On Aug 17, 2009, at 8:30 PM, Ian Kent wrote:
On Thu, 2009-08-13 at 12:18 -0300, Carlos André wrote:
Filled bug report:
https://bugzilla.redhat.com/show_bug.cgi?id=517349
Hi Carlos,
I have a patched source rpm to add a mount wait parameter to autofs
located at:
http://people.redhat.com/~ikent/autofs-5.0.1-0.rc2.131.bz517349.1
Could you build it and see if it works.
I haven't tested it at all but it is fairly straight forward.
For NFSv2/v3, we have an rpcbind query followed by a MNT request, on
separate transports that the kernel can set up with fast timeouts
(when the underlying transport implementations eventually support this).
It could be more challenging for NFSv4 because the initial connection
is done on the same NFS transport that will be used for normal
operation. The connection timeout parameters can't easily be changed
from giving up quickly to normal, at least with the current RPC
infrastructure. Perhaps the first few NFSv4 operations could be done
on a separate transport.
It is still unclear if this is the right way to do this and what the
consequences are in sending a term signal to mount. This mount request
will likely be followed by other requests for the same mount causing
an
accumulation of mount(8) processes waiting for RPC timeouts before
they
can answer the TERM signal.
One concern is that this would leave a number of outgoing privileged
ports in use (until the mount attempts time out, then after another
two minutes the port leaves TIME_WAIT). Clearly we need an in-kernel
solution to canceling mount requests without leaving a bunch of
resources tied up. Probably the best approach, in general terms, is
to tell mount(2) how long to wait, then let the kernel abort it as
needed.
Anyway, for information the patch included in the source rpm above is:
autofs-5.0.4 - add mount wait parameter
From: Ian Kent <raven@xxxxxxxxxx>
Often delays when trying to mount from a server that is not reponding
for some reason are undesirable. To try and prevent these delays we
provide a configuration setting to limit the time that we wait for
our spawned mount(8) process to complete before sending it a SIGTERM
signal. This patch adds a configuration parameter to allow us to
request we limit the time we wait for mount(8) to complete before
send it a TERM signal.
---
daemon/spawn.c | 3 ++-
include/defaults.h | 2 ++
lib/defaults.c | 13 +++++++++++++
man/auto.master.5.in | 7 +++++++
redhat/autofs.sysconfig.in | 9 +++++++++
samples/autofs.conf.default.in | 9 +++++++++
6 files changed, 42 insertions(+), 1 deletion(-)
--- autofs-5.0.1.orig/daemon/spawn.c
+++ autofs-5.0.1/daemon/spawn.c
@@ -312,6 +312,7 @@ int spawn_mount(unsigned logopt, ...)
unsigned int options;
unsigned int retries = MTAB_LOCK_RETRIES;
int update_mtab = 1, ret, printed = 0;
+ unsigned int wait = defaults_get_mount_wait();
char buf[PATH_MAX];
/* If we use mount locking we can't validate the location */
@@ -353,7 +354,7 @@ int spawn_mount(unsigned logopt, ...)
va_end(arg);
while (retries--) {
- ret = do_spawn(logopt, -1, options, prog, (const char **) argv);
+ ret = do_spawn(logopt, wait, options, prog, (const char **) argv);
if (ret & MTAB_NOTUPDATED) {
struct timespec tm = {3, 0};
--- autofs-5.0.1.orig/include/defaults.h
+++ autofs-5.0.1/include/defaults.h
@@ -24,6 +24,7 @@
#define DEFAULT_TIMEOUT 600
#define DEFAULT_NEGATIVE_TIMEOUT 60
+#define DEFAULT_MOUNT_WAIT -1
#define DEFAULT_UMOUNT_WAIT 12
#define DEFAULT_BROWSE_MODE 1
#define DEFAULT_LOGGING 0
@@ -62,6 +63,7 @@ struct ldap_schema *defaults_get_schema(
struct ldap_searchdn *defaults_get_searchdns(void);
void defaults_free_searchdns(struct ldap_searchdn *);
unsigned int defaults_get_append_options(void);
+unsigned int defaults_get_mount_wait(void);
unsigned int defaults_get_umount_wait(void);
const char *defaults_get_auth_conf_file(void);
unsigned int defaults_get_map_hash_table_size(void);
--- autofs-5.0.1.orig/lib/defaults.c
+++ autofs-5.0.1/lib/defaults.c
@@ -45,6 +45,7 @@
#define ENV_NAME_VALUE_ATTR "VALUE_ATTRIBUTE"
#define ENV_APPEND_OPTIONS "APPEND_OPTIONS"
+#define ENV_MOUNT_WAIT "MOUNT_WAIT"
#define ENV_UMOUNT_WAIT "UMOUNT_WAIT"
#define ENV_AUTH_CONF_FILE "AUTH_CONF_FILE"
@@ -323,6 +324,7 @@ unsigned int defaults_read_config(unsign
check_set_config_value(key, ENV_NAME_ENTRY_ATTR, value,
to_syslog) ||
check_set_config_value(key, ENV_NAME_VALUE_ATTR, value,
to_syslog) ||
check_set_config_value(key, ENV_APPEND_OPTIONS, value,
to_syslog) ||
+ check_set_config_value(key, ENV_MOUNT_WAIT, value, to_syslog)
||
check_set_config_value(key, ENV_UMOUNT_WAIT, value, to_syslog)
||
check_set_config_value(key, ENV_AUTH_CONF_FILE, value,
to_syslog) ||
check_set_config_value(key, ENV_MAP_HASH_TABLE_SIZE, value,
to_syslog))
@@ -652,6 +654,17 @@ unsigned int defaults_get_append_options
return res;
}
+unsigned int defaults_get_mount_wait(void)
+{
+ long wait;
+
+ wait = get_env_number(ENV_MOUNT_WAIT);
+ if (wait < 0)
+ wait = DEFAULT_MOUNT_WAIT;
+
+ return (unsigned int) wait;
+}
+
unsigned int defaults_get_umount_wait(void)
{
long wait;
--- autofs-5.0.1.orig/man/auto.master.5.in
+++ autofs-5.0.1/man/auto.master.5.in
@@ -175,6 +175,13 @@ Set the default timeout for caching fail
60). If the equivalent command line option is given it will override
this
setting.
.TP
+.B MOUNT_WAIT
+Set the default time to wait for a response from a spawned mount(8)
+before sending it a SIGTERM. Note that we still need to wait for the
+RPC layer to timeout before the sub-process exits so this isn't ideal
+but it is the best we can do. The default is to wait until mount(8)
+returns without intervention.
+.TP
.B UMOUNT_WAIT
Set the default time to wait for a response from a spawned umount(8)
before sending it a SIGTERM. Note that we still need to wait for the
--- autofs-5.0.1.orig/redhat/autofs.sysconfig.in
+++ autofs-5.0.1/redhat/autofs.sysconfig.in
@@ -14,6 +14,15 @@ TIMEOUT=300
#
#NEGATIVE_TIMEOUT=60
#
+# MOUNT_WAIT - time to wait for a response from umount(8).
+# Setting this timeout can cause problems when
+# mount would otherwise wait for a server that
+# is temporarily unavailable, such as when it's
+# restarting. The defailt of waiting for mount(8)
+# usually results in a wait of around 3 minutes.
+#
+#MOUNT_WAIT=-1
+#
# UMOUNT_WAIT - time to wait for a response from umount(8).
#
#UMOUNT_WAIT=12
--- autofs-5.0.1.orig/samples/autofs.conf.default.in
+++ autofs-5.0.1/samples/autofs.conf.default.in
@@ -14,6 +14,15 @@ TIMEOUT=300
#
#NEGATIVE_TIMEOUT=60
#
+# MOUNT_WAIT - time to wait for a response from umount(8).
+# Setting this timeout can cause problems when
+# mount would otherwise wait for a server that
+# is temporarily unavailable, such as when it's
+# restarting. The defailt of waiting for mount(8)
+# usually results in a wait of around 3 minutes.
+#
+#MOUNT_WAIT=-1
+#
# UMOUNT_WAIT - time to wait for a response from umount(8).
#
#UMOUNT_WAIT=12
Thanks!
2009/8/13 Carlos André <candrecn@xxxxxxxxx>:
2009/8/13 Ian Kent <ikent@xxxxxxxxxx>:
Carlos André wrote:
Today (2009-08-12) I'm using:
kernel-2.6.18-128.2.1.el5
autofs-5.0.1-0.rc2.102.el5_3.1
Thanks,
My mistake, the wait time I was referring to is used for umounts
during
expires and is present in rev rc2.102.
It shouldn't be hard to add this for mount as well.
Would you like me to put something together?
Sure! that 'll help me a lot (and for sure another ppl) :) Thanks :)
Probably would be good to test something out to see if we can
make a
difference with the killing mount after some configured timeout
but, if
we make progress, probably the best way to deal with it is for
you to
log a bug against rhel-5 so I can get it committed to the rhel
package.
The possible issue is that I'm not sure if the RPC subsystem in the
above rhel kernel will respond well to process death with potential
outstanding requests. But we'll see.
Ok, on my way :)
Thanks a lot!
Look my last test:
--------------------------------------------------------------
[root@KSTATION areas]# time ls testdown
ls: testdown: No such file or directory
real 3m9.025s
user 0m0.000s
sys 0m0.002s
Aug 12 12:57:07 KSTATION automount[15471]: sun_mount: parse(sun):
mounting root /misc/areas, mountpoint testdown, what
1.2.3.4:/areas/testdown, fstype nfs4, options
acl,sec=krb5p,proto=tcp,retry=0
Aug 12 12:57:07 KSTATION automount[15471]: do_mount:
1.2.3.4:/areas/testdown /misc/areas/testdown type nfs4 options
acl,sec=krb5p,proto=tcp,retry=0 using module nfs4
Aug 12 12:57:07 KSTATION automount[15471]: mount_mount:
mount(nfs):
root=/misc/areas name=testdown what=1.2.3.4:/areas/testdown,
fstype=nfs4, options=acl,sec=krb5p,proto=tcp,retry=0
Aug 12 12:57:07 KSTATION automount[15471]: mount_mount:
mount(nfs):
nfs options="acl,sec=krb5p,proto=tcp,retry=0", nosymlink=0, ro=0
Aug 12 12:57:07 KSTATION automount[15471]: mount_mount:
mount(nfs):
calling mkdir_path /misc/areas/testdown
Aug 12 12:57:07 KSTATION automount[15471]: mount_mount:
mount(nfs):
calling mount -t nfs4 -s -o acl,sec=krb5p,proto=tcp,retry=0
1.2.3.4:/areas/testdown /misc/areas/testdown
Aug 12 12:58:12 KSTATION automount[15471]: st_expire: state 1
path /misc
Aug 12 12:58:12 KSTATION automount[15471]: expire_proc: exp_proc =
3078093712 path /misc
Aug 12 12:58:13 KSTATION automount[15471]: expire_proc_indirect: 2
submounts remaining in /misc
Aug 12 12:58:13 KSTATION automount[15471]: expire_cleanup: got
thid
3078093712 path /misc stat 3
Aug 12 12:58:13 KSTATION automount[15471]: expire_cleanup:
sigchld:
exp 3078093712 finished, switching from 2 to 1
Aug 12 12:58:13 KSTATION automount[15471]: st_ready: st_ready():
state
= 2 path /misc
Aug 12 12:59:28 KSTATION automount[15471]: st_expire: state 1
path /misc
Aug 12 12:59:28 KSTATION automount[15471]: expire_proc: exp_proc =
3078093712 path /misc
Aug 12 12:59:28 KSTATION automount[15471]: expire_proc_indirect: 2
submounts remaining in /misc
Aug 12 12:59:28 KSTATION automount[15471]: expire_cleanup: got
thid
3078093712 path /misc stat 3
Aug 12 12:59:28 KSTATION automount[15471]: expire_cleanup:
sigchld:
exp 3078093712 finished, switching from 2 to 1
Aug 12 12:59:28 KSTATION automount[15471]: st_ready: st_ready():
state
= 2 path /misc
Aug 12 13:00:16 KSTATION automount[15471]: >> mount: mount to NFS
server '1.2.3.4' failed: timed out (giving up).
Aug 12 13:00:16 KSTATION automount[15471]: mount(nfs): nfs: mount
failure 1.2.3.4:/areas/testdown on /misc/areas/testdown
Aug 12 13:00:16 KSTATION automount[15471]: send_fail: token = 17
Aug 12 13:00:16 KSTATION automount[15471]: failed to mount /misc/
areas/testdown
Aug 12 13:00:43 KSTATION automount[15471]: st_expire: state 1
path /misc
--------------------------------------------------------------
2009/8/12 Ian Kent <ikent@xxxxxxxxxx>:
Carlos André wrote:
Hi Ian,
I'm getting crazy trying put "retry=" to work on mount... this
option
just DONT WORK if use proto=tcp and/OR kerberos (sec=krb5/
krb5i/krb5p)
like you can see on my previous emails...
Right, my mistake for not looking closely enough at post.
Maybe this is related to the same sort of problem we had with
mount in
the past, before the options parsing went into the kernel,
where other
services, like portmapper (or rpcbind), were being done with
different
timeout parameters before the RPC calls for mounting. That's
just an
example as NFSv4 shouldn't be sensitive to portmapper anyway.
But what version of autofs and kernel did you say you were using?
I appreciate any help.
Carlos.
2009/8/12 Ian Kent <ikent@xxxxxxxxxx>:
Chuck Lever wrote:
On Aug 11, 2009, at 8:41 AM, Carlos André wrote:
This long timeout is good if workstation need mount a
critical
directory using /etc/fstab on boot (for example)..
But in my case, using this loooong timeout doesnt make any
sense,
since autofs retry mount directory on-access. This in fact
gives me
alot of headaches, coz user login 'll just hangs if one
server goes
down for any reason, and will again hangs if user try
access directory
pointing to a NFS down server...
"retry=0" means the mount command will fail as soon as the
first
mount(2) system call fails. When you set SYN retries to 1,
this means
after 9 seconds, the connect fails, and that causes the
mount(2) system
call to fail.
Recent conversations with Ian suggested that a long timeout
was desired
for automounter as well as other cases. Ian, is there
something else we
need to consider to determine the correct retry timeout for
NFS/TCP
mount points handled via automounter? How should mount.nfs
wait so we
don't make other use cases worse? (Looks like most of the
history is
intact below).
Of course we know that autofs is entirely at the mercy of
mount(8) (and
mount.nfs in particular). This has always been a difficult
situation for
the automounter because interactive mount invocations should
wait. But I
believe automount mounts should always time out quickly, but
that leads
to its own set of problems, especially when home directories
are concerned.
I think adding "retry=0" is the right thing to do myself but
I'm not
certain that will work as we expect. I'll have to do some
experimentation.
How long do you think is appropriate for the automounter to
wait if the
server is down, in your case, Carlos?
Am losing something or there have was something weirdo...!?
------------------------------------------------
[root@KSTATION ~]# echo 5 > /proc/sys/net/ipv4/
tcp_syn_retries [DEFAULT]
[root@KSTATION ~]# time mount 1.2.3.4:/blabla /tmp/ -t nfs4
-o
proto=tcp,retry=1
mount: mount to NFS server '1.2.3.4' failed: timed out
(giving up).
real 3m9.000s
user 0m0.002s
sys 0m0.001s
[root@KSTATION ~]# time mount 1.2.3.4:/blabla /tmp/ -t nfs4
-o
sec=krb5p,proto=tcp,retry=1
mount: mount to NFS server '1.2.3.4' failed: timed out
(giving up).
real 3m9.000s
user 0m0.000s
sys 0m0.002s
[root@KSTATION ~]# time mount 1.2.3.4:/blabla /tmp/ -t nfs4
-o
proto=tcp,retry=0
mount: mount to NFS server '1.2.3.4' failed: timed out
(giving up).
real 3m9.001s
user 0m0.000s
sys 0m0.003s
[root@KSTATION ~]# time mount 1.2.3.4:/blabla /tmp/ -t nfs4
-o
sec=krb5p,proto=tcp,retry=0
mount: mount to NFS server '1.2.3.4' failed: timed out
(giving up).
real 3m9.001s
user 0m0.002s
sys 0m0.001s
[root@KSTATION ~]# echo 1 > /proc/sys/net/ipv4/
tcp_syn_retries [ 5 to 1 ]
[root@KSTATION ~]# time mount 1.2.3.4:/blabla /tmp/ -t nfs4
-o
proto=tcp,retry=1
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying). [x 6]
mount: mount to NFS server '1.2.3.4' failed: timed out
(giving up).
real 1m3.002s
user 0m0.000s
sys 0m0.002s
[root@KSTATION ~]# time mount 1.2.3.4:/blabla /tmp/ -t nfs4
-o
sec=krb5p,proto=tcp,retry=1
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying). [x 13]
mount: mount to NFS server '1.2.3.4' failed: timed out
(giving up).
real 2m6.000s
user 0m0.000s
sys 0m0.002s
[root@KSTATION ~]# time mount 1.2.3.4:/blabla /tmp/ -t nfs4
-o
proto=tcp,retry=0
mount: mount to NFS server '1.2.3.4' failed: timed out
(giving up).
real 0m9.003s
user 0m0.001s
sys 0m0.002s
[root@KSTATION ~]# time mount 1.2.3.4:/blabla /tmp/ -t nfs4
-o
sec=krb5p,proto=tcp,retry=0
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying). [x 13]
mount: mount to NFS server '1.2.3.4' failed: timed out
(giving up).
real 2m6.001s
user 0m0.001s
sys 0m0.002s
[root@KSTATION ~]#
------------------------------------------------
max timeout goes to 2m6s changing tcp_syn_retries from 5 to
1... and
using retry=0 without kerberos I got only 9s...
*sigh*
2009/8/10 Chuck Lever <chuck.lever@xxxxxxxxxx>:
On Aug 10, 2009, at 4:05 PM, Carlos André wrote:
Something funny: Using default tcp_syn_retries (5) i got
"3,6,12,24,48,96" secs interval... but if i change
tcp_syn_retries to
1 i got "3,6,3,6,3,6..." secs interval...
Right. Normally the RPC client calls the kernel's socket
connect
function,
which does 6 SYN retries. That one call usually takes
longer than
the RPC
client's connect timeout, so it only makes one connect
call, and then
fails.
Reducing the number of SYN retries per connect attempt
causes the RPC
client
to retry the connect call until its connect timeout
expires. Each
connect
call resets the SYN timeout to 3 seconds.
[root@KSERVER mnt]# time mount 1.2.3.4:/blabla tmp/ -t
nfs4 -o
sec=krb5p,proto=tcp
mount: mount to NFS server '1.2.3.4' failed: timed out
(giving up).
real 3m9.000s
user 0m0.000s
sys 0m0.002s
[root@KSERVER /]# echo 1 > /proc/sys/net/ipv4/
tcp_syn_retries
[root@KSERVER mnt]# time mount 1.2.3.4:/blabla tmp/ -t
nfs4 -o
sec=krb5p,proto=tcp ("retry=1" = no change)
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(retrying).
mount: mount to NFS server '1.2.3.4' failed: timed out
(giving up).
real 2m6.004s
user 0m0.000s
sys 0m0.004s
(3,6,3,6... secs interval)
2009/8/10 Carlos André <candrecn@xxxxxxxxx>:
No, i'm just using packages from CentOS repo...
And u're right about expo retries... with tcpdump i've
monitored
traffic and i got SYN retries in 3, 6, 12, 24, 48, 96
secs on port
2049...
I tried use "retry=1" option on mount without any
change... I dont
want change source or tcp timers... just NFSv4 client.
2009/8/10 Chuck Lever <chuck.lever@xxxxxxxxxx>:
On Aug 10, 2009, at 2:29 PM, Carlos André wrote:
Bruce, no... you're right. I'm describing a situation
where my
server
died... i need mount fail faster (10 or 15 secs max)
than 3 minutes
and 9 seconds...
The 189 second timeout is likely how long it takes the
kernel to
give up
trying to connect a TCP socket to the server (6 SYN
attempts with
exponential retries, or something like that). For
stock CentOS
5.3, I
think
user space does only a DNS lookup for normal NFSv4
mounts -- the
kernel
just
tries to connect a TCP socket to port 2049, with no
preceding rpcbind
request.
Carlos, let us know if you have replaced any NFS-
related CentOS
components
(kernel, nfs-utils) with something you've built yourself.
2009/8/7 J. Bruce Fields <bfields@xxxxxxxxxxxx>:
On Fri, Aug 07, 2009 at 09:42:18AM +0300, Benny
Halevy wrote:
On Aug. 07, 2009, 3:18 +0300, Carlos André <candrecn@xxxxxxxxx
>
wrote:
Anyone ?
2009/7/29 Carlos André <candrecn@xxxxxxxxx>:
PPL, I need put a CentOS 5.3 (updated) NFSv4
server to work with
Kerberos
and AutoFS, but i got a problem: If NFS server
goes down i get a
LOOOOOOONG
mount timeout on CentOS 5.3 (updated) NFSv4
client...
Since i need mount some (3 to 6) dirs at user
logon process, if
mount
hangs,
user logon hangs. Then i want configure it to
timeout (if server
down)
after
10-15 secs (MAX) on each mount attempt.
I already make a lab and tried a LOT of
combinations, there my
findings
(server DOWN IP: 172.16.0.10 / client IP:
172.16.1.10) using
basic
command
(time mount 172.16.0.10:/remotedir /localdir/ -t
nfs4 -o
sec=krb5,proto=<tcp/udp>) from NFS client:
- Once i try access mount point using AutoFS
(proto=tcp OR
proto=udp)
it
hangs for 189 secs (3m9s: real 3m9.001s) until
show error
(mount:
mount to
NFS server '172.16.0.10' failed: timed out (giving
up))
Sounds like you're hitting the server's grace period.
I thought he was describing a situation where the
server the server
is completely gone and isn't coming back, and
wondering how to make
the
mount fail faster. But I may be misunderstanding.
--b.
--
To unsubscribe from this list: send the line
"unsubscribe
linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
--
Chuck Lever
chuck[dot]lever[at]oracle[dot]com
--
Chuck Lever
chuck[dot]lever[at]oracle[dot]com
--
Chuck Lever
chuck[dot]lever[at]oracle[dot]com
--
Chuck Lever
chuck[dot]lever[at]oracle[dot]com
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
