arla-related hangs/pauses

[Nickolai Zeldovich]

On 09 Oct 2002, Love <lha at stacken.kth.se> wrote:

> If you do a ps axlwww, that do the process sleep on (xfs or xfs_lock) ?

I'd been unable to get this before, since when I observe the hang,
I can't open another shell until it's too late.  I'm keeping an extra
shell open now to catch it, but it seems like Google has stopped crawling
my AFS space temporarily (I've not seen the problem in a few days now).

> How many fcache-nodes are you using ?

fs getca says:

  Arla is using 65234 of the cache's available 307200 1K byte blocks
  (and 15018 of the cache's available 20000 vnodes)

The fcache-used count always stays at 15000 or a little above, and
always decays towards 15000:

  orbit~219> ls -l $large_dir > /dev/null ; fs getca ; sleep 10 ; fs getca
  Arla is using 64271 of the cache's available 307200 1K byte blocks
  (and 15120 of the cache's available 20000 vnodes)
  Arla is using 64123 of the cache's available 307200 1K byte blocks
  (and 15049 of the cache's available 20000 vnodes)

-- kolya