On the topic of corruption when mixing USB and SATA: Assume all hardware and firmware is working exactly correctly. No cable problems, no firmware bugs, the performance characteristics of the drives are identical and are in no way altered between SATA and USB, you still have a rather difficult problem. The file system writes are being broken up among member drives, guaranteeing they happen at different times and likely out of order compared to file system expectation. So it's really not for sure always crash safe. And the window of crash safeness meanders depending on the workload, whether these are new writes or overwrites, metadata heavy or data heavy. *shrug* It is totally unpredictable whether a crash is recoverable which is why everyone with experience on this list consistently tell people to "slow down, gather data, ask questions, don't make changes haphazardly and in a panic" etc. The whole setup must be assumed to be in a fragile state where any change will make things worse until proven otherwise. And then a big part of any troubleshooting is figuring out if there are in fact hardware or firmware bugs related to the problem. And the challenge for this list has always been that the more variables you've got in your setup, the harder to predict behavior, and harder to troubleshoot if something goes wrong because it adds complexity. But if it's all working correctly, is there something in the block, SCSI, SATA, or md layer that's fussy about any one member drive in an array having really slow commit times? I'm not sure. A bit off topic, but a data point, I have one 5400RPM and one 7200RPM USB HDDs in a Btrfs raid1. Each drive is LUKS formatted first, and the resulting dm-crypt devices are each formatted Btrfs with raid1 profile. For sure the writes complete faster with the 7200RPM drive. And when I do a scrub or balance, it's hundreds of GiB of data so the differential means the slow drive isn't done for more than an hour after the other. I've never had a problem. No errors. No corruption. So off hand I don't think there is an inherent problem with differential performance at the block, SCSI or SATA or USB layers. I'm not sure about md or all device mapper layers. Chris Murphy -- To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
