SYS@book> @ viewlatch.sql 'gcs partitioned table hash'
ADDR NAME LEVEL# GETS MISSES SLEEPS IMMEDIATE_GETS IMMEDIATE_MISSES WAITERS_WOKEN WAITS_HOLDING_LATCH SPIN_GETS WAIT_TIME
---------------- ---------------------------------------- ---------- ---------- ---------- ---------- -------------- ---------------- ------------- ------------------- ---------- ----------
0000000060018A18 gcs partitioned table hash 6 296 0 0 0 0 0 0 0 0
$ . i5.sh
105343
[1]- Done source peek.sh 'gcs partitioned table hash' 10 | timestamp.pl >|/tmp/peeks_${zdate}.txt
[2]+ Done seq 10 | xargs -I{} echo -e 'sqlplus -s -l / as sysdba <<< @latch_free\nsleep 1' | bash >|/tmp/latch_free_${zdate}.txt
SYS@book> @ viewlatch.sql 'gcs partitioned table hash'
ADDR NAME LEVEL# GETS MISSES SLEEPS IMMEDIATE_GETS IMMEDIATE_MISSES WAITERS_WOKEN WAITS_HOLDING_LATCH SPIN_GETS WAIT_TIME
---------------- ---------------------------------------- ---------- ---------- ---------- ---------- -------------- ---------------- ------------- ------------------- ---------- ----------
0000000060018A18 gcs partitioned table hash 6 446 1 0 0 0 0 0 1 0
--//并发执行150个进程获取gcs partitioned table hash latch,可以发现gets增加150,但是注意1个细节,出现1次MISSES,1次
--//SPIN_GETS。我一直以为大量share latch的请求不会出现阻塞进入SPIN,而实际上的测试还是有少量进入spin。
$ grep -v '^.*: $' /tmp/peeks_105343.txt | cut -c10- | uniq -c
1 SYSDATE LADDR
1 ------------------- ----------------
1 2021-02-08 10:53:43 0000000060018A18
1 Statement processed.
2 [060018A18, 060018A54) = 00000000 00000000 00000128 00000096 00000006 00000010 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
5 [060018A18, 060018A54) = 00000096 00000000 000001BE 00000096 00000006 0000005B 00000004 00000000 00000000 00000001 00000000 00000000 00000000 00000000 00000000
3 [060018A18, 060018A54) = 00000000 00000000 000001BE 00000096 00000006 0000005B 00000004 00000000 00000000 00000001 00000000 00000000 00000000 00000000 00000000
--//150 = 0x96
4.如果看jonathan Lewis<ORALCE 核心技术>中文版本P60:
表4-3
-------------------------------------------------------------------------
需要的栓锁访问 使用方法
-------------------------------------------------------------------------
以共享模式获取其他进程以共享模式持有的共享栓锁 休眠前仅自旋cou_count+2次.
--------------------------------------------------------------------------
--// 英文版本P72:
Table 4-3. Summary of Activity When a Latch Isn't Acquired on the First Attempt
------------------------------------------------------------------------------------------------------------------
Required Latch Access Method Used
------------------------------------------------------------------------------------------------------------------
Get exclusive latch Attempt immediate get, go into spin cycle
once (in this case the process spins 20,000
times), attach to wait list, attempt immediate
get, go to sleep
Get shareable latch in exclusive mode when another Go into spin cycle (in this case the process
process is holding it in some mode (whether shared, spins 2,000 times each cycle), attach to wait
exclusive, or blocking) list, repeat spin cycle, go to sleep if
unsuccessful
来自 “ ITPUB博客 ” ,链接:https://blog.itpub.net/69995498/viewspace-2761198/ ,如需转载,请注明出处,否则将追究法律责任。