I'm sharing a folder using nfs on a Linux system. It's very slow, and I found that there are many getattr requests.
After I mount my shared nfs folder on another machine, I'm trying to copy the files out. It's very slow. What I found via tcpdump is lots of getattr calls happening. Most of the requests are on the same file, which I don't understand either.
This is the mount command I'm using:
mount -t nfs -o rw,bg,hard,rsize=1048576,wsize=1048576,vers=3,nointr,timeo=600,actimeo=0,nolock,tcp
content of /etc/exportfs:
/dept/nfs *(no_subtree_check,rw,async)
What is the possible reason why there are so many getattr calls?
Is there any configuration problem?
It's because of the mount option: actimeo=0. This disables attribute caching on the client, which will send getattr requests to the server.
Related
I'm wondering if there is any way to practically speedup directory listings of a s3fs mount? I have a WebDAV server, only for read operations, that basically access my s3fs mount. The problem is that listing directories is slow, while transfer speed is fine.
So I started to look a bit around the web a stumbled across "JuiceFS", sadly this was also not an option for several reasons. Then I tried "vmtouch" to index the mounted s3 storage to local memory, this is also not working as it's a shared resourced managed by the fuse kernel extension.
Even using S3FS built-in cache does not solve the issue, instead it makes it even worse as the file first getting downloaded from s3 into the cache locally and then served via WebDav ...
Is there no way to just speedup directory listing using S3? Basically, this is all I need in the end and no fancy POSIX compatible Block Device like JuiceFS which basically creates its own logic on top of your s3 bucket ... Not what I was searching for.
Unfortunately s3fs 1.91 has poor readdir performance. There are a few open issues and pull requests that track future improvements:
Option to not use head requests
Consider changing -o notsup_compat_dir default
Consider changing -o noobj_cache default
Increase -o multireq_max
Issue parallel requests in get_object_attribute
You can toggle #2-4 via command-line flags today but #5 is still in-progress. #1 is the big win that would give a 100x speedup but trades off less POSIX compatibility, e.g., no UID/GID, no permissions. One alternative that you can try today is goofys which implements #1.
On an Ubuntu 18.01 LFS system (my private home server, only accessible from my home LAN), I have created an Apache/WSGI/flask web application that mounts encrypted drives (via cryptsetup luksOpen and then mount) on web request. Password for cryptsetup is sent via HTTP POST, a python script (that calls itself via sudo and then calls cryptsetup and mount via subprocess.run) is entered in /etc/sudoers.d... without the need for a password, so everything runs fine with sudo -u www-data ./starthepythonscript_that_mounts_stuff_via_sudo.py without a password. I can see mounted drives and cd into them after calling this script as just mentioned.
However, this behavior seems to be different when the script is called from the same user "www-data", but from the Apache WSGI Ubuntu service. In this case, mounting seems to succeed, but the mounts are not visible on the system (neither when listing contents of the corresponding mounted folder, nor when typing mount - e.g. as root - on the system that runs Apache). They just don't show up - is there a kind of sandbox mechanism implemented for the Apache service on Ubuntu?
My goal is to mount the drives via the Apache/WSGI/flask/sudo script, but in a "normal" way such that users on the same machine can see this via mount or cd into it.
Any hint is appreciated!
Nevermind - I have found the answer on unix.stackexchange.com - the apache service is started with PrivateTmp=true, which creates a private mount kernel namespace. One option is to remove that flag or set it to false - another would be to keep the setting and (from the root namespace) use nsenter -a -t APACHEPID bash to start a bash that sees the mounts.
I am trying to mount an nfs share to a solaris 10 machine at bootup without any luck so far.
The nfs share is accessible and mounts without any problems if I do so manually from the command line (mount -F nfs server_hostname:/exported_dir_path/ /mnt/tmpdir). But I don't know how to tell autofs to mount it at bootup.
We have another machine on the same network (also solaris) that has it working, but I can't figure out how is it configured differently from the non-working one.
I googled the problem and found that /etc/vfstab, /etc/auto_master, and /etc/hosts files need to have proper entries to make this work. I compared these files from the non-working machine with the ones on working machine but did not notice any differences.
Could someone please guide me to properly configure autofs to mount nfs shares on a solaris10 machine?
Thanks,
Aashish.
Is there a way to include configuration settings in Apache based on if a directory exists? Basically I have a portable hard drive that I transport between work and home that has some stuff I'm developing on it. I only want the Apache config to load a particular virtual host if the folder exists.
Since Apache 2.4.34 you can now use <IfFile>...</IfFile> which will check to see if a file exists. There's more details on the <IfFile> page.
No, there seems to be no direct way to do this.
The only thing that might be a solution is the IfDefine directive. You can define defines using the -d parameter to when the server is started.
The parameter-name argument is a define as given on the httpd command line via -Dparameter-, at the time the server was started.
You might be able to check for the existence of a directory in a batch or bash file, and set the -d parameter accordingly.
Whether that is an option, will depend on how your server is started from the portable hard drive.
I've come up with a solution that seems to work for Linux and OS X, and it hinges on "mountpoints". It might be possible to emulate it within Windows, as well, but you would probably have to get creative with FUSE and/or Cygwin.
If you create an empty folder in your home directory, such as "/Users/username/ExtraVhosts", you can add an apache directive to "Include /Users/username/ExtraVhosts/*".
Then, when you insert your thumb drive, you can mount somewhere and then use mountpoint "binding" to cross-link the ExtraVhosts folder to a folder on the mobile device.
An OS X example:
I have a thumb drive called 'Cherrybomb'
When I insert it, it always gets mounted to /Volumes/Cherrybomb
I can then use bindfs (sudo port install bindfs) to mount a subfolder of it, like so:
sudo bindfs /Volumes/Cherrybomb/Projects/vhosts /Users/username/ExtraVhosts
Then I can restart apache to read in the updated configuration:
sudo /opt/local/apache2/bin/apachectl restart
At that point, it's just a matter of adding entries in /etc/hosts for server aliases to get picked up.
The linux equivalent would be using the "--bind" parameter of the mount command.
One caveat: This makes it difficult to quickly unmount the USB drive, since it is always marked as "in use" by apache. Here's a removal procedure:
Close all open files and terminal sessions that are using the drive (the present-working-directory in terminal can cause unmount issues)
Stop apache: sudo /opt/local/apache2/bin/apachectl stop
umount /Users/username/ExtraVhosts
Then you can either unmount it graphically or manually (umount /Volumes/Cherrybomb).
If your work and home machines mount the drive to different locations, you could have multiple vhosts folders - home_vhost, work_vhost, etc - and use that in the binding step.
I hope this helps someone out :)
If you point apache to the mountpoint only there shouldn't be an issue. Just don't point Directory directives to directories within the drive.
eg, if you mount /dev/somedisk /mnt/somevhost, the
/mnt/somevhost directory will be there whether or not you have the drive mounted and apache will start. Apache doesn't care if the directory is empty so a <Directory "/mnt/somevhost"/> won't cause server to not start if the drive isn't mounted.
Work with UNIX not against it :-p This solution should be sufficient for development.
I have several web applications in production that utilize NFS mounts to share resources (usually static asset files) among web heads. In the event that an NFS mount becomes unavailable, Apache will hang requesting files that cannot be accessed, the kernel will log:
Nov 2 14:21:20 server2 kernel: nfs: server server1 not responding, still trying
I reproduced the behavior in RHEL5 running NFS v3 and Apache 2.2.3:
Create an NFS Mount on Server1 (contents of my /etc/exports)
/srv/test_share server2(rw)
Mount the NFS share on Server2 (contents of my /etc/fstab)
server1:/srv/test_share /mnt/test_share nfs defaults 0 0
Setup a virtual host in Apache with a simple HTML file referencing image files stored on the NFS sharen
Load the site, the html and image files all return 200
Unmount the NFS Share, loading the page returns 404s for the images referenced
Remount the NFS Share
Simulate an NFS crash by turning NFS off on Server1 - reloading the site hangs retrieving the referenced files.
Internet searches so far have not turned up a good solution. Basically the desired behavior would be for the web server to return 404s and not hang until the NFS mount recovers.
Cheers,
Ben
couple of options:
get your nfs mount options right, you need to do a soft mount so nfs access can be interupted. try soft,intr,timeo=10 instead of default
sync your document roots with something else like rsync, or script yourself a semi-atomatic checkout/export from your SCM, if you use one. SCM use is recommended anyway, gives you the possibility to revert to the last working version, for instance
use a real distributed filesystem (preferably fault tolerant like coda) or even a distributed block device system like drdb
option 2 and 3 give you disconnected operation and are therefore much more robust than nfs. drdb is sexy, but my advice would be option 2 with somwething like git or svn, simple and robust
I would not directly serve from the NFS mount, but instead from your local filesystem.
It wouldn't be too hard to setup a cron job that synced the NFS mount to the local file system every few minutes. Apache would serve its content from there, not depending on the NFS mount. If the mount goes down, Apache would still be able to serve the assets, although they might be out of date until the NFS mount comes back up.