Permalink
Fetching contributors…
Cannot retrieve contributors at this time
270 lines (198 sloc) 9.57 KB

Migrating to Riak 0.11 from earlier versions of Riak

Riak 0.10 to 0.11

Customized configuration, not switching backends

If you have customized your configuration file (etc/app.config), and you will be keeping your customized configuration file, your transition should be as easy as stopping the 0.10 node, installing the 0.11 binaries, and starting the node back up. The node should find its ring and data in the same location as before, and will continue working as if nothing changed.

Default configuration, or switching backends

If you’re using the default Riak configuration file, you’ll need to be aware that the default storage backend has changed from DETS to Basho’s new Bitcask. You’ll need to choose one of two upgrade paths in order to migrate your data from your old storage backend to the new one (these steps will also work, in general, from any backend to any other backend). Your options are: backup/restore, and online rolling upgrade.

Backup/Restore

The easiest way to migrate from one storage backend to another is to backup the cluster as it stands, shut it down, switch the backend (by upgrading or editing the config file), start the cluster back up, and then restore the data to the cluster. Follow these steps to do this:

  1. With a running cluster, use the “riak-admin backup” command to backup the cluster’s data:

    $ riak-admin backup riak@127.0.0.1 riak riak-backup.dat all

    This script will exit when backup is complete. If your cluster has a lot of data, you can backup each node individually by switching “all” to “node” in the command line above, and then running the command once for each node, changing the node name in the command line.

  2. Shut down the cluster.

    $ riak stop

    Run this command on each node.

  3. Upgrade each node by installing the new 0.11 package. If you’re just switching backends (not upgrading), modify the configuration file in this step.
  4. Start the cluster

    $ riak start

    Run this command on each node. The nodes will automatically reconnect to each other.

  5. Reload the cluster’s data

    $ riak-admin restore riak@127.0.0.1 riak riak-backup.dat

    If you used the whole-cluster (“all”) backup mode, you’re now finished. If you backed up each node individually, you’ll need to run this command for each of those backup files (note that they have node names appended to what you provided on the backup command line).

Once the restore script exits, all data should be restored to the cluster. After checking that this is true, you may remove the data directories for the old storage backend.

Online Rolling Upgrade

If you need your Riak cluster to remain running during the backend transition, and you have spare network capacity, you can use Riak’s easy node addition/removal process to switch backends.

  1. Configure a new node, with the new backend you want to use, and add it to the cluster.

    newnode$ riak start newnode$ riak-admin join oldclusternode@10.0.0.100

  2. Remove one of the old-backend nodes from the cluster.

    oldnode0$ riak-admin leave

  3. When the old-backend node completes handoff, it will exit (use ‘ps’ or similar to watch for it to shut down). Once it has exited, upgrade to 0.11 (or switch the backend) and restart the node.
  4. Repeat steps 2 and 3 for each other old-backend node in the cluster.
  5. When all nodes are running on the new backend, shutdown the new node you set up in step 1 (if you wish, or leave it up if you like).

Note: You can skip step 1 if you are running a cluster of more than one node, and your cluster can tolerate (capacity- and throughput-wise) being temporarily one node smaller. Step 1 is merely an attempt to add extra capacity to the system before beging an expensive operation.

Riak 0.9 and earlier to 0.11

If you are upgrading from a Riak version earlier than 0.10, your only option for upgrade is backup and restore. Please use the backup/restore method described in the 0.10-to-0.11-transition section above.

You will also be interested in the following section, which discusses migration from 0.9 to 0.10.

Migrating from Riak 0.9.x to 0.10

Overview

Riak has undergone significant restructuring in the transition from version 0.9.x to version 0.10. If you are using the binary builds, please skip directly to the Configuration and Clients sections. If you are building from source yourself, please review the whole of this document.

NOTE: If the only files you have changed in your Riak source clone are those underneath the “rel” directory (“rel/overlay/etc/app.config”, for example), the safest way to update is to make a fresh clone of the Riak repository, and then skip to the Configuration and Clients sections of this document for details about migrating your configurations and data.

Requirements

Erlang/OTP R13B04

Riak 0.10 uses new features (“NIFs”) provided by the latest Erlang/OTP release, R13B04. If you are building from source, you will need this release or a newer one.

Mercurial

Riak 0.10 has moved several of its components into external repositories. If you are building from source, you will need Mercurial installed to allow the Rebar build system to retrieve code from these external repositories.

Dependencies

Mochiweb, Webmachine, Erlang_js

mochiweb, webmachine, and erlang_js are now pulled into the “deps” subdirectory, instead of being included in the “apps” subdirectory. If you are pulling 0.10 code into a repository that formerly had 0.9.x in it, please remove the apps/mochiweb, apps/webmachine, and apps/erlang_js directories from your source tree.

There is a chance that your update will also leave an “apps/riak” directory hanging around. If it does, please remove this directory (Riak code has moved into the “apps/riak_core” and “apps/riak_kv” directories).

make deps

The “all” make target (and, by extension, the “rel” target as well), depend on a new “deps” target, which handles the fetching the dependencies (mochiweb, webmachine, erlang_js).

Source

Core/KV Split

We’ve drawn a line through 0.9.x Riak, and divided it into two things, one called “riak_core”, and the other called “riak_kv”.

The things that live in riak_core are those that deal with cluster membership. Ring-claiming and the like.

The things that live in riak_kv are those that deal with storing data. Get and Put FSMs, backends, etc.

Clients

We’ve also moved the clients out of the client_lib subdirectory, and into their own language-specific repositories on BitBucket. At http://bitbucket.org/basho/, you should find:

  • riak-python-client
  • riak-php-client
  • riak-erlang-client
  • riak-java-client
  • riak-javascript-client
  • riak-ruby-client

Configuration

app.config

Splitting the “riak” Erlang application into the “riak_core” and “riak_kv” Erlang applications means that configuration options for each component need to move around in etc/app.config.

Where before etc/app.config would have contained a section like:

{riak, [ %% many settings here ]},

Now, etc/app.config should contain two sections like:

{riak_core, [ %% core-specific settings ]}, {riak_kv, [ %% kv-specific settings ]},

The list of settings that moved to the riak_core section are:

  • choose_claim_fun
  • cluster_name - string, defaults to “default”
  • default_bucket_props
  • gossip_interval - integer, defaults to 60k msec
  • ring_creation_size - integer, defaults to 64
  • ring_state_dir - string
  • target_n_val - integer, defaults to 3
  • wants_claim_fun
  • web_ip - string. Used to be “riak_web_ip”
  • web_logdir - string.
  • web_port - integer. Used to be “riak_web_port”

IMPORTANT: Note the rename of “riak_web_*” to just “web_*”

The list of settings that moved to the riak_kv section are:

  • add_paths - list, defaults to []
  • handoff_concurrency - integer, defaults to 4
  • js_source_dir - string
  • js_vm_count - integer
  • mapred_name - string
  • raw_name - string
  • riak_kv_stat - boolean.
  • stats_urlpath - string
  • storage_backend - atom. Backend names are now prefixed as “riak_kv_” instead of just “riak_”.
  • pb_ip - string
  • pb_port - integer

IMPORTANT: The default backend has changed names from riak_dets_backend to riak_kv_dets_bakend. Other backends have changed names as well. This rename does not affect you if you are using the Innostore backend.

If you did not have any of these settings defined in etc/app.config, you still do not need to define them in your new etc/app.config.

Ring Storage

Periodically, Riak nodes save the state of their ring to disk. In 0.9, these files were named “data/ring/riak_ring.*”, but in 0.10, they’re named “data/ring/riak_core_ring.*”. Renaming the old files to the new scheme is all you need to do to make the switch.

If you referenced any Riak modules in your bucket properties, you will also need to change those references to point to the new module names after your cluster is running.

Your Data

The rest of your cluster’s data, stored in the “data” directory (“data/dets” or “data/innodb”, for example) should be safe to either leave in place, or copy to your new install location, depending on how you upgraded.