Hi,
I started to notice that pve nodes get fenced, when the swap space, provided by zramswap, becomes full. This only happens when the pve node is performing backups of its guests. In such a case, messages like these are to be found in the system log.
A couple of RIP: 0010:_raw_spin_lock events later, the node will lock-up and get fenced by the cluster. I wondered if zramswap could be the cuplrit, but that one seems to be proven and stable code. However, I also stumbled across this, when searching the net for possible reasons: [PATCH -V3] swap: Reduce lock contention on swap cache from swap slots allocation
Anybody else noticed something similar?
Cheers,
budy
I started to notice that pve nodes get fenced, when the swap space, provided by zramswap, becomes full. This only happens when the pve node is performing backups of its guests. In such a case, messages like these are to be found in the system log.
A couple of RIP: 0010:_raw_spin_lock events later, the node will lock-up and get fenced by the cluster. I wondered if zramswap could be the cuplrit, but that one seems to be proven and stable code. However, I also stumbled across this, when searching the net for possible reasons: [PATCH -V3] swap: Reduce lock contention on swap cache from swap slots allocation
Anybody else noticed something similar?
Cheers,
budy
