On Wed, 11 Jun 2008, Raghavendra G wrote:
Hi Brent,
glusterfs--mainline--3.0--patch-188 should fix server process consuming lots
of memory with io-threads loaded. Can you please confirm?
Indeed it did. However, when doing multiple 10 GB dd writes on my four
client/servers, 2 of my clients died (the other two finished their tasks
without issue and stayed operational). There were a large number of
"frame :" lines in glusterfs.log, and the log ended with:
frame : type(2) op(0)
frame : type(2) op(0)
2008-06-11 20:43:20 C [common-utils.c:155:gf_print_bytes] : xfer ==
142359133061, rcvd == 130379872
[0xb7fb4420]
/lib/tls/i686/cmov/libc.so.6(abort+0x101)[0xb7e4ea01]
/usr/lib/glusterfs/1.4.0qa19/xlator/mount/fuse.so[0xb758dd84]
/lib/tls/i686/cmov/libpthread.so.0[0xb7f764fb]
/lib/tls/i686/cmov/libc.so.6(clone+0x5e)[0xb7ef8e5e]
---------
The server processes all stayed up and kept very reasonable memory
consumption.
As before, this is 4 nodes (4 exports per node, plus AFRed namespace)
acting as clients and servers, io-threads and posix locks loaded on the
server-side, AFR and unify used on the client-side.
Thanks,
Brent
