thanks md raid
I'm glad my RAID1 worked as expected... just hoping I don't encounter any read timeouts on the non-TLER drive before my rebuild finishes:
# mdadm --manage --add -v /dev/md2 /dev/sdb2 mdadm: added /dev/sdb2
# cat /proc/mdstat
Personalities : [raid1]
md2 : active raid1 sdb2[3] sda2[2]
976510840 blocks super 1.2 [2/1] [_U]
[>....................] recovery = 1.9% (19032256/976510840)
finish=350.1min speed=45575K/secunused devices: <none>
# iostat -k 1 -x /dev/sd[ab]
avg-cpu: %user %nice %system %iowait %steal %idle
1.63 0.00 20.65 0.00 0.00 77.72
Device: rrqm/s wrqm/s r/s w/s rkB/s wkB/s
avgrq-sz avgqu-sz await svctm %util
sda 0.00 0.00 1584.00 0.00 101376.00 0.00
128.00 2.84 1.80 0.39 61.60
sdb 0.00 3.00 0.00 1584.00 0.00 101568.00
128.24 27.72 17.70 0.63 99.60
Quoted 3 lines · Daniel Pocock
I'm glad my RAID1 worked as expected... just hoping I don't encounter any read timeouts on the non-TLER drive before my rebuild finishes:
You have an inverse understanding of ERC. Drives without ERC will retry forever, or until an upper layer puts a stop to its efforts. Drives with a 7 second ERC will return a hard error after 7 seconds.
So the only way you'll get a timeout with your rebuild is if the healthy drive spends 30 seconds retying a sector read.
Signature · 1 line
Stan
Quoted 5 lines · Stan Hoeppner
Quoted 3 lines · Daniel Pocock
I'm glad my RAID1 worked as expected... just hoping I don't encounter any read timeouts on the non-TLER drive before my rebuild finishes:
Quoted 7 lines
You have an inverse understanding of ERC. Drives without ERC will retry forever, or until an upper layer puts a stop to its efforts. Drives with a 7 second ERC will return a hard error after 7 seconds.
So the only way you'll get a timeout with your rebuild is if the healthy drive spends 30 seconds retying a sector read.
I was thinking about the more obscure case - that some other URE followed by an attempt at write access on the good drive fails and it becomes degraded
Quoted 12 lines · Daniel Pocock
Quoted 9 lines · Stan Hoeppner
Quoted 3 lines · Daniel Pocock
I'm glad my RAID1 worked as expected... just hoping I don't encounter any read timeouts on the non-TLER drive before my rebuild finishes:
You have an inverse understanding of ERC. Drives without ERC will retry forever, or until an upper layer puts a stop to its efforts. Drives with a 7 second ERC will return a hard error after 7 seconds.
Quoted 7 lines
Quoted 3 lines
So the only way you'll get a timeout with your rebuild is if the healthy drive spends 30 seconds retying a sector read.
I was thinking about the more obscure case - that some other URE followed by an attempt at write access on the good drive fails and it becomes degraded
If drives were that damn fragile modern computing wouldn't exist. The odds of your UPS taking a dump during a rebuild are greater than the scenario you just described. You need to put more thought into UPS failure scenarios than ERC.
I mention this specifically because my "desktop" APC Backups XS 900 did the unthinkable the other day. Apparently it decided the batteries were bad at the very moment it ran its hard scheduled self test. Class, what happens with all APC UPSes when the scheduled self test runs and the batteries have been flagged bad? Answer: it drops the load and causes your system to reboot.
One would think APC would be smart enough to have the firmware skip the self test until after the batteries have been replaced, specifically to prevent an unplanned power event, the whole purpose of a UPS. I guess one of their actuaries figured they'd get more battery sales if they keep downing your system until you replace them...
Signature · 1 line
Stan