Unix & Linux Stack Exchange is a question and answer site for users of Linux, FreeBSD and other Un*x-like operating systems. It only takes a minute to sign up.
Sign up to join this community
I have a situation where I may need to execute an SAP command line tool hdbsql when memory gets too low (to help clean out the HANA table cache).
How I can best approach this? I did have an idea to extract the free memory value from the top command (or get free/max*100 to get a %, which is better) in a shell script scheduled using crontab, but I cannot find anything on a possible approach on this anywhere, so I cannot even start anything.
You could use awk to calculate the percentage and the test utillity to determine if the value exceeds 90%, for example. The cronjob could look like this:
/usr/bin/test 90 -le $(/usr/bin/awk '$1=="MemTotal:"{t=$2} $1=="MemFree:"{f=$2} END{printf "%d", (t-f)/(t/100)}' /proc/meminfo) && command-to-cleanup
The awk part extracts the needed values out of /proc/meminfo, and then calculates the percentage of used memory. The test utillity checks if 90 is is less or equal (-le) to the calculated value. The part after the && would then be your tool that cleans up the memory (.. && command-to-cleanup). That cronjob can run every minute, for example:
* * * * * root /usr/bin/test 90 -le $(/usr/bin/awk '$1=="MemTotal:"{t=$2} $1=="MemFree:"{f=$2} END{printf "%d", (t-f)/(t/100)}' /proc/meminfo) && command-to-cleanup
IMHO the percentage of the free memory on a linux system is NOT a good indicator in this situation because of the way linux memory management works (memory is also used for buffering and caching, there's also swapping, etc) - plenty of info out there on the subject. Which means using such indicator would cause cleaning the cache earlier than actually necessary which I imagine comes with a price in overall performance.
I found that the percentage of swap used is usually a much better indication of the system being short on memory while taking all these into account: as long as it can still allocate memory it tipically won't use the swap.
As a bonus this doubles as a performance indicator as well - a system using over 4G of swap for example is usually barely crawling (think how much it takes to copy a 4G disk file from one place to another).
The swap info be found in /proc/meminfo, so the approach that chaos suggested can be adapted accordingly:
SwapTotal: 16777212 kB
SwapFree: 16777212 kB
You may need to experiment a bit to find out which swap usage percentage value is better fit for your particular case.
