ADRCI and listener purging

I won’t go into any details on ADRCI as others have already done a better job than I could. The key piece of information for this post is that the MMON process shoulders the responsibility for ADR purging. So what if there isn’t an MMON process?

Note: All ADRCI “show control” output on this page has been edited for readability

Below is the ADR purge details for my local listener. These settings were defined using “set control (SHORTP_POLICY = 168,LONGP_POLICY = 336)”. “168” = 7 days and “336” = 14 days.

adrci
set home diag/tnslsnr/n01/listener
show control

ADRID      SHORTP_POLICY LONGP_POLICY LAST_MOD_TIME              LAST_AUTOPRG_TIME LAST_MANUPRG_TIME
---------- ------------- ------------ -------------------------- ----------------- ----------------------------
1794992421 168           336          2012-01-25 20:41:09.052526

adrci> host
$ date

Thu Feb 16 21:26:16 GMT 2012

$ cd  /u01/app/oracle/diag/tnslsnr/n01/listener/alert
$ ls -ltr

total 30260
-rw-r----- 1 grid oinstall 10485934 Jan 25 03:45 log_1.xml
-rw-r----- 1 grid oinstall 10485779 Feb  4 03:11 log_2.xml
-rw-r----- 1 grid oinstall  9954308 Feb 16 21:43 log.xml

Notice that the “LAST_AUTOPRG_TIME” column is blank. It looks like this system is not being purged automatically. Also notice that the date of this test was Feb 16th and the oldest file in the “alert” directory is dated “Jan 25”. According to M.O.S note “Which Files Are Part Of SHORTP_POLICY And LONGP_POLICY In ADR? [ID 975448.1]” the “alert” directory is under LONGP_POLICY which is 14 days so the file “log_1.xml” should have been purged.

Below I force a manual purge

adrci> purge
adrci> host

$ cd  /u01/app/oracle/diag/tnslsnr/n01/listener/alert
$ ls -ltr

total 20004
-rw-r----- 1 grid oinstall 10485779 Feb  4 03:11 log_2.xml
-rw-r----- 1 grid oinstall  9960304 Feb 16 21:53 log.xml

It has now gone. And LAST_MANUPRG_TIME has been set to record the fact.

ADRID      SHORTP_POLICY LONGP_POLICY LAST_MOD_TIME              LAST_AUTOPRG_TIME LAST_MANUPRG_TIME
---------- ------------- ------------ -------------------------- ----------------- --------------------------
1794992421 168           336          2012-01-25 20:41:09.052526                   2012-02-16 21:53:51.761639

So my recommendation is that alongside ADR’s auto purging and your own maintenance on the alert.log & listener.log files in the “trace” directories you should also routinely call something like below:

for i in `adrci exec="show homes"|grep listener`;do
 echo "adrci exec=\"set home $i;purge\""
 adrci exec="set home $i;purge";
 # maybe check $? here
done

I noticed this M.O.S note recently too. I’ve never seen the symptoms but it’s worth taking a look.

ADR Auto Purge Does Not Work On All RAC Instances, ASM and Standby Databases [ID 1326983.1]

Controlling The SCAN Listener Log Location

I was lucky enough to attend UKOUG Conference 2011 last year. I saw several great presentations, one of which was a presentation on troubleshooting RAC by Julian Dyke. During this presentation Julian commented that we need to watch out for SCAN listeners having their ADR home inside the Grid Infrastructure home instead of with the Database listeners in the ADR home under $ORACLE_BASE.

Well this was one of the few things in his presentation I was fit to comment on so I tried to attract Julian’s attention by waving my arm, sadly my timid nature meant I went unnoticed. So this blog post is me going public with my attempted interruption.

First an example of how the log location for SCAN listeners typically differs to that of Database listeners. I personally do not like this behaviour as I like my Grid Infrastructure home to have resonably stable size, which it can have once the log directory’s rotation is mature, however the ADR home for listeners is not cleared down automatically so you can end up with a directory growing unchecked.

[grid@n01 n01]$ $GRID_HOME/bin/lsnrctl status LISTENER|grep Log
Listener Log File         /u01/app/oracle/diag/tnslsnr/n01/listener/alert/log.xml
[grid@n01 n01]$ $GRID_HOME/bin/lsnrctl status LISTENER_SCAN3|grep Log
Listener Log File         /u01/app/grid/11203/ghome_1/log/diag/tnslsnr/n01/listener_scan3/alert/log.xml

And we know from the Oracle documentation that ADR_BASE is defined as:

Purpose
To specify the base directory into which tracing and logging incidents are stored when ADR is enabled.

Default
The default on the server side is ORACLE_BASE, or ORACLE_HOME/log, if ORACLE_BASE is not defined.

So… let’s check ORACLE_BASE for CRS resources:

[grid@n01 ~]$ $GRID_HOME/bin/crsctl stat res ora.LISTENER.lsnr -p|grep -i base
USR_ORA_ENV=ORACLE_BASE=/u01/app/oracle
[grid@n01 ~]$ $GRID_HOME/bin/crsctl stat res ora.LISTENER_SCAN3.lsnr -p|grep -i base
[grid@n01 ~]$

Notice above that the Database listener has $ORACLE_BASE set and the Scan listener does not. Hence the log location.

This can be controlled for Database listeners by using the SETENV switch for SRVCTL but this does not work for Scan listeners.

[grid@n01 ~]$ $GRID_HOME/bin/srvctl setenv listener -l listener_scan1 -t "ORACLE_BASE=/u01/app/oracle"
PRKO-3162 : The actions setenv, getenv, and unsetenv are not suppported for Single Client Access Name listener listener_scan1
PRCN-2066 : Failed to retrieve cluster listener for listener_scan1 because it has the type of Single Client Access Name Listener

There is another way of setting environment variables for cluster resources and that is via crsctl modify resource.

[grid@n01 ~]$ $GRID_HOME/bin/crsctl modify resource ora.LISTENER_SCAN3.lsnr -attr USR_ORA_ENV=ORACLE_BASE=/u01/app/oracle
[grid@n01 ~]$ $GRID_HOME/bin/crsctl stat res ora.LISTENER_SCAN3.lsnr -p|grep -i base
USR_ORA_ENV=ORACLE_BASE=/u01/app/oracle
[grid@n01 ~]$ $GRID_HOME/bin/srvctl stop scan_listener -i 3
[grid@n01 ~]$ $GRID_HOME/bin/srvctl start scan_listener -i 3
[grid@n01 ~]$ lsnrctl status  listener_scan3 | grep Log
Listener Log File         /u01/app/oracle/diag/tnslsnr/n01/listener_scan3/alert/log.xml

To my mind the whole thing would seem a little cleaner if there was an “SRVCTL SETENV SCAN_LISTENER” command.

After writing this post I thought I’d check around for other information to support this post and, lo and behold, there is a My Oracle Support article covering almost exactly what I’ve written above. The article also talks about using the LISTENER.ORA attribute ADR_BASE_LISTENER_SCANn to control this.

Diagnostic Destination For Listeners Is Under Grid_home/Log Instead Of Oracle_base/Diag [ID 1269109.1]