【翻译自mos文章】当 使用DCD 和TCPS时，rman duplicate hang住

当 使用DCD 和TCPS时，rman duplicate hang住。

来源于：
RMAN Duplicate hangs when using DCD and TCPS (文档 ID 1676197.1)

适用于：
Oracle Database - Enterprise Edition - Version 11.2.0.1 and later
Information in this document applies to any platform.

症状：
在datafile copy 阶段，RMAN active duplicate for standby hang住。SSL Oracle Net 和Dead Connection Detection (DCD) 正在使用。

这个hang 是 间歇性的（intermittent）,也就是说，有时duplicate 是能工作的，在其他时候，会hang住 很多天，直到进程从操作系统和database中kill掉。
rman debug 揭示了下面的信息会repeat：

RMAN-06731: command backup:x% complete, time left HH:MM:SS

样例RMAN debug输出如下：

RMAN-12016: using channel ORA_DISK_8
RMAN-08580: channel ORA_DISK_1: starting datafile copy
RMAN-08522: input datafile file number=00012 name=+OFD_DATA/ofmim01q/datafile/ofm_tbs_oaam_indx.272.810048785
...
RMAN-08581: channel ORA_DISK_4: datafile copy complete, elapsed time: 00:00:16
RMAN-08592: output file name=+OFN_DAT/ofmiy01q/datafile/ofm_ias_iau.373.842790419 tag=TAG20140321T065222
RMAN-08581: channel ORA_DISK_7: datafile copy complete, elapsed time: 00:00:16
RMAN-06731: command backup:94.1% complete, time left 00:21:05


// 
// RMAN-06731 and % complete repeats here
// Process is completely stalled 

 

RMAN-06731: command backup:94.1% complete, time left 00:21:05 

在primary database上，我们可以看到8个session hang住，等待事件"remote db file write" 的wait time会简单的增加

SQL> select SID ,SERIAL# , INST_ID , USERNAME, OSUSER || '@' || MACHINE OSINFO, SUBSTR(PROGRAM,0,20) PROGRAM,
  2  TO_CHAR(LOGON_TIME,'yyyy-mm-dd hh24:mi:ss') LOGON_TIME, EVENT, SECONDS_IN_WAIT SIW from gv$session where type <> 'BACKGROUND' and PROGRAM like 'rman%'
  3  ORDER BY USERNAME, INST_ID, SID;

 SID SERIAL# INST_ID USERNAME OSINFO            PROGRAM              LOGON_TIME           EVENT                                 SIW
---- ------- ------- -------- ----------------- -------------------- -------------------- ------------------------------ ----------
 632    5635       2 SYS      oracle@auq5583l   rman@auq5583l (TNS V 2014-04-08 17:29:27  SQL*Net message from client            34
 758    2535       2 SYS      oracle@auq5583l   rman@auq5583l (TNS V 2014-04-08 17:29:30  SQL*Net message from client             4
 948     441       2 SYS      oracle@auq5583l   rman@auq5583l (TNS V 2014-04-08 17:29:35  remote db file write                52036
1010     369       2 SYS      oracle@auq5583l   rman@auq5583l (TNS V 2014-04-08 17:29:36  remote db file write                35532
1073     215       2 SYS      oracle@auq5583l   rman@auq5583l (TNS V 2014-04-08 17:29:37  remote db file write                52935
1136     291       2 SYS      oracle@auq5583l   rman@auq5583l (TNS V 2014-04-08 17:29:37  remote db file write                54014
1199     753       2 SYS      oracle@auq5583l   rman@auq5583l (TNS V 2014-04-08 17:29:38  remote db file write                41651
1325    1595       2 SYS      oracle@auq5583l   rman@auq5583l (TNS V 2014-04-08 17:29:39  remote db file write                42730
1388    2121       2 SYS      oracle@auq5583l   rman@auq5583l (TNS V 2014-04-08 17:29:39  remote db file write                50771
1451    1351       2 SYS      oracle@auq5583l   rman@auq5583l (TNS V 2014-04-08 17:29:40  remote db file write                47650

hang住的进程必须被从databae和os级别kill掉。

原因：
Unfortunately expire_time + TCPS combination is not supported by oracle as NTZ layer(used for TCPS communication) uses routines that not async-signal-safe.
Using async-signal-safe routines can cause unpredictable results like hang, crash etc.

解决方案：
Do not use DCD with SSL Oracle Net. Remove sqlnet.expire_time from the sqlnet.ora file or set it to 0 (zero).

If you need to keep the connection alive due to firewall issues, consider using the operating system's TCP KEEPALIVE parameters instead. eg:

TCP_KEEPIDLE (the amount of time until the first keepalive packet is sent)
TCP_KEEPCNT (the number of probes to send)
TCP_KEEPINTVL (the interval between keepalive packets)

Otherwise, if you need to use DCD, you must use non-SSL Oracle Net.