[Lustre-discuss] new errors w adaptive timeouts?

[Thomas Roth]

Hi all,

since the upgrade to version 1.6.7._patched, our MDT prints huge amounts of:

00000100:00000400:4:1240598372.080847:0:4842:0:(service.c:753:ptlrpc_at_send_early_reply())
@@@ Couldn't add any time (42/30), not sending early reply
  req at ffff8107faf6f450 x18776/t0
o400->ddd8a2c9-047b-b157-61ae-383b164adaf6 at NET_0x200008cb57238_UUID:0/0
lens 128/0 e 0 to 0 dl 1240598414 ref 2 fl New:H/0/0 rc 0/0

At least it seems that these two messages appear together always. Of
course I'm completely clueless and would ask the experts whether this is
something dangerous or remarkable.

The problem (is it?) of "not sending early reply" did not show up in the
logs in version 1.6.5.1, although I had  activated adaptive timeouts
then also.

We have set:
at_max = 6000
at_history = 6000
at_extra = 30
at_early_margin = 50

(The enormous numbers are of course an attempt to allow the MDT to
prolong its patience to whatever our shaky network requires.)

Regards,
Thomas