Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
simon написал: > I used drbd 8.3.7 on HA. When Master host is dead and HA swatches from > Master to Slave, the drbd can't switch because it spends 10 minutes to mount > its partition. But the time is timeout to HA.(in HA, default overtime is 2 > miniutes). > > > > Why does drbd spent that long time? For me it was new problem this summer. Even while I temporary replace broken onboard e1001e by semi-compatible d-link/VIA card (with slow PCI transfer and lost packets) I have no this error, but SUDDENLY it happened this sammer with same configs, starting kernel 3.5.0. IMHO new kernels (including tcp stack and network drivers) and drbd control some things like multitasking/preemption with different ways and need more to care. First, check hardware health. Then - all (not directly this) network cards on same switch to PAUSE frame support (using ethtool, make Y if not sure - some other cases like PSPacer is other subject). Then check kernel config and PREEMPT states and related things. For example, now I virtualize semi-dead server and inside of qemu-kvm - CONFIG_PARAVIRT_TIME_ACCOUNTING=y (with PREEMPT and TREE_PREEMPT_RCU). -- WBR, Dzianis Kahanovich AKA Denis Kaganovich, http://mahatma.bspu.unibel.by/