3

Once a month I've been doing 

echo check > /sys/block/md2/md/sync_action

in order to force a software raid consistency check. The problem is this is a 7.2T RAID 5. The consistency check takes days to complete and while it's running, everything on my system slows to a crawl. Is there any way to nice this (say by temporarily setting /proc/sys/dev/raid/speed_limit_max to something smaller for the duration of the check?

In general it pains me that even in the 4.13 linux kernel, extended disk I/O causes the system to grind to a halt; should be a solved problem by now.

pgoetz
  • 171
  • 1
  • 5
  • 2
    Possible duplicate of [Reducing Linux Software RAID Rebuild Speed?](https://superuser.com/questions/625722/reducing-linux-software-raid-rebuild-speed) – Nattgew Nov 07 '17 at 16:25
  • 1
    Not a duplicate. I read the question you link to before posting. The question you're referring to concerns RAID rebuild times. I'm asking about consistency checks for an (apparently) healthy RAID. – pgoetz Nov 07 '17 at 21:52
  • So you're looking for a way to throttle the check process? You recognize that if this is possible the process will take longer, and even with throttling, there will still be a slowdown. – music2myear Nov 09 '17 at 18:19
  • What I'm really looking for is a way to nice the check process (with full understanding that this will make it take longer). I've never understood why certain kinds of sustained I/O slow the system down so much. Maybe that has something to do with the hardware architecture? How interrupts work, or something like this? – pgoetz Nov 10 '17 at 21:28
  • I have similar issues. Linux softraid should intelligently reduce sanity check speed as system load increases, but this seems not to happen. My server can get load factors in the 40-60 range as tasks get backed up - the backlog clears up in a minute or two but can then reappear a few minutes later, depending on system demands. Manually overriding /proc/sys/dev/raid/speed_limit_max solves the problem, of course, but at the expense of the sanity check being much slower if the setting is left there. – Jim MacKenzie Jun 06 '21 at 16:25

0 Answers0