Hi
very interesting post. Could someone explain two things?
1) Per-system tw_count is greater than per-system max_tw_buckets limit:
tw_count < sysctl_max_tw_buckets
2) Per-VE counter is greater than per-VE max_tw_buckets limit:
tw_count < sysctl_tcp_max_tw_buckets_ub
I didn't understand when Vaverin talk about Per-System tw_count. What is tw_count?
And in the bottom of his post, he speaks about increase net.ipv4.tcp_max_tw_kmem_fraction or increase kmemsize of VE in conf. Really is a mistery for me how to solve the issue of Time Wait Bucket Overflow. Until now when issue appears I increase kmemsize of VE and restart the VE. But never Ive tried to increase kmem_fraction. What can I do in first time, increase kmemsize or kmem_fraction?
In the other hand, vzcfgvalidate show "success" result in conf of VEs with "Time Wait Bucket Overlflow" issue.
Thanks!