System swap exaggerated.
- Thread starter frankz
- Start date
Thanks Fabian, I knew linux doesn't use swap memory only when the total memory is full. Maybe I was wrong to set the question, that is I was wondering how mail is the only node of the three that uses this swap, since the others are configured like this. Thanks comunqiue for your reply
Hi frankz.
Thank you for asking the question, because I have exactly the same behavior.
Hi frankz.
Thank you for asking the question, because I have exactly the same behavior.
Hi I wanted to confirm that suspending the LXC machine in another node the problem is the same. Furthermore, Zabbix, which I use to monitor the nodes as well, signals me a low-space alert, due to the swap consumption. I think this needs to be reviewed for good, as the backup of the LXC VM also fails.Thank you for asking the question, because I have exactly the same behavior.
Not sure if this is the case here, but: when one Linux system does more I/O than another Linux system, it more often makes a choice between not using additional memory for cache and swapping out memory than has not been used for some time. Swap is only reduced when the memory on disk is actually requested by software and moved back into RAM. Therefore, swap keeps growing when the swapped-out memory is not actively used. Maybe that's the difference between your two systems.
If a system is constantly reading from swap (and writing other parts of memory to swap), I would consider it a problem because the additional disk I/O when using memory will slow down the system a lot. However, if swap is filled up over time and almost never ever read/reduced, I would not call this a problem. Apparently some software allocated memory that is hardly ever actually used and it is a good thing that it is not taking up high-speed RAM.
As I said, I'm not sure this it the case for your particular setup, but a monitor that complains about swap being full, while there is no actual (heavy) swap I/O going on, does not understand how Linux uses swap. [EDIT: Some typos an a bit of clarification.]
If a system is constantly reading from swap (and writing other parts of memory to swap), I would consider it a problem because the additional disk I/O when using memory will slow down the system a lot. However, if swap is filled up over time and almost never ever read/reduced, I would not call this a problem. Apparently some software allocated memory that is hardly ever actually used and it is a good thing that it is not taking up high-speed RAM.
As I said, I'm not sure this it the case for your particular setup, but a monitor that complains about swap being full, while there is no actual (heavy) swap I/O going on, does not understand how Linux uses swap. [EDIT: Some typos an a bit of clarification.]
Last edited:
First of all thanks for posting. Considering what was written above, I purposely wanted to move the VM (LXC) to another node and the swap of the nod increases in volume. Technically I assume from the parameters of the VM that the swap declared in the properties is not virtualized but shared with the node, so if it were, it is clear that the node itself is involved in the available space. assuming you have several lxcs that make use of a minimum of 1024 Gb of memory and 2048 Gb of swap, the node would be saturated. Not true for all VM machines, but I use zoneminder in LXC and it eats up a lot of memory .... other lxc like centos 8, ubuntu 21 or other are moderate.
All containers share the same swap space with each other and the Proxmox host. Just as they all share the same physical memory. Please note that setting Memory (MiB) to M and Swap (MiB) to S for a container limits the use of the swap space to (S+M) MiB. Of course, it can be less if the swap has less space available (because of the shared nature). This also means that setting the Swap (MiB) to 0 can still result in the use of swap space because the limit is then M MiB. In your case, CT110 can use up to 3.50 GiB of swap space and also 1.50 GiB of actual memory.
To clearly answer you original questions in post #1: Writing and filing the shared swap space is usually a good thing. That's probably why you see a lot of swap usage but everything seems fine. It's reading memory back from swap space (all the time while also writing) that tends to cause slow downs, as I tried to explain in post #12.
To clearly answer you original questions in post #1: Writing and filing the shared swap space is usually a good thing. That's probably why you see a lot of swap usage but everything seems fine. It's reading memory back from swap space (all the time while also writing) that tends to cause slow downs, as I tried to explain in post #12.