Graham Ollis π > Yars > Yars



Annotate this POD


View/Report Bugs
Module Version: 1.23   Source  


Yars - Yet Another RESTful-Archive Service


version 1.23


Create a configuration in ~/etc/Yars.conf

 url: http://localhost:9001
 start_mode: hypnotoad
   pid_file: <%= home %>/var/run/
   listen: [ 'http://localhost:9001' ]
   - url: http://localhost:9001
       - root: <%= home %>/var/data/disk1
         buckets: <%= json [ 0..9, 'a'..'f' ] %>

Create needed directories and run the server

 % mkdir -p ~/var/run ~/var/data/disk1
 % yars start

Upload a file:

 % md5sum foo.jog
 469f9b131cce1631ddd449fbef9059ba  foo.jpg
 % yarsclient upload foo.jpg

Download a file

 % yarsclient download foo.jpg 469f9b131cce1631ddd449fbef9059ba


Yars is a simple RESTful server for data storage.

Properly configured it provides consistent WRITE availability, and eventual READ availability. Once files are written to the storage cluster they are immutable (new files can -- even with the same filename) can also be written to the cluster.

It allows files to be PUT and GET based on their md5 sums and filenames, and uses a distributed hash table to store the files across any number of hosts and disks.

Files are assigned to disks and hosts based on their md5s in the following manner :

The first N digits of the md5 are considered the "bucket" for a file. e.g. for N=2, 256 buckets are then distributed among the disks in proportion to the size of each disk. The bucket distribution is done manually as part of the configuration (with the aid of an included tool, yars_generate_diskmap).

The server is controlled with the command line tool yars.

The basic operations of a running yars cluster are supporting requests of the form

 PUT http://$host/file/$filename
 GET http://$host/file/$md5/$filename
 HEAD http://$host/file/$md5/$filename
 GET http://$host/bucket_map

to store and retrieve files, where $host may be any of the hosts in the cluster, $md5 is the md5 of the content, and $filename is a filename for the content to be stored. See Yars::Routes for documentation of other routes.

Failover is handled in the following manner:

If the host to which a file is assigned is not available, then the file will be "stashed" on the filesystem for the host to which it was sent. If there is no space there, other hosts and disks will be tried until an available one is found. Because of this failover mechanism, the "stash" must be checked whenever a GET request is handled. A successful GET will return quickly, but an unsuccessful one will take longer because all of the stashes on all of the servers must be checked before a "404 Not Found" is returned.

Another tool yars_fast_balance is provided which takes files from stashes and returns them to their correct locations.

A client Yars::Client is also available (in a separate distribution), for interacting with a yars server.


simple single server configuration

This creates a single Yars server using hypnotoad with sixteen buckets.

Create a configuration file in ~/etc/Yars.conf with this content:

 # The first half of the configuration specifies the
 # generic Clustericious / web server settings for
 # the server
 start_mode : 'hypnotoad'
 url : http://localhost:9001
 hypnotoad :
   pid_file : <%= home %>/var/run/
   listen :
      - http://localhost:9001
 # The rest defines the servers, disks and buckets
 # used by the Yars cluster.  In this single server
 # example, there is only one server and one disk
 servers :
 - url : http://localhost:9001
   disks :
     - root : <%= home %>/var/data/disk1
       buckets : <%= json [ 0..9, 'a'..'f' ] %>

The configuration file is a Mojo::Template template with helpers provided by Clustericious::Config::Helpers.

Create the directories needed for the server:

 % mkdir -p ~/var/run ~/var/data

Now you can start the server process

 % yars start

check status

Now verify that it works:

 % curl http://localhost:9001/status

You can also verify that it works with yarsclient:

 % yarsclient status
 app_name: Yars
 server_hostname: iscah
 server_url: http://localhost:9001
 server_version: '1.11'

Or via Yars::Client:

 % perl -MYars::Client -MYAML::XS=Dump -E 'say Dump(Yars::Client->new->status)'
 app_name: Yars
 server_hostname: iscah
 server_url: http://localhost:9001
 server_version: '1.11'

upload and downloads

Now try storing a file:

 % echo "hi" | curl -D headers.txt -T - http://localhost:9001/file/test_file1
 % grep Location headers.txt 
 Location: http://localhost:9001/file/764efa883dda1e11db47671c4a3bbd9e/test_file1

You can use the Location header to fetch the file at a later time

 % curl http://localhost:9001/file/764efa883dda1e11db47671c4a3bbd9e/test_file1

With yarsclient

 % echo "hi" > test_file2
 % md5sum test_file2
 764efa883dda1e11db47671c4a3bbd9e  test_file2
 % yarsclient upload test_file2
 ... some time later ...
 % yarsclient downbload test_file2 764efa883dda1e11db47671c4a3bbd9e

You can see the HTTP requests and responses using the --trace option:

 % yarsclient --trace upload test_file2
 % yarsclient --trace download test_file2 764efa883dda1e11db47671c4a3bbd9e

And from Perl:

 use 5.010;
 use Yars::Client;
 use Digest::MD5 qw( md5_hex );
 my $y = Yars::Client->new;
 # filename as first argument,
 # reference to content as second argument
 $y->upload("test_file3", \"hi\n");
 # you can also skip the content like this:
 # $y->upload("test_file3");
 # to upload content from a local file
 my $md5 = md5_hex("hi\n");
 $y->download("test_file3", $md5);

Multiple servers

set up the URL

When configuring a cluster of several hosts, the url value in the configuration must have the correct hostname or IP address for each host that the server is running on. One way to handle this would be to have a configuration file for each host:

 # ~/etc/Yars.conf on yars1
 url: http://yars1:9001

 # ~/etc/Yars.conf on yars2
 url: http://yars2:9001

A less tedious way is to use the hostname or hostname_full helper from Clustericious::Config::Helpers. This allows you to use the same configuration for all servers in the cluster:

 # works for yars1, yars2 but not for
 # a client host
 url: http://<%= hostname %>:9001

abstract the webserver configuration

If you have multiple Clustericious services on the same host, or if you share configurations between multiple hosts, it may be useful to use the <%= extends_config %> helper and put the web server configuration in a separate file. For example:

 # ~/etc/Yars.conf
 % my $url = "http://" . hostname . ":9001";
 url: <%= $url %>
 % extends_config 'hypnotoad', url => $url, name => 'yars';

 # ~/etc/hypnotoad.conf
 hypnotoad :
   pid_file : <%= home %>/var/run/<%= $name %>.pid
   listen :
      - <%= $url %>

Now if you were also going to use PlugAuth on the same host they could share the same hypnotoad.conf file with different parameters:

 # ~/etc/PlugAuth.conf
 % my $url = "http://" . hostname . ":3001";
 url: <%= $url %>
 % extends_config 'hypnotoad', url => $url, name => 'plugauth';

generate the disk map

Given a file with a list of hosts and disks like this called diskmap.txt:

 yars1 /disk/1a
 yars1 /disk/1b
 yars2 /disk/2a
 yars2 /disk/2b
 yars3 /disk/3a
 yars3 /disk/3b

You can generate a disk map using the yars_generate_diskmap command:

 % yars_generate_diskmap 2 diskmap.txt > ~/etc/yars_diskmap.conf

This will generate a diskmap configuration with the buckets evenly allocated to the available disks:

 servers :
 - url : http://yars1:9001
   disks :
   - root : /disk/1a
     buckets : [ 00, 06, 0c, 12, 18, 1e, 24, 2a, 30, 36, 3c, 42, 48,
                 4e, 54, 5a, 60, 66, 6c, 72, 78, 7e, 84, 8a, 90, 96, 9c,
                 a2, a8, ae, b4, ba, c0, c6, cc, d2, d8, de, e4, ea, f0,
                 f6, fc ]
   - root : /disk/1b
     buckets : [ 01, 07, 0d, 13, 19, 1f, 25, 2b, 31, 37, 3d, 43, 49,
                 4f, 55, 5b, 61, 67, 6d, 73, 79, 7f, 85, 8b, 91, 97, 9d,
                 a3, a9, af, b5, bb, c1, c7, cd, d3, d9, df, e5, eb, f1,
                 f7, fd ]
 - url : http://yars2:9001
   disks :
   - root : /disk/2a
     buckets : [ 02, 08, 0e, 14, 1a, 20, 26, 2c, 32, 38, 3e, 44, 4a,
                 50, 56, 5c, 62, 68, 6e, 74, 7a, 80, 86, 8c, 92, 98, 9e,
                 a4, aa, b0, b6, bc, c2, c8, ce, d4, da, e0, e6, ec, f2,
                 f8, fe ]
   - root : /disk/2b
     buckets : [ 03, 09, 0f, 15, 1b, 21, 27, 2d, 33, 39, 3f, 45, 4b,
                 51, 57, 5d, 63, 69, 6f, 75, 7b, 81, 87, 8d, 93, 99, 9f,
                 a5, ab, b1, b7, bd, c3, c9, cf, d5, db, e1, e7, ed, f3,
                 f9, ff ]
 - url : http://yars3:9001
   disks :
   - root : /disk/3a
     buckets : [ 04, 0a, 10, 16, 1c, 22, 28, 2e, 34, 3a, 40, 46, 4c,
                 52, 58, 5e, 64, 6a, 70, 76, 7c, 82, 88, 8e, 94, 9a, a0,
                 a6, ac, b2, b8, be, c4, ca, d0, d6, dc, e2, e8, ee, f4,
                 fa ]
   - root : /disk/3b
     buckets : [ 05, 0b, 11, 17, 1d, 23, 29, 2f, 35, 3b, 41, 47, 4d,
                 53, 59, 5f, 65, 6b, 71, 77, 7d, 83, 89, 8f, 95, 9b, a1,
                 a7, ad, b3, b9, bf, c5, cb, d1, d7, dd, e3, e9, ef, f5,
                 fb ]

which you can now extend from the Yars.conf file:

 # ~/etc/Yars.conf
 % my $url = "http://" . hostname . ":9001";
 url: <%= $url %>
 % extends_config 'hypnotoad', url => $url, name => 'yars';
 % extends_config 'yars_diskmap';

Also, if for whatever reason you are unable to use the hostname or hostname_full helper in your Yars.conf file, it helps to keep your diskmap configuration in a separate file that can be shared between the different Yars server configuration files.

You can now run yars start on each host to start the servers. clad may be useful for starting "yars start" on multiple hosts at once.

client configuration

If you are using the hostname or hostname_full helpers to generate the URL in the serve configuration, then you won't be able to share that configuration with client systems. In addition you can specify one or more failover hosts for Yars::Client and yarsclient to use when the primary is not available:

 # ~/etc/Yars.conf on client systems
 url: http://yars2:9001
   - http://yars1:9001

randomizing the server choices

In order to more evenly spread the load over each node in the Yars cluster, you can randomize the servers that the client considers the "primary" and the "failover(s)":

 # ~/etc/Yars.conf on client systems
 % use List::Util qw( shuffle );
 % my @url = shuffle map { "http://yars$_:9001" } 1..3;
 url: <%= $url[0] %>
   - <%= $url[1] %>

Accelerated downloads with nginx

One of the advantages of Clustericious is that it integrates with a number of different webservers. You can do testing with hypnotoad, which comes with Mojolicious (and thus a prerequisite of Clustericious and Yars), and then deploy to production with a more capable webserver, such as nginx. The integration with nginx allows for handing off some of the workload to nginx; hypnotoad is good for serving dynamic web applications, but nginx is better for serving static files. So with this next configuration we will show you how to configure Yars to handle the selection of servers and disks and hand off the actual serving of the static file to nginx.

Once again we put the nginx configuration in its own file so that we can reuse it with other Clustericious services.

 # ~/etc/nginx.conf
   - hypnotoad
   - nginx
 # we use hypnotoad to server the dynamic part of the app
 # and listen to the same port on localhost
     -<%= $port %>
   pid_file: <%= home %>/var/run/<%= $name %>
   proxy: 1
 # and we proxy requests on the main IP address through
 # nginx
   '-p': <%= home %>/var/run/<%= $name %>-nginx
   '-c': <%= home %>/var/run/<%= $name %>-nginx/conf/nginx.conf
     filename: <%= home %>/var/run/<%= $name %>-nginx/conf/nginx.conf
     content: |
         # autogenerated file
         events {
           worker_connections 4096;
         http {
           server {
             listen <%= hostname %>:<%= $port %>;
             location / {
               proxy_pass<%= $port %>;
               proxy_http_version 1.1;
 % if($name eq 'yars') {
               # to accelerate downloads, for Yars only
               # we set the X-Yars-Use-X-Accel header to
               # any value.  This will trigger Yars to
               # use nginx's X-Accel-Redirect to serve
               # actual static files back to the client.
               proxy_set_header X-Yars-Use-X-Accel yes;
 % }
 % if($name eq 'yars') {
             # we need to make the static files available
             # to nginx.  The /static prefix is to ensure
             # that routes (future and present) do not
             # conflict with physical files on your disk.
             location /static/disk/ {
               # internal makes sure that these files
               # won't be served to external clients
               # without going through the yars interface
               alias /disk/;
 % }

and once again, our Yars.conf file is short and sweet:

 % my $port = 9001;
 url: http://<%= hostname %>:<%= $port %>
 % extends_config 'nginx', port => $port, name => 'yars';
 % extends_config 'yars_diskmap';

If you are storing large files in your Yars cluster the nginx default maximum request size will probably not be adequate. If you see an error message like this:

 % yarsclient upload large-file.iso
 [ERROR] 2016/09/30 11:23:48 (204) (413) Request Entity Too Large

Then you need to set client_max_body_size appropriately.

 http {
   server {
     client_max_body_size 0; # zero for no max

Accelerate by not checking the md5 twice

By default, Yars checks the MD5 of files before serving them to the client. Yars::Client and yarsclient both also check the MD5 sum after downloading. This saves bandwidth if automated processes attempt to redownload the same file if it is corrupted on the disk of the server. The chance of error is likely much higher on the network than it is on the disk, and if you prefer to do the check just on the client side, then you can use set the download_md5_verify to zero.

 % my $port = 9001;
 url: http://<%= hostname %>:<%= $port %>
 % extends_config 'nginx', port => $port, name => 'yars';
 % extends_config 'yars_diskmap';
 download_md5_verify: 0

When you download files with other clients like curl or wget, the MD5 check will still happen on the server side. You may request this check be skipped by setting the X-Yars-Skip-Verify header to any value.


Thanks to Brian Duggan (BDUGGAN) for doing most of the initial work on Yars, and David Golden (XDG, DAGOLDEN) for describing Yars strength as "Write availability and eventual read consistency and availability".



Perl API interface to Yars.


Command line client interface to Yars.


HTTP REST routes useable for interfacing with Yars.


Automated upload / download of files to Yars for performance testing.


Yars is built on the Clustericious framework, itself heavily utilizing Mojolicious


Original author: Marty Brandon

Current maintainer: Graham Ollis <>


Brian Duggan

Curt Tilmes


This software is copyright (c) 2013 by NASA GSFC.

This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.

syntax highlighting: