pass zram to agents as well

[lennart]

relates to #1326

[rennokki]

@mysticaltech Not sure if this would fix this too:

control_plane_nodepools = [
  {
    name        = "cp-cax11"
    server_type = "cax11"
    location    = "fsn1"
    labels      = []
    taints      = []
    count       = 3
    swap_size   = "4G"
    zram_size   = "4G"
  },
]

agent_nodepools = [
  {
    name        = "arm-x2"
    server_type = "cax21"
    location    = "fsn1"
    labels      = []
    taints      = []
    count       = 1
    swap_size   = "4G"
    zram_size   = "4G"
  },
  {
    name        = "amd64-x2"
    server_type = "cpx21"
    location    = "fsn1"
    labels      = []
    taints      = []
    count       = 1
    swap_size   = "4G"
    zram_size   = "4G"
  },
]

[mysticaltech]

@lennart Thanks for this, appreciate it!

[mysticaltech]

@rennokki Here are possible explanations, but please next time, that kind of things are best in a dedicated issue.

The memory usage graph you've provided shows that you are significantly exceeding your available memory resources. Let's analyze the situation based on the nodepool configuration you provided:

Nodepools Configuration:

• Control Plane Nodepools (cp-cax11):

  • RAM per Node: 8 GB
  • Node Count: 3
  • Total RAM: 24 GB

• Agent Nodepools:

  • arm-x2 (cax21):
    • RAM per Node: 16 GB
    • Node Count: 1
    • Total RAM: 16 GB
  • amd64-x2 (cpx21):
    • RAM per Node: 8 GB
    • Node Count: 1
    • Total RAM: 8 GB

Total Cluster Memory:

• Total Available RAM: ( 24 , \text{GB} , (\text{cp-cax11}) + 16 , \text{GB} , (\text{cax21}) + 8 , \text{GB} , (\text{cpx21}) ) = 48 GB

Memory Usage Graph Analysis:

• Total Reserved Memory: 0.23 GB out of 10 GB (2.3%)
• Total Used Memory: 15 GB out of 11 GB (136.36%)

The discrepancy here is that your memory usage exceeds the capacity your graphs are set to display (10 GB), which is inconsistent with the actual memory capacity from your nodepools (48 GB). This discrepancy could be due to several reasons:

1. Misconfiguration in Reporting: The monitoring tool or dashboard might not be configured correctly to represent the total memory available across all nodepools. It's possible it's only tracking one subset or an incorrect aggregation.
2. Overcommitment of Memory: Kubernetes and many cloud-native systems allow overcommitting resources, where you allocate more virtual resources than physically available, relying on the fact that not all resources are used at the same time.
3. Possible Inclusion of Swap/ZRAM: You mentioned swap and zram sizes, both set at 4 GB across your nodes, but it's not clear if these are being counted in your total memory calculations or usage in the dashboard.

Recommendations:

• Verify Dashboard Configuration: Ensure that the dashboard aggregates all memory from all nodepools.
• Check Node Resource Allocation: Confirm that all nodes are online and correctly reporting their resources. Node misconfigurations or failures could lead to incorrect total memory calculations.
• Monitor and Optimize Usage: If your usage often exceeds physical limits, consider optimizing your applications for memory usage or adding more nodes to distribute the load.