Comments

• 

  11 days ago by zzhou | Reply

  Summary

  At this stage, it's challenging to start the cluster more than 20 nodes. It can be challenging just join one more node either the cold start of the cluster gradually, or grow the nodes online.

  Potential further research direction: probably to tune corosync timeout and knet configure for the bigger cluster size

  Misc.: to improve tool to grow node gradually https://github.com/ClusterLabs/crmsh/pull/1235

  Observations

  Environment:

  • VM, CPU=8, RAM=16G
  • CPU: grow 4 nodes all at once. Half of #CPU of the system
  • MEM: One node container consumes 300M

  Challenges:

  • difficult to setup 20 node cluster all at once

  • difficult to grow 4 nodes all at once on a 32 node cluster (even a 20 node cluster)

    • CPU overloaded
      • DC election can't finish
      • Token timeout
    • corosync membership
      • [TOTEM ] Token has not been received in 41025 ms
      • Repeat: Completed service synchronization, ready to provide service.
    • Network
      • corosync-cfgtool -R might mess KNET layer for a while
        
      • hanode1 kernel: neighbour: arp_cache: neighbor table overflow!
      • [KNET ] loopback: send local failed. error=Resource temporarily unavailable
      • kernel: podman2: port 19(veth38) entered blocking state
    • Mem can be consumed quickly. Memory leak?

×
Edit Comment 9040
# A First Level Header
## A Second Level Header

Use one asterisk to *emphasize*

Use two asterisks for **strong emphasis**

- Use hyphens
- for unordereed
- lists

This is an [link to example.com](http://example.com/)

This is an image ![an openSUSE geeko icon](https://en.opensuse.org/images/d/d0/Icon-distribution.png)

This is a user link @hans

This is a project link hw#some-cool-title
More Complex Markdown Help
Formatting Help
• Edit
• Preview
Summary =================== At this stage, it's challenging to start the cluster more than 20 nodes. It can be challenging just join one more node either the cold start of the cluster gradually, or grow the nodes online. Potential further research direction: probably to tune corosync timeout and knet configure for the bigger cluster size Misc.: to improve tool to grow node gradually https://github.com/ClusterLabs/crmsh/pull/1235 Observations =================== Environment: - VM, CPU=8, RAM=16G - CPU: grow 4 nodes all at once. Half of #CPU of the system - MEM: One node container consumes 300M Challenges: - difficult to setup 20 node cluster all at once - difficult to grow 4 nodes all at once on a 32 node cluster (even a 20 node cluster) - CPU overloaded - DC election can't finish - Token timeout - corosync membership - [TOTEM ] Token has not been received in 41025 ms - Repeat: Completed service synchronization, ready to provide service. - Network - `corosync-cfgtool -R` might mess KNET layer for a while - hanode1 kernel: neighbour: arp_cache: neighbor table overflow! - [KNET ] loopback: send local failed. error=Resource temporarily unavailable - kernel: podman2: port 19(veth38) entered blocking state - Mem can be consumed quickly. Memory leak?
 
×
Reply to zzhou
Summary
At this stage, it's challenging to start the cluster more than 20 nodes. It can be challenging just join one more node either the cold start of the cluster gradually, or grow the nodes online.
Potential further research direction: probably to tune corosync timeout and knet configure for the bigger cluster size
Misc.: to improve tool to grow node gradually https://github.com/ClusterLabs/crmsh/pull/1235
Observations
Environment:
• VM, CPU=8, RAM=16G
• CPU: grow 4 nodes all at once. Half of #CPU of the system
• MEM: One node container consumes 300M
Challenges:
• difficult to setup 20 node cluster all at once

• difficult to grow 4 nodes all at once on a 32 node cluster (even a 20 node cluster)

  • CPU overloaded
    • DC election can't finish
    • Token timeout
  • corosync membership
    • [TOTEM ] Token has not been received in 41025 ms
    • Repeat: Completed service synchronization, ready to provide service.
  • Network
    • corosync-cfgtool -R might mess KNET layer for a while
      
    • hanode1 kernel: neighbour: arp_cache: neighbor table overflow!
    • [KNET ] loopback: send local failed. error=Resource temporarily unavailable
    • kernel: podman2: port 19(veth38) entered blocking state
  • Mem can be consumed quickly. Memory leak?
---
# A First Level Header
## A Second Level Header

Use one asterisk to *emphasize*

Use two asterisks for **strong emphasis**

- Use hyphens
- for unordereed
- lists

This is an [link to example.com](http://example.com/)

This is an image ![an openSUSE geeko icon](https://en.opensuse.org/images/d/d0/Icon-distribution.png)

This is a user link @hans

This is a project link hw#some-cool-title
More Complex Markdown Help
Formatting Help
• Edit
• Preview