Home / user-431332

schweik's questions

Martin Hope
schweik
Asked: 2021-07-14 12:14:27 +0800 CST
  • 0

I got 12 hard disks from completely crashed server, some of them are failing, too. I successfully saved bit copy of all of them, even not always error free. I found 512MB superblock on each end of disk starting with a word DSREGION. The 512MB of space is not heavily used and from some text fragments one can judge, that the built-in RAID controller was some LSI model. The question remains: how can I assemble the RAID arrays using mdadm. How can I tell the mdadm where it can read the superblock while the command:

# mdadm -E /dev/loopX

does not detect superblock, even if DDF is set explicitly:

# mdadm -E -e DDF /dev/loopX

If there is a hint how to parse the 512MB superblock I can do the parsing manualy and then I can assemble the RAID arrays explicitly using mdadm -A.

Or if there is a possibility to tell mdadm where it has to look for the superblock metadata.

J.

P.S. I hoped I can find something useful at www.snia.org but did not find any detail.

raid mdadm metadata
Martin Hope
schweik
Asked: 2021-05-27 23:32:19 +0800 CST
  • 4

On mdadm softraid 6 (about 12 disks, 60TB) there appeared accidental misswrites in large files (>100GB). All system was checked - RAM, NIC, LSI RAID card. The most suspected is the LSI because its battery BBU is flat and write back / write through is not correctly set. In theory each chunk of data is backed up with XOR redundancy, RAID5 1x, RAID 6 2x. But this comes in act only if some of active disks fails.

In the mdadm is there a command which could start the complete data consistency check taking in account the XOR backup? It means can I identify the miswritten chunks?

After I discarded the BBU I want to know, which files are good and which files are corrupted and have to be replaced. If there is no way to solve it, I should have to create the array from the scratch and get all the files from their backup.

raid mdadm
Martin Hope
schweik
Asked: 2018-10-19 01:02:36 +0800 CST
  • 0

I use RAID1 - mirror long time. I have beleaved yet, that RAID1 does stripping on read operation for improve the speed.

While migrating form 160GB disk to 3TB mirror, I started the file copy, watching the real speed on mirrored volume as on mirror members. I'm disappointed that all the read concerns the /dev/sdb only while the other member is sleeping. Do somebody know how to force the stripe-read operations on RAID1?

OS Debian Linux 9 Stretch, mdadm, ext3/4

performance
Martin Hope
schweik
Asked: 2017-10-11 03:12:26 +0800 CST
  • 0

While preparing the mini forensic system I hung on a problem: Let us imagine, we prepare a usb-flash booted linux for lot of diferent workstations. While setting up the network (ipv4 dhcp) we can find more than one LAN interfaces, and someone is connected, someone not, but hopefully one of them is connected into the private LAN (eg. 192.168.20.0/24), and only this interface should be set up.

Perhaps it can be solved with some udev rules, but I cannot find where they are and how they could be modified. At present I use my own /etc/rc.local script toghether with some modification of dhclient.conf (refuse network), but even though it runs somehow (better said: it walks) it is not reliable and many times it does not finish succesfully the local cifs mounts, even on the same workstation.

Can somebody show me the way how to solve it?

Details:

  • the private LAN is isolated with one managable switch
  • local dnsmasq server responses exclusively in this private LAN
  • each workstation can have 1 - 4 (wire) NIC
  • some workstations are connected not only into private LAN, but also can have some NIC configured by some other dhcp-server or can have static IP - but those interfaces should be ignored

The goal is: after the boot to get dhcp address and connect to the local network data sorage.

Thank a lot

linux-networking