Skip to Content
avatar image
Former Member

Hana inactive/cached memory not reclaimed nor usable

Dear hana experts,

I am facing a situation where I'm having trouble to understand linux/hana memory behavior.

Last week, we faced a bunch of OOM dumps on our 1TB hana prod system (1.00.122), but when checking hana resident memory, it was showing around 250GB of usage. As nothing else than hana is running on this server, the occurences of OOM dumps were a bit suprising...

So I check this from the OS :

- free command shows 1TB of used memory and 5GB of free memory.

- vmstat shows 750GB of free memory and 700GB of inactive memory

- top command shows hdbindexserver is using 250GB of memory.

- cat /proc/meminfo shows :

MemTotal: 1056503664 kB

MemFree: 5632228 kB

MemAvailable: 743517648 kB

Cached: 746952804 kB

So it seems the memory is stuck in the "cached" area, which count as inactive memory. And here's the confusing part, although 700GB of memory are inactive it's displayed as available memory by vmstat and /proc/meminfo. So how comes that hana cannot grab this memory? Even the hana studio is reporting 750GB of free memory...

From what I've red about linux memory management, the inactive memory should be granted if there's no more free memory available. But it was not happening in my case.

The blog https://blogs.sap.com/2016/09/01/how-to-analyze-and-retain-unused-memory-from-the-hanabodsbw-servers/ was a very similar case to mine so I applied the solution of forcing a drop cache and it worked. But it still seems like an abnormal behavior to me, and I'm not sure if this is a safe way to reclaim the memory.

Any idea on how to explain this behavior and how can I prevent that from happening in the future?

Thanks in advance!

Add comment
10|10000 characters needed characters exceeded

  • Get RSS Feed

0 Answers