Skip to content

LogCabin is a distributed storage system built on Raft that provides a small amount of highly replicated, consistent storage. It is a reliable place for other distributed systems to store their core metadata and is helpful in solving cluster management issues.

License

Notifications You must be signed in to change notification settings

mongodb-labs/logcabin

 
 

Repository files navigation

logo

Overview

LogCabin is a distributed system that provides a small amount of highly replicated, consistent storage. It is a reliable place for other distributed systems to store their core metadata and is helpful in solving cluster management issues. LogCabin uses the Raft consensus algorithm internally and is actually the very first implementation of Raft. It's released under the ISC license (equivalent to BSD).

External resources:

Information about releases is in RELEASES.md.

This README will walk you through how to compile and run LogCabin.

Questions

The best place to ask questions about the LogCabin implementation is on the logcabin-dev mailing list. You might also try #logcabin on the freenode IRC network, although there aren't always people around. Use GitHub Issues to report problems or suggest features.

For questions and discussion about the Raft consensus algorithm, which LogCabin implements, use the raft-dev mailing list.

Building

Build Status

Pre-requisites:

  • Linux x86-64 (v2.6.32 and up should work)
  • git (v1.7 and up should work)
  • scons (v2.0 and v2.3 are known to work)
  • g++ (v4.4 through v4.9 and v5.1 are known to work) or clang (v3.4 through v3.7 are known to work with libstdc++ 4.9, and libc++ is also supported; see CLANG.md for more info)
  • protobuf (v2.6.x suggested, v2.5.x should work, v2.3.x is not supported)
  • crypto++ (v5.6.1 is known to work)
  • doxygen (optional; v1.8.8 is known to work)

In short, RHEL/CentOS 6 should work, as well as anything more recent.

Get the source code:

git clone git://github.com/logcabin/logcabin.git
cd logcabin
git submodule update --init

Build the client library, server binary, and unit tests:

scons

For custom build environments, you can place your configuration variables in Local.sc. For example, that file might look like:

BUILDTYPE='DEBUG'
CXXFLAGS=['-Wno-error']

To see which configuration parameters are available, run:

scons --help

Running basic tests

It's a good idea to run the included unit tests before proceeding:

build/test/test

You can also run some system-wide tests. This first command runs the smoke tests against an in-memory database that is embedded into the LogCabin client (no servers are involved):

build/Examples/SmokeTest --mock && echo 'Smoke test completed successfully'

To run the same smoke test against a real LogCabin cluster will take some more setup.

Running a real cluster

This section shows you how to run the HelloWorld example program against a three-server LogCabin cluster. We'll run all the servers on localhost for now:

  • Server 1 will listen on 127.0.0.1:5254
  • Server 2 will listen on 127.0.0.1:5255
  • Server 3 will listen on 127.0.0.1:5256

Port 5254 is LogCabin's default port and is reserved by IANA for LogCabin. The other two belong to others and are hopefully not in use on your network.

We'll first need to create three configuration files. You can base yours off of sample.conf, or the following will work for now:

File logcabin-1.conf:

serverId = 1
listenAddresses = 127.0.0.1:5254

File logcabin-2.conf:

serverId = 2
listenAddresses = 127.0.0.1:5255

File logcabin-3.conf:

serverId = 3
listenAddresses = 127.0.0.1:5256

Now you're almost ready to start the servers. First, initialize one of the server's logs with a cluster membership configuration that contains just itself:

build/LogCabin --config logcabin-1.conf --bootstrap

The server with ID 1 will now have a valid cluster membership configuration in its log. At this point, there's only 1 server in the cluster, so only 1 vote is needed: it'll be able to elect itself leader and commit new entries. We can now start this server (leave it running):

build/LogCabin --config logcabin-1.conf

We don't want to stop here, though, because the cluster isn't fault-tolerant with just one server! We're going to start two more servers and then add them both to the first server's cluster.

Let's start up the second server in another terminal (leave it running):

build/LogCabin --config logcabin-2.conf

Note how this server is just idling, awaiting a cluster membership configuration. It's still not part of the cluster.

Start the third server also (LogCabin checks to make sure all the servers in your new configuration are available before committing to switch to it, just to keep you from doing anything stupid):

build/LogCabin --config logcabin-3.conf

Now use the reconfiguration command to add the second and third servers to the cluster:

ALLSERVERS=127.0.0.1:5254,127.0.0.1:5255,127.0.0.1:5256
build/Examples/Reconfigure --cluster=$ALLSERVERS set 127.0.0.1:5254 127.0.0.1:5255 127.0.0.1:5256

This Reconfigure command is a special LogCabin client. It first queries each of the servers given in its positional command line arguments (space-delimited after the command "set") to retrieve their server IDs and listening addresses (as set in their configuration files). Then, it connects to the cluster given by the --cluster option (comma-delimited) and asks the leader to set the cluster membership to consist of those servers. Note that the existing cluster members should be included in the positional arguments if they are to remain in the cluster; otherwise, they will be evicted from the cluster.

If this succeeded, you should see that the first server has added the others to the cluster, and the second and third servers are now participating. It should have output something like:

Current configuration:
Configuration 1:
- 1: 127.0.0.1:5254

Attempting to change cluster membership to the following:
1: 127.0.0.1:5254 (given as 127.0.0.1:5254)
2: 127.0.0.1:5255 (given as 127.0.0.1:5255)
3: 127.0.0.1:5256 (given as 127.0.0.1:5256)

Membership change result: OK

Current configuration:
Configuration 4:
- 1: 127.0.0.1:5254
- 2: 127.0.0.1:5255
- 3: 127.0.0.1:5256

Note: If you're sharing a single magnetic disk under heavy load for all the servers, the cluster may have trouble maintaining a leader. See issue 57 for more details on symptoms and a workaround.

Finally, you can run a LogCabin client to exercise the cluster:

build/Examples/HelloWorld --cluster=$ALLSERVERS

That program doesn't do anything very interesting. Another tool called TreeOps exposes LogCabin's data structure on the command line:

echo -n hello | build/Examples/TreeOps --cluster=$ALLSERVERS write /world
build/Examples/TreeOps --cluster=$ALLSERVERS dump

See the --help for a complete listing of the available commands.

You should be able to kill one server at a time and maintain availability, or kill more and restart them and maintain safety (with an availability hiccup).

If you find it annoying to pass --cluster=$ALLSERVERS everywhere, you can also use a DNS name to return all the IP addresses. However, you will need distinct IP addresses for each server, not just distinct ports.

If you have your own application, you can link it against build/liblogcabin.a. You'll also need to link against the following libraries:

  • pthread
  • protobuf
  • cryptopp

Running cluster-wide tests

The procedure described above for running a cluster is fairly tedious when you just want to run some tests and tear everything down again. Thus, scripts/smoketest.py automates it. Create a file called scripts/localconfig.py to override the smokehosts and hosts variables found in scripts/config.py:

smokehosts = hosts = [
  ('192.168.2.1', '192.168.2.1', 1),
  ('192.168.2.2', '192.168.2.2', 2),
  ('192.168.2.3', '192.168.2.3', 3),
]

The scripts use this file to when launching servers using SSH. Each tuple in the (smoke)hosts list represents one server, containing:

  1. the address to use for SSH,
  2. the address to use for LogCabin TCP connections, and
  3. a unique ID.

Each of these servers should be accessible over SSH without a password and should have the LogCabin directory available in the same filesystem location. The script currently assumes this directory to be on a shared filesystem, such as an NFS mount or localhost.

You may optionally create a smoketest.conf file, which can define various options that apply to all the servers. The servers' listen addresses will be merged with your smoketest.conf automatically.

Now you're ready to run:

scripts/smoketest.py && echo 'Smoke test completed successfully'

This script can also be hijacked/included to run other test programs.

Documentation

To build the documentation from the source code, run:

scons docs

The resulting HTML files will be placed in docs/doxygen.

You can also find this documentation at https://logcabin.github.io.

Installation

To install a bunch of things on your filesystem, run:

scons install

Along with the binaries, this installs a RHEL 6-compatible init script.

If you don't want these files to pollute your filesystem, you can install the files to any given directory as follows (replace pathtoinstallprefix in both places with wherever you'd like the files to go):

scons --install-sandbox=pathtoinstallprefix pathtoinstallprefix

Finally, you can build a binary RPM as follows:

scons rpm

This creates a file called build/logcabin-0.0.1-0.1.alpha.0.x86_64.rpm or similar that you can then install using RPM, with the same effect as scons install.

Contributing

Please use GitHub to report issues and send pull requests.

All commits should pass the pre-commit hooks. Enable them to run before each commit:

ln -s ../../hooks/pre-commit .git/hooks/pre-commit

About

LogCabin is a distributed storage system built on Raft that provides a small amount of highly replicated, consistent storage. It is a reliable place for other distributed systems to store their core metadata and is helpful in solving cluster management issues.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C++ 91.5%
  • Python 4.8%
  • Protocol Buffer 3.0%
  • Other 0.7%