Yars - Yet Another RESTful-Archive Service
Yars is a simple RESTful server for data storage.
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.
The following sequence of commands will start yars on a single host (with 16 buckets) :
$ mkdir ~/etc $ cat > ~/etc/Yars.conf --- start_mode : 'hypnotoad' url : http://localhost:9999 hypnotoad : pid_file : /tmp/yars.pid listen : - http://localhost:9999 servers : - url : http://localhost:9999 disks : - root : /usr/local/data/disk1 buckets : [ <%= join ',', '0'..'f' %> ] ^D $ yars start
Now, verify that it works :
And try to PUT and GET a file :
echo "hi" | lwp-request -em PUT http://localhost:9999/file/here # (notice the "Location" header GET http://localhost:9999/file/764efa883dda1e11db47671c4a3bbd9e/here
Also you can use Yars::Client :
echo "hi" > myfile yarsclient upload myfile yarsclient download myfile 764efa883dda1e11db47671c4a3bbd9e
Or to see the requests and responses :
yarsclient --trace root upload myfile yarsclient --trace root download myfile 764efa883dda1e11db47671c4a3bbd9e
To install Yars on a cluster of several hosts, the configuration for each host should be identical, except that the 'url' should reflect the host on which the server is running.
To accomplish this, the above configuration may be divided into two files, one with the bucket map, and another with the server specific information.
yars1 ~$ cat > ~/etc/Yars.conf : ---- extends_config 'disk_map'; url : http://yars1:9999 hypnotoad : pid_file : /tmp/yars.pid listen : - http://yars1:9999 yars2 ~$ cat > ~/etc/Yars.conf : ---- extends_config 'disk_map'; url : http://yars2:9999 hypnotoad : pid_file : /tmp/yars.pid listen : - http://yars2:9999 Then on both servers : $ cat > ~/etc/disk_map.conf : servers : - url : http://yars1:9999 disks : - root : /usr/local/data/disk1 buckets : [ <%= join ',', '0'..'9' %> ] - url : http://yars2:9999 disks : - root : /usr/local/data/disk1 buckets : [ <%= join ',', 'a'..'f' %> ]
Then run "yars start" on both servers and voila, you have an archive.
See also, clad, for a tool to facilitate running "yars start" on multiple hosts at once.
Called by the server to start up, we change the object classes to use Yars::Message::Request for incoming requests.
Original author: Marty Brandon
Current maintainer: Graham Ollis <firstname.lastname@example.org>
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.