Published on August 1st, 2016 📆 | 5727 Views ⚑0
orgalorg — Parallel SSH Command Execution
orgalorg provides easy way of synchronizing files across cluster and running arbitrary SSH commands. orgalorg works through SSH & tar, so no unexpected protocol errors will arise. In default mode of operation (lately referred as sync mode) orgalorg will perform steps in the following order:
- Acquire global cluster lock (check more detailed info above).
- Create, upload and extract specified files in streaming mode to the specified nodes into temporary run directory.
- Start synchronization tool on each node, that should relocate files from temporary run directory to the destination.
So, orgalorg expected to work with third-party synchronization tool, that will do actual files relocation and can be quite intricate, but orgalorg can work without that tool and perform simple files sync (more on this later).
- Zero-configuration. No config files. Everything is done via command line flags.
- Running SSH commands or shell scripts on any number of hosts in parallel. All output from nodes will be returned back, keeping stdout and stderr streams mapping of original commands.
- Synchronizing files and directories across cluster with prior global cluster locking. After synchronization is done, arbitrary command can be evaluated.
- Synchronizing files and directories with subsequent run of complex multi-step scenario with steps synchronization across cluster.
- User-friendly progress indication.
- Both strict or loose modes of failover to be sure that everything will either fail on any error or try to complete, no matter of what.
- Interactive password authentication as well as SSH public key authentication.
- Ability to run commands through
- Grouped mode of output, so stdout and stderr from nodes will be grouped by node name. Alternatively, output can be returned as soon as node returns something.
go get github.com/reconquest/orgalorg
Before doing anything else orgalorg will perform global cluster lock. That lock is acquired atomically, and no other orgalorg instance can acquire lock if it is already acquired. Locking is done via flock’ing specified file or directory on each of target nodes, and will fail, if flock fails on at least one node. Directory can be used as lock target as well as ordinary file.
--lock-file can be used to specify lock target different from
After acquiring lock, orgalorg will run heartbeat process, which will check, that lock is still intact. By default, that check will be performed every 10 seconds. If at least one heartbeat is failed, then orgalorg will abort entire sync procedure. User can stop there by using
-L flag, effectively transform orgalorg to the distributed locking tool.
-o <host>... in later examples will mean any supported combination of host-specification arguments, like
-o node1.example.com -o node2.example.com.
Evaluating command on hosts in parallel
orgalorg -o <host>... -C uptime
Evaluating command on hosts given by stdin
axfr is a tool of your choice for retrieving domain information from your infrastructure DNS.
axfr | grep phpnode | orgalorg -s -C uptime
Evaluate command under root (passwordless sudo required)
orgalorg -o <host>... -x -C whoami
Tailing logs from many hosts in realtime
orgalorg -o <host>... -C tail -f /var/log/syslog
Copying SSH public key for remote authentication
orgalorg -o <host>... -p -i ~/.ssh/id_rsa.pub -C tee -a ~/.ssh/authorized_keys
Synchronizing configs and then reloading service (like nginx)
orgalorg -o <host>... -xn 'systemctl reload nginx' -S /etc/nginx.conf
Evaluating shell script
orgalorg -o <host>... -i script.bash -C bash
Install package on all nodes and get combined output from each node
orgalorg -o <host>... -lx -C pacman -Sy my-package --noconfirm
Evaluating shell oneliner
orgalorg -o <host>... -C sleep '$(($RANDOM % 10))' '&&' echo done
Running poor-man interactive parallel shell
orgalorg -o <host>... -i /dev/stdin -C bash -s
Obtaining global cluster lock
orgalorg -o <host>... -L
Next orgalorg calls will fail with message, that lock is already acquired, until first instance will be stopped.
Useful for setting cluster into maintenance state.
Obtaining global cluster lock on custom directory
orgalorg -o <host>... -L -r /etc
Files will be sent from local node to the amount of specified nodes.
orgalorg will perform streaming transfer, so it’s safe to synchronize large files without major memory consumption.
By default, orgalorg will upload files to the temporary run directory. That behaviour can be changed by using
-rflag. Then, files will be uploaded to the specified directory.
User can specify
-U flag to transform orgalorg to the simple file upload tool. In that mode orgalorg will upload files to the specified directory and then exit.
orgalorg preserves all file attributes while tranfer as well as user and group IDs. That behaviour can be changed by using
--no-preseve-gid command line options.
By default, orgalorg will keep source file paths as is, creating same directory layout on the target nodes. E.g., if orgalorg told to upload file
a while current working directory is
/b/c/, orgalorg will upload file to the
<root>/b/c/a on the remote nodes. That behaviour can be changed by specifying
-e flag. Then, orgalorg will not preserve source file base directory.
orgalorg will try to upload files under specified user (current user by default). However, if user has
NOPASSWD record in the sudoers file on the remote nodes,
-x can be used to elevate to root before uploading files. It makes possible to login to the remote nodes under normal user and rewrite system files.
After file upload orgalorg will execute synchronization tool (
/usr/lib/orgalorg/sync). That tool is expected to relocate synced files from temporary directory to the target directory. However, that tool can perform arbitrary actions, like reloading system services.
To specify custom synchronization tool user can use
-n flag. Full shell syntax is supported in the argument to that option.
Tool is also expected to communicate with orgalorg using sync protocol (described below), however, it’s not required. If not specified, orgalorg will communicate with that tool using stdin/stdout streams. User can change that behaviour using
-m flag, which will cause orgalorg to treat specified sync tool as simple shell command. User can even provide stdin to that program by using
Tool can accept number of arguments, which can be specified by using
orgalorg will communicate with given sync tool using special sync protocol, which gives possibility to perform some actions with synchronization across entire cluster. orgalorg will start sync tool as it specified in the command line, without any modification. After start, orgalorg will communicate with running sync tool using stdin and stdout streams. stderr will be passed to user untouched. All communication messages should be prefixed by special prefix, which is send by orgalorg in the hello message. All lines on stdout that are not match given prefix will be printed as is, untouched. Communcation begins from the hello message.
orgalorg -> sync tool
Start communication session. All further messages should be prefixed with given prefix.
orgalorg -> sync tool
<prefix> NODE <node>
orgalorg will send node list to the sync tools on each running node.
orgalorg -> sync tool
Start messages will be sent at the end of the nodes list and means that sync tool can start doing actions.
sync tool -> orgalorg
<prefix> SYNC <description>
Sync tool can send sync messages after some steps are done to be sure, that every node in cluster are performing steps gradually, in order.
When orgalorg receives sync message, it will be broadcasted to every connected sync tool.