on all RAC has been updated.
snapshot of RAC load stat [ total queued PQ: 0, total running(admitted)
PQ: 0, total granted slaves in RAC: 0, total target in RAC: 8 ]
根据前面的实验我们可以清楚的看到11gR2中 auto dop的操作过程，做出如下的简单总结：
1. 根据系统默认dop进行计算计算出一个dop;
2. 获取当前实例的负载信息(rac环境会获取所有节点的负载信息);
3. 根据负载信息，结合前面的dop进行计算，得出一个dop值;
4. 多次更新实例负载信息，这里应该还会参考过去的负载信息进行多次计算。
从上面来看，这里计算了2次，我猜测此时这2分实例负载信息类似该实例过去的一份awr快照一样；
5. 最后计算出一个合理的dop值，然后生成执行计划；
6. 执行sql语句(执行之前会进行salve进程的分配)；
7. 执行完毕以后，释放salve进程;
8. 更新实例负载信息(我猜测这次的负载信息会被下次计算作为参考).
个人见解，希望能够抛砖引玉!
--------------------------------------ORACLE-DBA----------------------------------------
最权威、专业的Oracle案例资源汇总之【学习笔记】Oracle 11gR2新特性之In-Memory Parallel execution深入研究
扫一扫
603
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
