"Cross Instance call Enqueue"是一种在一个或多个instance实例间调用后台进程行为时用到的队列锁,具体调用的后台进程行为包括检查点checkpoint、日志切换logfile switch、shutdown实例、载入数据文件头等等。需要注意的是这种Enqueue Lock并不仅仅在RAC中使用，即便是单节点也会用到。CI锁的数量取决于并行执行Cross Instance Call调用的进程的总数。

SQL> col ksqsttyp for a20
SQL> col ksqstrsn for a20
SQL> col ksqstexpl for a80
SQL> set linesize 200 pagesize 2000;
SQL> select ksqsttyp,ksqstrsn,ksqstexpl from x$ksqst where ksqsttyp='CI';KSQSTTYP             KSQSTRSN             KSQSTEXPL
-------------------- -------------------- --------------------------------------------------------------------------------
CI                   contention           Coordinates cross-instance function invocationsSQL> show parameter cluster_databaseNAME                                 TYPE        VALUE
------------------------------------ ----------- ------------------------------
cluster_database                     boolean     FALSE
cluster_database_instances           integer     1SQL> select * from v$enqueue_stat where eq_type='CI';INST_ID EQ TOTAL_REQ# TOTAL_WAIT#  SUCC_REQ# FAILED_REQ# CUM_WAIT_TIME
---------- -- ---------- ----------- ---------- ----------- -------------1 CI        595           0        595           0             0

当系统中出现有大量这种跨实例后台进程调用时，将出现CI队列锁的争用。假设在一个RAC场景中，同时有大量的回话开始对不同的数据表执行TRUNCATE截断操作，TRUNCATE的一个前提是在所有实例上(因为对象表的dirty buffer可能分布在多个实例上)发生对象级别的检查点(object level checkpoint)，检查点发生时CKPT进程会通知DBWR写出指定对象表相关的脏块，DBWR需要扫描Buffer Cache以找出脏块，而如果Buffer Cache很大那么扫描将花费大量的时间，而在此过程中前台进程将一直排他地持有着本地的CI队列锁，这就将造成CI锁的严重争用。 为了减少CI队列锁地争用，我们第一步所要做的是找出实际的Cross Instance call跨实例调用的类型。这里要另外提一下的是在10g以前不管是v$session_wait或statspack中都不会将enqueue锁等待事件的具体enqueue lock类型写明，一般需要我们从p1/p2/p3列中找出enqueue的具体身份，例如"WAIT #1: nam='enqueue' ela= 910796 p1=1128857606 p2=1 p3=4"，这里的p1为1128857606也就是16进制的43490006，高位的'4349'转换为ascii码也就是'CI'，而这里的p2/p3对应为V$lock中的ID1/ID2，ID1=1代表了"Reuse (checkpoint and invalidate) block range"，ID2=4代表了"Mounted excl, use to allocate mechanism"。 具体ID1/ID2代表的含义在不同版本中有所变化，可以参考下表:

Id1, Id2 Combination:Oracle 10gR127 TO 29  *Same as 9i R230    process waiters after row cache requeue31    Active Change Directory extent relocation32    block change tracking state change33    kgl mulitversion obsolete34    set previous resetlogs data35    set recovery destination pointer36    fast object reuse request37    test ksbcic()38    ASM diskgroup discovery wait39    ASM diskgroup release40    ASM push DB updates41    ASM add ACD chunk42    ASM map resize message43    ASM map lock message44    ASM map unlock message (phase 1)45    ASM map unlock message (phase 2)46    ASM generate add disk redo marker47    ASM check of PST validity48    ASM offline disk CIC49    Logical Standby Sync Point SCN50    update SQL Tuning Base existence bitvector51    PQ induced Checkpointing52    ASM F1X0 relocation53    Scheduler autostart54    KZS increment grant/revoke counter55    ASM disk operation message56    ASM I/O error emulation57    DB Supp log cursor invalidation58    Cache global range invalidation59    Cache global object invalidation60    ASM Pre-Existing Extent Lock wait61    Perform a ksk action through DBWR62    ASM diskgroup refresh waitOracle 10gR230 to 62 *Same as 10gR163    KCBO object checkpoint64    KCBO object pq checkpoint65    global health check event66    Oracle Label Security refresh67    thread internal enable68    cross-instance registration69    KGL purge unused subheaps70    clear pin instance flag71    Rolling operations CICOracle 9iR2Id1   Meaning~~~   ~~~~~~25 TO 26  *Same as 9i R127    set Database Force Logging mode28    invalidate cached file address translations29    Cursor Unauthorize Mode30    snapshot too old diagnosis31    process waiters after row cache requeueOracle 9iR1Id1    Meaning~~~    ~~~~~~~~18 TO 24  *Same as Oracle 8i25     Update Dscn Tracking (ktcndt)26     Purge dictionary Object number CacheOracle 8iId1     Meaning~~~     ~~~~~~~0 TO 17    *Same as Oracle 8(please see "Oracle8*" for the Meaning)18      Object reuse request19      Rolling release checks20      Propagate begin backup scn for a file21      Refresh top plan (for db scheduler)22      Clear checkpoint progress record23      Drop temp file24      Quiesce database RestrictedId2      Meaning~~~      ~~~~~~~0x01     Used to pass in parameters0x02     Used to invoke the function in backgroud process0x03     Used to indicate the foreground has not returned0x04     Mounted excl, use to allocate mechanism0x05     Used to queue up interested clientsOracle 8*Id1   Meaning~~~  ~~~~~~~0       Checkpoint block range1       Reuse (checkpoint and invalidate) block range2       LGWR Checkpointing and Hot Backup3       DBWR syncronization of SGA with control file4       Log file add/drop/rename notification5       Write buffer for CR read6       Test call7       Invalidate KCK cache in all instances8       Alter rollback segment optimal9       Signal Query Servers/coordinator10      Create Remote parallel query Server11      Set Global Partitions12      Stop Disk Writes13      Drop Sort Segments14      Release unused space from Sort Segments15      Instance Recovery for Parallel operation Group16      Validate parallel slave Lock Value17      Check transaction state objects18      Flush blocks in object19      Rolling release checks20      Propagate begin backup scn for a file21      Clear checkpoint progress recordOracle 7Id1    Meaning~~~  ~~~~~~~0    Flush buffers for reuse as new class1   LGWR checkpointing and Hot Backup2  DBWR synchronization of SGA with control file3  Log file add/drop/rename notification4  Write buffer for CR read5   Test Call6  Invalidate KCK cache in all instances7  Alter rollback segment optimal8     Signal Query Servers/coordinator9   Create Remote Parallel Query Server10   Set Global Partitions11     Stop Disk Writes12  Drop Sort Segments13    Release unused space from Sort Segments14   Instance Recovery for Parallel operation Group15    Validate parallel slave Lock Value16    Check Transaction State ObjectsId2  Meaning~~~  ~~~~~~~1    Pass in Parameters2     Invoke the call in background process3  Foreground has not returned yet4    Used to allocate the CI call5   Used to queue up interested clients

本文转自maclean_007 51CTO博客，原文链接:http://blog.51cto.com/maclean/1277792