37

I'm currently trying to work out a method of tidying up Oracle Recover log files that are created by Cron...

Currently, our Oracle standby recover process is invoked by Cron every 15mins using the following command:

0,15,30,45 * * * * /data/tier2/scripts/recover_standby.sh SID >> /data/tier2/scripts/logs/recover_standby_SID_`date +\%d\%m\%y`.log 2>&1

This creates files that look like:

$ ls -l /data/tier2/scripts/logs/
total 0
-rw-r--r-- 1 oracle oinstall 0 Feb  1 23:45 recover_standby_SID_010213.log
-rw-r--r-- 1 oracle oinstall 0 Feb  2 23:45 recover_standby_SID_020213.log
-rw-r--r-- 1 oracle oinstall 0 Feb  3 23:45 recover_standby_SID_030213.log
-rw-r--r-- 1 oracle oinstall 0 Feb  4 23:45 recover_standby_SID_040213.log
-rw-r--r-- 1 oracle oinstall 0 Feb  5 23:45 recover_standby_SID_050213.log
-rw-r--r-- 1 oracle oinstall 0 Feb  6 23:45 recover_standby_SID_060213.log
-rw-r--r-- 1 oracle oinstall 0 Feb  7 23:45 recover_standby_SID_070213.log
-rw-r--r-- 1 oracle oinstall 0 Feb  8 23:45 recover_standby_SID_080213.log
-rw-r--r-- 1 oracle oinstall 0 Feb  9 23:45 recover_standby_SID_090213.log
-rw-r--r-- 1 oracle oinstall 0 Feb 10 23:45 recover_standby_SID_100213.log
-rw-r--r-- 1 oracle oinstall 0 Feb 11 23:45 recover_standby_SID_110213.log
-rw-r--r-- 1 oracle oinstall 0 Feb 12 23:45 recover_standby_SID_120213.log

I basically want to delete off files older than x days old, which I thought logrotate would be perfect for...

I've configured logrotate with the following config file:

/data/tier2/scripts/logs/recover_standby_*.log {
    daily
    dateext
    dateformat %d%m%Y
    maxage 7
    missingok
}

Is there something I'm missing to get the desired outcome?

I guess I could remove the date from the Crontab log file, and then have logrotate rotate that file, however then the date in the log file wont reflect the day the logs were generated... i.e. Recoveries on 010313 would be in file with a date of 020313 due to logrotate firing on 020313 and rotating the file...

Any other ideas? And thank-you in advance for any responses.

Regards

Gavin

fatmcgav
  • 431
  • 1
  • 4
  • 8
  • You normally have to restart the service in order to have logrotate working. Did you try to? – fedorqui Feb 13 '13 at 17:02
  • 2
    I've been trying it by running logrotate manually.. Haven't done a logrotate service restart... However will give that a go... Edit: Actually, logrotate on RHEL is running through Cron... So shouldn't need a restart to pick up any changes... – fatmcgav Feb 13 '13 at 17:18
  • 4
    Don't use that date naming format. Always put year, then month, then day. If you see someone else doing this, correct them. Thank you. – toddkaufmann Dec 14 '17 at 22:42

6 Answers6

46

Logrotate removes files according to order in lexically sorted list of rotated log file names, and also by file age (using last modification time of the file)

  • rotate is maximal number of rotated files, you may find. If there is higher number of rotated log files, their names are lexically sorted and the lexically smallest ones are removed.

  • maxage defines another criteria for removing rotated log files. Any rotated log file, being older than given number of days is removed. Note, that the date is detected from the file last modification time, not from file name.

  • dateformat allows specific formatting for date in rotated files. Man page notes, that the format shall result in lexically correct sorting.

  • dateyesterday allows using dates in log file names one day back.

To keep given number of days in daily rotated files (e.g. 7), you must set rotate to value of 7 and you may ignore maxage, if your files are created and rotated really every day.

If log creation does not happen for couple of days, a.g. for 14 days, number of rotated log files will be still the same (7).

maxage will improve the situation in "logs not produced" scenarios by always removing too old files. After 7 days of no log production there will be no rotated log files present.

You cannot use dateformat as OP shows, as it is not lexically sortable. Messing up with dateformat would probably result in removing other rotated log files than you really wanted.

Tip: Run logrotate from command line with -d option to perform a dry run: you will see what logrotate would do but doesn't actually do anything. Then perform a manual run using -v (verbose) so that you can confirm that what is done is what you want.

Solution: clean logs created by cron

The concept is:

Let cron to create and update the log files, but make small modification to create files, following logrotate standard file names when using default dateext

/data/tier2/scripts/logs/recover_standby_SID.log-`date +\%Y\%m\%d`.log

Use logrotate only for removing too old log files

  • aim at not existing log file /data/tier2/scripts/logs/recover_standby_SID.log
  • use missingok to let logrotate cleanup to happen
  • set rotate high enough, to cover number of log files to keep (at least 7, if there will be one "rotated" log file a day, but you can safely set it very high like 9999)
  • set maxage to 7. This will remove files which have last modification time higher than 7 days.
  • dateext is used just to ensure, logrotate searches for older files looking like rotated.

Logrotate configuration file would look like:

data/tier2/scripts/logs/recover_standby_SID.log {
    daily
    missingok
    rotate 9999
    maxage 7
    dateext
}

Solution: rotate directly by logrotate once a day

I am not sure, how is source recovery standby file created, but I will assume, Oracle or some script of yours is regularly or continually appending to a file /data/tier2/scripts/logs/recover_standby_SID.log

The concept is:

  • rotate the file once a day by logrotate
  • working directly with log file containing recovery data /data/tier2/scripts/logs/recover_standby_SID.log
  • daily will cause rotation once a day (in terms of how cron understands daily)
  • rotate must be set to 7 (or any higher number).
  • maxage set to 7 (days)
  • dateext to use default logrotate date suffix
  • dateyesterday used to cause date suffixes in rotated files being one day back.
  • missingok to clean older files even when no new content to rotate is present.

Logrotate config would look like:

data/tier2/scripts/logs/recover_standby_SID.log {
    daily
    missingok
    rotate 7
    maxage 7
    dateext
    dateyesterday
}

Note, that you may need to play a bit with copytruncate and other similar options which are related to how is the source log file created by external process and how it reacts to the act of rotation.

Thomas BDX
  • 2,632
  • 2
  • 27
  • 31
Jan Vlcinsky
  • 42,725
  • 12
  • 101
  • 98
  • 5
    This is from far the best answer. Thank you Jan – Dionys May 03 '16 at 08:34
  • 1
    The first solution doesn't work for me. It says "log /var/log/tomcat8/catalina.log does not exist -- skipping" and doesn't perform check for old files. Creating a dummy catalina.log file helps, but starts to rotate the dummy file and fails if external program has already created the log for new day: "destination ... already exists, skipping rotation" – Yoory N. Dec 01 '17 at 10:48
31

You can use find command to do that task easily! It will delete all 7 Days old files. Put it in crontab and run nightly basis:

$ cd /data/tier2/scripts/logs/    
$ /usr/bin/find . -mtime +7 -name "*.log" -print -delete

Or Better way

$ /usr/bin/find /data/tier2/scripts/logs/ -mtime +7 -name "*.log" -print -delete;
Jeremy L
  • 7,686
  • 4
  • 29
  • 36
Satish
  • 16,544
  • 29
  • 93
  • 149
  • 15
    Do not use rm! For error once I did not write the correct folder and BOOM all system was emptied! It is bettere to use logrotate – gdm Jul 11 '14 at 16:15
  • 5
    `find ... -exec /bin/rm {} \;` === `find ... -delete`. Easier to read, faster (rm is fork/exec per file) and less chance to mess up. – temoto Mar 15 '16 at 08:26
  • 1
    @giuseppe Only `logrotate` doesn't delete files that are not properly sorted or are unique. The best I could do is get the files truncated. – Alexis Wilke Aug 22 '18 at 05:09
6

(Updated) Your options are:

  • As Satish answered, abandon logrotate and put a find script in cron
  • You could even, use logrotate and put a find script in the postrotate command

Initially, I thought that changing the dateformat to match your logs might work but as Reid Nabinger pointed out the date format was not compatible with logrotate anyway. Recently, I tried to configure the same thing but for Java rotated logs that I wanted logrotate to delete. I tried the configuration below but it kept trying to delete all logs

/opt/jboss/log/server.log.* {
    missingok
    rotate 0
    daily
    maxage 30
}

I ended up just implementing what Satish suggested - a simple find with rm script in cron.

Alexis Wilke
  • 19,179
  • 10
  • 84
  • 156
Lars Nordin
  • 2,785
  • 1
  • 22
  • 25
  • Regarding the initial question (delete timestamped log files), this is the only correct answer. I needed it for RADIUS log files, which are already timestamped. – mgutt Dec 09 '21 at 08:05
3

(Unable to comment as not enough reputation)

I had a similar issue. By all accounts logrotate is useless for use on filenames with built in datestamps.

If all else was equal I would probably go with find in a cron job.

For my own reasons I wanted to use logrotate and eventually found a way: https://stackoverflow.com/a/23108631

In essence it was a way of encapsulating the cron job in a logrotate file. Maybe not the prettiest or most efficient but like I said, I had my reasons.

Community
  • 1
  • 1
northern-bradley
  • 907
  • 8
  • 11
3

As per @Jan Vlcinsky, you can let logrotate add the date - just use dateyesterday to get the right date.

Or, if you want to put in the date yourself, you can 'aim' at the name without the date , and then the names with the date will be cleaned up.

However, what I found is that if I don't have a log file there, logrotate doesn't do the cleanup of the files with dates.

But if you're prepared to have an empty log file lying around, then it can be made to work.

For example, to clean up /var/log/mylogfile.yyyymmdd.log after 7 days, touch /var/log/mylogfile.log, then configure logrotate as follows:

/var/log/mylogfile.log
{
        daily
        rotate 7
        maxage 7
        dateext
        dateformat .%Y%m%d
        extension .log
        ifempty
        create
}

This entry, combined with the existence of mylogfile.log, triggers logrotate to clean up old files, as if they had been created by logrotate.

daily, rotate plus maxage cause old log files to be deleted after 7 days (or 7 old log files, whichever comes first).

dateext, dateformat plus extension causes logrotate to match our filesnames.

And ifempty plus create ensure that there continues to be an empty file there, or the log rotation would stop.

Another tip for testing, be prepared to edit /var/lib/logrotate.status to reset the 'last rotated' date or logrotate won't do anything for you.

Ben Aveling
  • 842
  • 7
  • 13
  • Instead of removing logrotate.status, you may run better using the --force commandline option - in combination with --debug you can check what logrotate would do if rotation would be triggered. – frank42 Aug 29 '22 at 07:54
1

FYI I know that this is an old question, but the reason why it does not work for you is because your dateformat is not lexically sortable. From the manpage:

   dateformat format_string
          Specify the extension for dateext using the notation similar to strftime(3)  function.  Only
          %Y  %m  %d  and %s specifiers are allowed.  The default value is -%Y%m%d. Note that also the
          character separating log name from the extension is part of the dateformat string. The  sys-
          tem  clock must be set past Sep 9th 2001 for %s to work correctly.  Note that the datestamps
          generated by this format must be lexically sortable (i.e., first the year,  then  the  month
          then  the  day.  e.g.,  2001/12/01 is ok, but 01/12/2001 is not, since 01/11/2002 would sort
          lower while it is later).  This is because when using the rotate option, logrotate sorts all
          rotated filenames to find out which logfiles are older and should be removed.

The solution is to either change to one that goes year-month-date, or call an external process to perform the cleanup.