Merge pull request #32 from basho-labs/2.0.0 2.0.0
Upgrade Riak TS version to 1.5.1
Merge pull request #31 from basho-labs/2.0.0-rc2 2.0.0 rc2
- IMPORTANT Persistent volumes/Data dir: the path for this is now
$(executor-dir)/data, to avoid the contents of any of the artifacts over-writing the symlink Mesos creates for the persistent volume inside the executor's directory. The Executor is currently hard-coded to use this path. In future versions, this will be configurable. If you need to use this path in your custom Riak config, it is accessible via the
- Riak config: No longer always require a
riak.confto be set for Riak TS clusters: instead, no default
riak.confis provided and we use the one that ships with the Riak you install. If you wish to customise the config that is used for your cluster, you can still do so in the same way (instructions here), but be aware that it is necessary for the executor to override the configuration items contained in this file.
- We now only supply artifacts built using
Mesos-1.0.0protobuf definitions: these are fully backwards compatible with older Mesos versions, so there is no need for the e.g.
- Derive the
rolevalues from the
framework-nameby default. These can be set from your
config.jsonbut it is not necessary to do so.
Full change details:
- #91 Inspect the riak archive for the path under which it will extract, and pass that to the executor.
- #89 new default values for
.riak.framework-nameif unset in
- #84 #87 : Rework how config files for Riak are managed: no longer do we automatically set a default config file, and we allow the config files to be deleted once set.
Note that for this to be compatible with Executor versions before
1.5.0, you MUST set a config and advanced-config file for your clusters.
IMPORTANT: if you upgrade to this release and use a custom configuration, you must change your
riak.confso that it sets
platform_data_dir = ../../data. If you use the default config that ships with the scheduler, this has been updated already.
- #85 Move persistent volume path; ensure no archive contents clobber PV path
If you are running a Riak cluster on Mesos with a previous version of this scheduler and want to update, do not simply update the
config.json, reinstall the scheduler and perform a
riak-mesos cluster restart: this will lead to data loss.
Instead, you will need to update the
config.jsonand reinstall the scheduler, then slowly migrate your cluster to new nodes, allowing transfers to finish before removing each old node.
For a much more in-depth write-up of the procedure, see here: https://gist.github.com/sanmiguel/aa06f1e6026601677af2ced87c6a35f9#file-live-upgrade-procedure-sh
If in doubt, please contact us via the issues.
- #81 Adds backwards compatible upgrade to
- #80 Add CONTRIBUTING.md: an initial guide to working with the code in this project
A note on
erl_mesos: the upgrade here extends the underlying protobuf definitions to those from
mesos-1.0.0, but does not yet use any
1.0.0-specific functionality. This release should be compatible with earlier versions of mesos without problem: if you encounter an issue, please report it.
- #82 Fix order of archive extraction in executor
- #34 New supported variable from
RiakRootPath- the location to look for Riak config files under
- #34 Add
erlpmd_patches-relvariant, to work with official
- #34 Automatically calculate the
- #33 Upgrade to erl_mesos
v1.0.0which uses mesos-1.0.0 protobufs (backwards compatible with mesos-0.28.x)
- #31 No longer expect the Scheduler to have a configuration file with all the right template variables in the right places. Now, we either use the config file that shipped with our Riak node, or one from the Scheduler if it's set. Once that's in place, we append
riak-mesos.confto the end, which contains all of the settings we MUST control in a Mesos environment (ports, data location etc)
v1.0.0 of the Riak Mesos Framework