Brent,
I have now switched all clients and servers to 2.4-Mainline. The read-ahead
crash is fixed. The spurious disconnect errors remain but I suspect it may be
tied to large reads. I cannot get through a large mailbox reindex without a
disconnect in the middle which is now recovered from but leaves 300-1000
unindexed files. I turned off stat-prefetch because my application didn't need
it but I didn't notice any errors with it on.
One thing I am noticing. When I run:
find /mnt/gluster/folder -type f -exec md5sum {} \;
on a tree with over 50,000 files, it does not run to completion. After about
30,000 files or so if fails with cannot find file or folder errors. It does not
seem to be tied to any errors in the logs. find without the exec and du run
correctly on the same tree. Do you have a way of duplicating this test?
Harris
----- Original Message -----
From: "Brent A Nelson" <address@hidden>
To: "Anand Avati" <address@hidden>
Cc: "Harris Landgarten" <address@hidden>, address@hidden
Sent: Thursday, May 24, 2007 11:30:16 AM (GMT-0500) America/New_York
Subject: Re: [Gluster-devel] Segfault in read_ahead in 1.3.0_pre4
I was getting the same behavior in my testing. After I reported it, the
readahead crash was quickly patched, but the random disconnect is still
very much a mystery...
I noticed that you are using stat-prefetch; have you encountered any
issues? I was finding that du's on complex directories could return
abnormal results and/or errors, so it seemed that heavy metadata queries
were occassionally glitchy. Without stat-prefetch, it's been fine. If
you've been having good luck with it, maybe I should try again.
Thanks,
Brent
On Thu, 24 May 2007, Anand Avati wrote:
Harris,
this bug was fixed a few days back and the fix is available in the
glusterfs--mainline--2.4 repository latest checkout.
thanks,
avati
2007/5/24, Harris Landgarten <address@hidden>:
I am running glusterfs in a very basic configuration on Amazon EC
instances. I have a 2 brick cluster and 2 clients. One of the clients is
running Zimbra and I am using the cluster as secondary storage for the mail
store. I have repeatedly tried to reindex a mailbox with 31000 items. Most
of the email is on the cluster. The entire process takes about 2 hours.
Part way through I get at least one TCP Disconnect which seems random. With
read_ahead enabled on the client, the disconnect results in a segfault and
the mount point disappears. When I disabled read_ahead on the client, the
disconnect was recovered from, and the process completed. This is the
backtrace from the read_ahead segfault:
[May 23 20:00:37] [CRITICAL/client-protocol.c:218/call_bail()]
client/protocol:bailing transport
[May 23 20:00:37] [DEBUG/tcp.c:123/cont_hand()] tcp:forcing poll/read/write
to break on blocked socket (if any)
[May 23 20:00:37] [ERROR/common-utils.c:55/full_rw()] libglusterfs:full_rw:
0 bytes r/w instead of 113 (errno=115)
[May 23 20:00:37] [DEBUG/protocol.c:244/gf_block_unserialize_transport()]
libglusterfs/protocol:gf_block_unserialize_transport: full_read of header
failed
[May 23 20:00:37] [DEBUG/client-protocol.c:2605/client_protocol_cleanup()]
protocol/client:cleaning up state in transport object 0x8078a08
[May 23 20:00:37] [CRITICAL/common-utils.c:215/gf_print_trace()]
debug-backtrace:Got signal (11), printing backtrace
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:/usr/lib/libglusterfs.so.0(gf_print_trace+0x2d)
[0xb7f2584d]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:[0xbfffe420]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:/usr/lib/glusterfs/1.3.0-pre4/xlator/performance/read-ahead.so(ra_page_error+0x47)
[0xb755e587]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:/usr/lib/glusterfs/1.3.0-pre4/xlator/performance/read-ahead.so
[0xb755ecf0]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:/usr/lib/glusterfs/1.3.0-pre4/xlator/performance/write-behind.so
[0xb7561809]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:/usr/lib/glusterfs/1.3.0-pre4/xlator/cluster/unify.so
[0xb7564919]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:/usr/lib/glusterfs/1.3.0-pre4/xlator/protocol/client.so
[0xb756d17b]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:/usr/lib/glusterfs/1.3.0-pre4/xlator/protocol/client.so
[0xb75717a5]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:/usr/lib/libglusterfs.so.0(transport_notify+0x1d)
[0xb7f26d2d]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:/usr/lib/libglusterfs.so.0(sys_epoll_iteration+0xe7)
[0xb7f279d7]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:/usr/lib/libglusterfs.so.0(poll_iteration+0x1d)
[0xb7f26ddd]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:glusterfs [0x804a15e]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:/lib/libc.so.6(__libc_start_main+0xdc) [0xb7dca8cc]
[May 23 20:00:37] [CRITICAL/common-utils.c:217/gf_print_trace()]
debug-backtrace:glusterfs [0x8049e71]
Segmentation fault (core dumped)
This is a sample of the debug log with read_ahead turned off
[May 24 05:35:05] [CRITICAL/client-protocol.c:218/call_bail()]
client/protocol:bailing transport
[May 24 05:35:05] [DEBUG/tcp.c:123/cont_hand()] tcp:forcing poll/read/write
to break on blocked socket (if any)
[May 24 05:35:05] [ERROR/common-utils.c:55/full_rw()] libglusterfs:full_rw:
0 bytes r/w instead of 113 (errno=115)
[May 24 05:35:05] [DEBUG/protocol.c:244/gf_block_unserialize_transport()]
libglusterfs/protocol:gf_block_unserialize_transport: full_read of header
failed
[May 24 05:35:05] [DEBUG/client-protocol.c:2605/client_protocol_cleanup()]
protocol/client:cleaning up state in transport object 0x80783d0
[May 24 05:35:05] [CRITICAL/tcp.c:81/tcp_disconnect()]
transport/tcp:client1: connection to server disconnected
[May 24 05:35:05] [DEBUG/tcp-client.c:180/tcp_connect()] transport: tcp:
:try_connect: socket fd = 4
[May 24 05:35:05] [DEBUG/tcp-client.c:202/tcp_connect()] transport: tcp:
:try_connect: finalized on port `1022'
[May 24 05:35:05] [DEBUG/tcp-client.c:226/tcp_connect()]
tcp/client:try_connect: defaulting remote-port to 6996
[May 24 05:35:05] [DEBUG/tcp-client.c:262/tcp_connect()] tcp/client:connect
on 4 in progress (non-blocking)
[May 24 05:35:05] [DEBUG/tcp-client.c:301/tcp_connect()]
tcp/client:connection on 4 still in progress - try later
[May 24 05:35:05] [ERROR/client-protocol.c:204/client_protocol_xfer()]
protocol/client:transport_submit failed
[May 24 05:35:05] [DEBUG/client-protocol.c:2605/client_protocol_cleanup()]
protocol/client:cleaning up state in transport object 0x80783d0
[May 24 05:35:11] [DEBUG/tcp-client.c:310/tcp_connect()]
tcp/client:connection on 4 success, attempting to handshake
[May 24 05:35:11] [DEBUG/tcp-client.c:54/do_handshake()]
transport/tcp-client:dictionary length = 50
[May 24 07:20:10] [DEBUG/stat-prefetch.c:58/stat_prefetch_cache_flush()]
stat-prefetch:flush on: /
[May 24 07:20:20] [DEBUG/stat-prefetch.c:58/stat_prefetch_cache_flush()]
stat-prefetch:flush on: /backups/sessions
[May 24 07:57:12] [CRITICAL/client-protocol.c:218/call_bail()]
client/protocol:bailing transport
[May 24 07:57:12] [DEBUG/tcp.c:123/cont_hand()] tcp:forcing poll/read/write
to break on blocked socket (if any)
[May 24 07:57:12] [ERROR/common-utils.c:55/full_rw()] libglusterfs:full_rw:
0 bytes r/w instead of 113 (errno=115)
[May 24 07:57:12] [DEBUG/protocol.c:244/gf_block_unserialize_transport()]
libglusterfs/protocol:gf_block_unserialize_transport: full_read of header
failed
[May 24 07:57:12] [DEBUG/client-protocol.c:2605/client_protocol_cleanup()]
protocol/client:cleaning up state in transport object 0x80783d0
[May 24 07:57:12] [CRITICAL/tcp.c:81/tcp_disconnect()]
transport/tcp:client1: connection to server disconnected
[May 24 07:57:12] [DEBUG/tcp-client.c:180/tcp_connect()] transport: tcp:
:try_connect: socket fd = 4
[May 24 07:57:12] [DEBUG/tcp-client.c:202/tcp_connect()] transport: tcp:
:try_connect: finalized on port `1023'
[May 24 07:57:12] [DEBUG/tcp-client.c:226/tcp_connect()]
tcp/client:try_connect: defaulting remote-port to 6996
[May 24 07:57:12] [DEBUG/tcp-client.c:262/tcp_connect()] tcp/client:connect
on 4 in progress (non-blocking)
[May 24 07:57:12] [DEBUG/tcp-client.c:301/tcp_connect()]
tcp/client:connection on 4 still in progress - try later
[May 24 07:57:12] [ERROR/client-protocol.c:204/client_protocol_xfer()]
protocol/client:transport_submit failed
[May 24 07:57:12] [DEBUG/client-protocol.c:2605/client_protocol_cleanup()]
protocol/client:cleaning up state in transport object 0x80783d0
[May 24 07:57:12] [DEBUG/tcp-client.c:310/tcp_connect()]
tcp/client:connection on 4 success, attempting to handshake
[May 24 07:57:12] [DEBUG/tcp-client.c:54/do_handshake()]
transport/tcp-client:dictionary length = 50
This is the client config with read_ahead
### Add client feature and attach to remote subvolume
volume client1
type protocol/client
option transport-type tcp/client # for TCP/IP transport
# option ibv-send-work-request-size 131072
# option ibv-send-work-request-count 64
# option ibv-recv-work-request-size 131072
# option ibv-recv-work-request-count 64
# option transport-type ib-sdp/client # for Infiniband transport
# option transport-type ib-verbs/client # for ib-verbs transport
option remote-host xx.xxx.xx.xxx # IP address of the remote brick
# option remote-port 6996 # default server port is 6996
# option transport-timeout 30 # seconds to wait for a reply
# from server for each request
option remote-subvolume brick # name of the remote volume
end-volume
### Add client feature and attach to remote subvolume
volume client2
type protocol/client
option transport-type tcp/client # for TCP/IP transport
# option ibv-send-work-request-size 131072
# option ibv-send-work-request-count 64
# option ibv-recv-work-request-size 131072
# option ibv-recv-work-request-count 64
# option transport-type ib-sdp/client # for Infiniband transport
# option transport-type ib-verbs/client # for ib-verbs transport
option remote-host yy.yyy.yy.yyy # IP address of the remote brick
# option remote-port 6996 # default server port is 6996
# option transport-timeout 30 # seconds to wait for a reply
# from server for each request
option remote-subvolume brick # name of the remote volume
end-volume
volume bricks
type cluster/unify
subvolumes client1 client2
option scheduler alu
option alu.limits.min-free-disk 4GB
option alu.limits.max-open-files 10000
option alu.order disk-usage:read-usage:write-usage:open-files-usage
option alu.disk-usage.entry-threshold 2GB
option alu.disk-usage.exit-threshold 10GB
option alu.open-files-usage.entry-threshold 1024
option alu.open-files-usage.exit-threshold 32
option alu.stat-refresh.interval 10sec
end-volume
#
### Add writeback feature
volume writeback
type performance/write-behind
option aggregate-size 131072 # unit in bytes
subvolumes bricks
end-volume
### Add readahead feature
volume readahead
type performance/read-ahead
option page-size 65536 # unit in bytes
option page-count 16 # cache per file = (page-count x page-size)
subvolumes writeback
end-volume
### Add stat-prefetch feature
### If you are not concerned about performance of interactive commands
### like "ls -l", you wouln't need this translator.
volume statprefetch
type performance/stat-prefetch
option cache-seconds 2 # timeout for stat cache
subvolumes readahead
end-volume
This is the brick config:
### Export volume "brick" with the contents of "/home/export" directory.
volume brick
type storage/posix # POSIX FS translator
option directory /mnt/export # Export this directory
end-volume
volume iothreads
type performance/io-threads
option thread-count 8
subvolumes brick
end-volume
### Add network serving capability to above brick.
volume server
type protocol/server
option transport-type tcp/server # For TCP/IP transport
# option ibv-send-work-request-size 131072
# option ibv-send-work-request-count 64
# option ibv-recv-work-request-size 131072
# option ibv-recv-work-request-count 64
# option transport-type ib-sdp/server # For Infiniband transport
# option transport-type ib-verbs/server # For ib-verbs transport
# option bind-address 192.168.1.10 # Default is to listen on all
interfaces
# option listen-port 6996 # Default is 6996
# option client-volume-filename /etc/glusterfs/glusterfs-client.vol
subvolumes iothreads
# NOTE: Access to any volume through protocol/server is denied by
# default. You need to explicitly grant access through # "auth"
# option.
option auth.ip.brick.allow * # Allow access to "brick" volume
end-volume
_______________________________________________
Gluster-devel mailing list
address@hidden
http://lists.nongnu.org/mailman/listinfo/gluster-devel
--
Anand V. Avati
_______________________________________________
Gluster-devel mailing list
address@hidden
http://lists.nongnu.org/mailman/listinfo/gluster-devel