Posts Tagged ‘backlog’

Howto debug and remount NFS hangled filesystem on Linux

Monday, August 12th, 2019

nfsnetwork-file-system-architecture-diagram

If you're using actively NFS remote storage attached to your Linux server it is very useful to get the number of dropped NFS connections and in that way to assure you don't have a remote NFS server issues or Network connectivity drops out due to broken network switch a Cisco hub or other network hop device that is routing the traffic from Source Host (SRC) to Destination Host (DST) thus, at perfect case if NFS storage and mounted Linux Network filesystem should be at (0) zero dropped connectios or their number should be low. Firewall connectivity between Source NFS client host and Destination NFS Server and mount should be there (set up fine) as well as proper permissions assigned on the server, as well as the DST NFS should be not experiencing I/O overheads as well as no DNS issues should be present (if NFS is not accessed directly via IP address).
In below article which is mostly for NFS novice admins is described shortly few of the nuances of working with NFS.
 

1. Check nfsstat and portmap for issues

One indicator that everything is fine with a configured NFS mount is the number of dropped NFS connections
or with a very low count of dropped connections, to check them if you happen to administer NFS

nfsstat

 

linux:~# nfsstat -o net
Server packet stats:
packets    udp        tcp        tcpconn
0          0          0          0  


nfsstat is useful if you have to debug why occasionally NFS mounts are getting unresponsive.

As NFS is so dependent upon portmap service for mapping the ports, one other point to check in case of Hanged NFSes is the portmap service whether it did not crashed due to some reason.

 

linux:~# service portmap status
portmap (pid 7428) is running…   [portmap service is started.]

 

linux:~# ps axu|grep -i rpcbind
_rpc       421  0.0  0.0   6824  3568 ?        Ss   10:30   0:00 /sbin/rpcbind -f -w


A useful commands to debug further rcp caused issues are:

On client side:

 

rpcdebug -m nfs -c

 

On server side:

 

rpcdebug -m nfsd -c

 

It might be also useful to check whether remote NFS permissions did not changed with the good old showmount cmd

linux:~# showmount -e rem_nfs_server_host


Also it is useful to check whether /etc/exports file was not modified somehow and whether the NFS did not hanged due to attempt of NFS daemon to reload the new configuration from there, another file to check while debugging is /etc/nfs.conf – are there group / permissions issues as well as the usual /var/log/messages and the kernel log with dmesg command for weird produced NFS client / server or network messages.

nfs-utils disabled serving NFS over UDP in version 2.2.1. Arch core updated to 2.3.1 on 21 Dec 2017 (skipping over 2.2.1.) If UDP stopped working then, add udp=y under [nfsd] in /etc/nfs.conf. Then restart nfs-server.service.

If the remote NFS server is running also Linux it is useful to check its /etc/default/nfs-kernel-server configuration

At some stall cases it might be also useful to remount the NFS (but as there might be a process on the Linux server) trying to read / write data from the remote NFS mounted FS it is a good idea to check (whether a process / service) on the server is not doing I/O operations on the NFS and if such is existing to kill the process in question with fuser
 

linux:~# fuser -k [mounted-filesystem]
 

 

2. Diagnose the problem interactively with htop


    Htop should be your first port of call. The most obvious symptom will be a maxed-out CPU.
    Press F2, and under "Display options", enable "Detailed CPU time". Press F1 for an explanation of the colours used in the CPU bars. In particular, is the CPU spending most of its time responding to IRQs, or in Wait-IO (wio)?
 

3. Get more extensive Mount info with mountstats

 

nfs-utils package contains mountstats command which is very useful in debugging further the issues identified

$ mountstats
Stats for example:/tank mounted on /tank:
  NFS mount options: rw,sync,vers=4.2,rsize=524288,wsize=524288,namlen=255,acregmin=3,acregmax=60,acdirmin=30,acdirmax=60,soft,proto=tcp,port=0,timeo=15,retrans=2,sec=sys,clientaddr=xx.yy.zz.tt,local_lock=none
  NFS server capabilities: caps=0xfbffdf,wtmult=512,dtsize=32768,bsize=0,namlen=255
  NFSv4 capability flags: bm0=0xfdffbfff,bm1=0x40f9be3e,bm2=0x803,acl=0x3,sessions,pnfs=notconfigured
  NFS security flavor: 1  pseudoflavor: 0

 

NFS byte counts:
  applications read 248542089 bytes via read(2)
  applications wrote 0 bytes via write(2)
  applications read 0 bytes via O_DIRECT read(2)
  applications wrote 0 bytes via O_DIRECT write(2)
  client read 171375125 bytes via NFS READ
  client wrote 0 bytes via NFS WRITE

RPC statistics:
  699 RPC requests sent, 699 RPC replies received (0 XIDs not found)
  average backlog queue length: 0

READ:
    338 ops (48%)
    avg bytes sent per op: 216    avg bytes received per op: 507131
    backlog wait: 0.005917     RTT: 548.736686     total execute time: 548.775148 (milliseconds)
GETATTR:
    115 ops (16%)
    avg bytes sent per op: 199    avg bytes received per op: 240
    backlog wait: 0.008696     RTT: 15.756522     total execute time: 15.843478 (milliseconds)
ACCESS:
    93 ops (13%)
    avg bytes sent per op: 203    avg bytes received per op: 168
    backlog wait: 0.010753     RTT: 2.967742     total execute time: 3.032258 (milliseconds)
LOOKUP:
    32 ops (4%)
    avg bytes sent per op: 220    avg bytes received per op: 274
    backlog wait: 0.000000     RTT: 3.906250     total execute time: 3.968750 (milliseconds)
OPEN_NOATTR:
    25 ops (3%)
    avg bytes sent per op: 268    avg bytes received per op: 350
    backlog wait: 0.000000     RTT: 2.320000     total execute time: 2.360000 (milliseconds)
CLOSE:
    24 ops (3%)
    avg bytes sent per op: 224    avg bytes received per op: 176
    backlog wait: 0.000000     RTT: 30.250000     total execute time: 30.291667 (milliseconds)
DELEGRETURN:
    23 ops (3%)
    avg bytes sent per op: 220    avg bytes received per op: 160
    backlog wait: 0.000000     RTT: 6.782609     total execute time: 6.826087 (milliseconds)
READDIR:
    4 ops (0%)
    avg bytes sent per op: 224    avg bytes received per op: 14372
    backlog wait: 0.000000     RTT: 198.000000     total execute time: 198.250000 (milliseconds)
SERVER_CAPS:
    2 ops (0%)
    avg bytes sent per op: 172    avg bytes received per op: 164
    backlog wait: 0.000000     RTT: 1.500000     total execute time: 1.500000 (milliseconds)
FSINFO:
    1 ops (0%)
    avg bytes sent per op: 172    avg bytes received per op: 164
    backlog wait: 0.000000     RTT: 2.000000     total execute time: 2.000000 (milliseconds)
PATHCONF:
    1 ops (0%)
    avg bytes sent per op: 164    avg bytes received per op: 116
    backlog wait: 0.000000     RTT: 1.000000     total execute time: 1.000000 (milliseconds)


nfs-utils disabled serving NFS over UDP in version 2.2.1. Arch core updated to 2.3.1 on 21 Dec 2017 (skipping over 2.2.1.) If UDP stopped working then, add udp=y under [nfsd] in /etc/nfs.conf. Then restart nfs-server.service.
 

4. Check for firewall issues
 

If all fails make sure you don't have any kind of firewall issues. Sometimes firewall changes on remote server or somewhere in the routing servers might lead to stalled NFS mounts.

 

To use properly NFS as you should know as a minimum you need to have opened as ports is Port 111 (TCP and UDP) and 2049 (TCP and UDP) on the NFS server (side) as well as any traffic inspection routers on the road from SRC (Linux client host) and NFS Storage destination DST server.

There are also ports for Cluster and client status (Port 1110 TCP for the former, and 1110 UDP for the latter) as well as a port for the NFS lock manager (Port 4045 TCP and UDP) but having this opened or not depends on how the NFS is configured. You can further determine which ports you need to allow depending on which services are needed cross-gateway.
 

5. How to Remount a Stalled unresponsive NFS filesystem mount

 

At many cases situation with remounting stalled NFS filesystem is not so easy but if you're lucky a standard mount and remount should do the trick.

Most simple way to remout the NFS (once you're sure this might not disrupt any service) – don't blame me if you break something is with:
 

umount -l /mnt/NFS_mnt_point
mount /mnt/NFS_mnt_point


Note that the lazy mount (-l) umount opt is provided here as very often this is the only way to unmount a stalled NFS mount.

Sometimes if you have a lot of NFS mounts and all are inacessible it is useful to remount all NFS mounts, if the remote NFS is responsive this should be possible with a simple for bash loop:

for P in $(mount | awk '/type nfs / {print $3;}'); do echo $P; echo "sudo umount $P && sudo mount $P" && echo "ok :)"; done


If you cd /mnt/NFS_mnt_point and try ls and you get

$ ls
.: Stale File Handle

 

You will need to unmount the FS with forceful mount flag

umount -f /mnt/NFS_mnt_point
 

Sum it up


In this article, I've shown you a few simple ways to debug what is wrong with a Stalled / Hanged NFS filesystem present on a NFS server mounted on a Linux client server.
Above was explained the common issues caused by NFS portmap (rpcbind) dependency, how to its status is fine, some further diagnosis with htop and mountstat was pointed. I've pointed the minimum amount of TCP / UDP ports 2049 and 111 that needs to be opened for the NFS communication to work and finally explained on how to remount a stalled NFS single or all attached mount on a NFS client to restore to normal operations.
As NFS is a whole ocean of things and the number of ways it is used are too extensive this article is just a general info useful for the NFS dummy admin for more robust configs read some good book on NFS such as Managing NFS and NIS, 2nd Edition – O'Reilly Media and for Kernel related NFS debugging make sure you check as a minimum ArchLinux's NFS troubleshooting guide and sourceforge's NFS Troubleshoting and Optimizing NFS Performance guides.

 

Tags: , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,
Posted in Linux, NFS, Storage | No Comments »

Optimizing Linux TCP/IP Networking to increase Linux Servers Performance

Tuesday, April 8th, 2008

optimize-linux-servers-for-network-performance-to-increase-speed-and-decrease-hardware-costs-_tyan-exhibits-hpc-optimized-server-platforms-featuring-intel-xeon-processor-e7-4800-v3-e5-2600-supercomputing-15_full

Some time ago I thought of ways to optimize my Linux Servers network performance.

Even though there are plenty of nice articles on the topic on how to better optimize Linux server performance by tunning up the kernel sysctl (variables).

Many of the articles I found was not structed in enough understandable way so I decided togoogle around and  found few interesting websites which gives a good overview on how one can speed up a bit and decrease overall server loads by simply tuning few basic kernel sysctl variables.

Below article is a product of my research on the topic on how to increase my GNU / Linux servers performance which are mostly running LAMP (Linux / Apache / MySQL / PHP) together with Qmail mail servers.

The article is focusing on Networking as networking is usual bottleneck for performance.
Below are the variables I found useful for optimizing the Linux kernel Network stack.

Implementing the variables might reduce your server load or if not decrease server load times and CPU utilization, they would at lease increase thoroughput so more users will be able to access your servers with (hopefully) less interruptions.
That of course would save you some Hardware costs and raise up your Servers efficiency.

Here are the variables themselves and some good example:
 

# values.net.ipv4.ip_forward = 0 ( Turn off IP Forwarding )

net.ipv4.conf.default.rp_filter = 1

# ( Control Source route verification )
net.ipv4.conf.default.accept_redirects = 0

# ( Disable ICMP redirects )
net.ipv4.conf.all.accept_redirects = 0 ( same as above )
net.ipv4.conf.default.accept_source_route = 0

# ( Disable IP source routing )
net.ipv4.conf.all.accept_source_route = 0
( - || - )net.ipv4.tcp_fin_timeout = 40

# ( Decrease FIN timeout ) - Useful on busy/high load
serversnet.ipv4.tcp_keepalive_time = 4000 ( keepalive tcp timeout )
net.core.rmem_default = 786426 - Receive memory stack size ( a good idea to increase it if your server receives big files )
net.ipv4.tcp_rmem = "4096 87380 4194304"
net.core.wmem_default = 8388608 ( Reserved Memory per connection )
net.core.wmem_max = 8388608
net.core.optmem_max = 40960
( maximum amount of option memory buffers )

# like a homework investigate by yourself what the variables below stand for :)
net.ipv4.tcp_max_tw_buckets = 360000
net.ipv4.tcp_reordering = 5
net.core.hot_list_length = 256
net.core.netdev_max_backlog = 1024

 

# Below are newly added experimental
#net.core.rmem_max = 16777216
#net.core.wmem_max = 16777216
##kernel.msgmni = 1024
##kernel.sem = 250 256000 32 1024
##vm.swappiness=0
kernel.sched_migration_cost=5000000

 

Also a good sysctl.conf file which one might want to substitite or use as a skele for some productive server is ready for download here


Even if you can't reap out great CPU reduction benefits from integrating above values or similar ones, your overall LAMP performance to end customers should increase – at some occasions dramatically, at others little bit but still noticable.

If you're unsure on exact kernel variable values to use check yourself what should be the best values that fits you according to your server Hardware – usually this is done by experimenting and reading the kernel documentation as provided for each one of uplisted variables.

Above sysctl.conf is natively created to run on Debian and on other distributions like CentOS, Fedora Slackware some values might either require slight modifications.

Hope this helps and gives you some idea of how network optimization in Linux is usually done. Happy (hacking) tweakening !

Tags: , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,
Posted in Computer Security, Linux, Networking, Performance Tuning, System Administration | 2 Comments »