Hi, Thanks for the reply :-) I actually uses write-back. Here are the tables : ----8<---- cache_cacher: 0 1024 linear 254:32 2097152 cache_origin: 0 2048 linear 8:0 0 cache: 0 2048 cache 253:3 253:4 253:2 512 1 writeback default 4 sequential_threshold 2147483648 random_threshold 4 cache_metadata: 0 2097152 linear 254:32 0 ---->8---- 8:0 is /dev/sda, the iSCSI target. 254:32 is /dev/vdc, a virtio volume accessing a LV on a SSD. The thing is : the cache status properlay state that new blocks in cache are filled, they are even marked as dirty. Which suggest they are not in the backing device. But actually, it blocks when iSCSI detects the network failure. Can dm-cache handle writes without querying the backing device until it needs demotion ? Regards, -- Étienne BERSAC <etienne.bersac@xxxxxxxxxx> NON STOP Systems
Attachment:
signature.asc
Description: This is a digitally signed message part
-- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
