Today, while installing RAC, the installation began to wait at “Performing remote operations” step. I’ve tried to check log files, but there was no information written

Then I tried to trace the currently running process, but didn’t get enough information about the wait. At last, I checked both servers and found firewall running. After stopping firewall, (stopping, deinstalling and starting the installation again) the setup was able to pass that step and copied the Grid home files to the remote server

Today, while checking both V$PROCESS and V$SESSION view, I realized that there’re some processes that are missing from V$SESSION view:

SQL> select count(*) from v$process where addr not in (select paddr from v$session);

 COUNT(*)

———-
120

How can a process be created without having any corresponding entry in V$SESSION view?

I started to check those processes on OS level. I got SPID of one of those processes and firstly tried to debug it using ORADEBUG from sqlplus:

sqlplus / as sysdba

ORADEBUG SETOSPID 3422
ORADEBUG TRACEFILE_NAME;

By getting the trace file name, I’ve checked the file but didn’t find any helpful message inside.

Then I decided to debug it on OS level. As the OS was HP-UX, I used truss executable to trace the unix system calls:

/usr/local/bin/truss -o truss.out -vall -laefdD -p 3422

And checked the output again:

tail -f truss.out
( Attached to process 18997 (“oraclePRODDB (LOCAL=NO)”) [64-bit] )
18997/9934051: 18446744073.7095 read(16, 0x600000000021a616, 2064) [sleeping]

This output wasn’t helpful (however, Tanel uses pfiles executable to get all opened files in the process by getting the first parameter of the read function – which is 16) , so I decided to use the next executable – lsof – which lists the opened files by the process:

/usr/local/bin/lsof -p 3422

COMMAND PID USER FD TYPE DEVICE SIZE/OFF NODE NAME
oracle 20472 oracle cwd DIR 64,0×4 2048 3660 /home/oracle/product/10g/dbs
oracle 20472 oracle txt REG 64,0×4 284477360 101589 /home/oracle/product/10g/bin/oracle
oracle 20472 oracle mem REG 64,0×4 5337208 25311 /home/oracle/product/10g/lib/libnnz10.so
oracle 20472 oracle mem REG 64,0×7 21281 27339 /usr/lib/tztab
oracle 20472 oracle mem REG 64,0×5 312440 54074 /opt/star-ncf-prod/ep_patch/usr/lib/hpux64/libxti.so.1
oracle 20472 oracle mem REG 64,0×7 4371800 14392 /usr/lib/hpux64/libc.so.1
oracle 20472 oracle mem REG 64,0×7 1649152 91060 /usr/lib/hpux64/libnsl.so.1
oracle 20472 oracle mem REG 64,0×7 67664 91056 /usr/lib/hpux64/libnss_compat.so.1
oracle 20472 oracle mem REG 64,0×4 20041776 8569 /home/oracle/product/10g/lib/libjox10.so
oracle 20472 oracle mem REG 64,0×7 1046480 11645 /usr/lib/hpux64/libpthread.so.1
oracle 20472 oracle mem REG 64,0×7 55088 491 /usr/lib/hpux64/libuca.so.1
oracle 20472 oracle mem REG 64,0×7 706832 91078 /usr/lib/hpux64/libunwind.so.1
oracle 20472 oracle mem REG 64,0×7 6444800 91068 /usr/lib/hpux64/libm.so.1
oracle 20472 oracle mem REG 64,0×7 78168 90961 /usr/lib/hpux64/libdl.so.1
oracle 20472 oracle mem REG 64,0×7 89208 91053 /usr/lib/hpux64/libnss_dns.so.1
oracle 20472 oracle mem REG 64,0×7 44832 3881 /usr/lib/hpux64/librt.so.1
oracle 20472 oracle mem REG 64,0×4 199296 8277 /home/oracle/product/10g/lib/libdbcfg10.so
oracle 20472 oracle mem REG 64,0×4 134552 13143 /home/oracle/product/10g/lib/libocrutl10.so
oracle 20472 oracle mem REG 64,0×4 533792 13142 /home/oracle/product/10g/lib/libocrb10.so
oracle 20472 oracle mem REG 64,0×4 731352 25330 /home/oracle/product/10g/lib/libocr10.so
oracle 20472 oracle mem REG 64,0×4 1296848 25478 /home/oracle/product/10g/lib/libhasgen10.so
oracle 20472 oracle mem REG 64,0×4 315008 13264 /home/oracle/product/10g/lib/libskgxp10.so
oracle 20472 oracle mem REG 64,0×7 1163152 90960 /usr/lib/hpux64/dld.so
oracle 20472 oracle mem REG 64,0×7 184592 90962 /usr/lib/hpux64/uld.so
oracle 20472 oracle 0u CHR 3,0×2 0t0 74 /dev/null
oracle 20472 oracle 1u CHR 3,0×2 0t0 74 /dev/null
oracle 20472 oracle 2u CHR 3,0×2 0t0 74 /dev/null
oracle 20472 oracle 3u CHR 3,0×2 0t0 74 /dev/null
oracle 20472 oracle 4u CHR 3,0×2 0t0 74 /dev/null
oracle 20472 oracle 5u CHR 3,0×2 0t0 74 /dev/null
oracle 20472 oracle 6u CHR 3,0×2 0t0 74 /dev/null
oracle 20472 oracle 7u IPv4 0xe0000001cbfe0900 0t0 TCP *:* (IDLE)
oracle 20472 oracle 8u REG 64,0×4 849408 13244 /home/oracle/product/10g/rdbms/mesg/oraus.msb
oracle 20472 oracle 14u REG 64,0×4 268644352 79279 /home/oracle/product/10g/cdata/localhost/local.ocr
oracle 20472 oracle 27u IPv4 0xe0000001c34f0740 0t3603 TCP ***.***.***.***:1521->name.ofthehost.com:4934 (ESTABLISHED)
oracle 20472 oracle 28w FIFO 0xe00000017b3ba748 0t0 184090157
oracle 20472 oracle 29r FIFO 0xe0000001f99d3e48 0t0 184090158

At the end of the output, I saw an IP address and the host name where the process were connecting. After checking that server, I saw that it was opening a connection to the database and created a process on OS. Then it wasn’t able to create a session and didn’t detect the dead process and kill it. So it remained in the V$PROCESS view and on the OS as a process and was missing in the V$SESSION view.

I killed all missing sessions from OS level and rebooted the server which was opening connections to the database. It seems it has memory or network issues, so I asked sysadmins to check it

You can also check the following metalink note and learn how you can enable Dead Connection Detection (DCD)

A discussion of Dead Connection Detection, Resource Limits, V$SESSION, V$PROCESS and OS processes [ID 601605.1]

If you’ve installed Oracle RAC (Real Application Clusters) and want to test how Load Balancing works, you can run the following shell script and check GV$SESSION view:

#!/bin/bash
. /home/oracle/.bash_profile
for ((i=1; i <= 50 ; i++))
do
nohup sqlplus -S system/oracle@racdb<<eof &
begin
dbms_lock.sleep(10);
end;
/

eof
done

This will open 50 sessions in the background. Check GV$SESSION view before and after  running this query:

SQL> select inst_id,count(*) from gv$session where username is not null group by inst_id;

INST_ID COUNT(*)
———- ———-
1           10
2          9

Run the following command from the different session:

[oracle@node1 ~] ./check_load_balancing.sh

SQL> /

INST_ID COUNT(*)
———- ———-
1           33
2           36

Wait for 10 seconds (as we’ve defined “10” seconds at DBMS_LOCK.SLEEP procedure) and run the query again :

SQL> /

INST_ID COUNT(*)
———- ———-
1           10
2           9

SQL>

From the output you can see that all new sessions were distributed between two nodes appropriately.