Implement Couchbase cluster provisioning

Implement Couchbase cluster provisioning.

Launchpad Blueprint: https://blueprints.launchpad.net/trove/+spec/couchbase-cluster-support

Problem Description

Couchbase is by nature a distributed database. Documents are distributed evenly across all nodes based on the hash of the document key. There are no query routers or configuration servers like in some other databases (MongoDB). Each document may have up to three replicas in addition to the active copy. What this means is that there will be three extra copies of that data within the cluster. Couchbase clients are provided with an updated ‘cluster map’ which they use to lookup data.

When a node gets added to the cluster the existing documents are incrementally redistributed within the cluster (process also known as rebalancing). Updated cluster maps are continuously provided to the clients. Any active client connections should not be interrupted throughout the process.

When a node gets removed from the cluster the replicas on the other nodes are promoted to the active and cluster maps updated so that the clients can keep serving the requests. The promoted documents are then re-replicated within the remaining nodes (rebalanced).

See References section for further details.

Proposed Change

This specification proposes the following cluster-related actions:

  • create
  • grow
  • shrink
  • delete
  • upgrade

Couchbase imposes cluster wide quota on the memory that get evenly distributed between node services. All nodes (including future nodes) need to be able to accommodate this quota. We therefore require the cluster to be homogeneous. The quota size is configurable by ‘cluster_ramsize_pc’ Trove configuration as the percentage of the total node RAM.

Create Cluster

  • Provision a requested number of instances while initializing them as Couchbase nodes (all nodes receive the same randomly generated Administrative credentials from the API).
  • Wait for the instances to become active.
  • Execute cluster initialization on the first node (coordinator).
  • Add the remaining nodes to the cluster via the coordinator node.
  • Wait for the cluster rebalance to finish.

Grow Cluster

  • Retrieve the current Administrative credentials from the first node (coordinator) of the existing cluster.
  • Provision and initialize a requested number of instances as Couchbase nodes with the current credentials (above).
  • Wait for the instances to become active.
  • Add the new nodes to the cluster via the coordinator node.
  • Wait for the cluster rebalance to finish.

Shrink Cluster

  • Load the removed instances.
  • Select one of the remaining nodes and use it as the coordinator.
  • Remove the nodes from the cluster via the coordinator.
  • Wait for the cluster rebalance to finish.
  • Delete decommissioned instances.

Configuration

The following configuration values will be implemented in the Couchbase configuration group:

cfg.BoolOpt('cluster_support', default=True,
            help='Enable clusters to be created and managed.'),
cfg.StrOpt('api_strategy',
           default='trove.common.strategies.cluster.experimental.'
           'couchbase.api.CouchbaseAPIStrategy',
           help='Class that implements datastore-specific API logic.'),
cfg.StrOpt('taskmanager_strategy',
           default='trove.common.strategies.cluster.experimental'
           '.couchbase.taskmanager.CouchbaseTaskManagerStrategy',
           help='Class that implements datastore-specific task manager '
                'logic.'),
cfg.StrOpt('guestagent_strategy',
           default='trove.common.strategies.cluster.experimental'
           '.couchbase.guestagent.CouchbaseGuestAgentStrategy',
           help='Class that implements datastore-specific Guest Agent API '
                'logic.'),
cfg.IntOpt('cluster_ramsize_pc', default=80, min=0, max=80,
           help='Per node RAM quota in for the Data service expressed as a'
           ' percentage of the available memory.'
           ' Minimum of 256MB will be used if the given percentage amounts'
           ' for less.'),

Database

None

Public API

None

Public API Security

None

Python API

None

CLI (python-troveclient)

None

Internal API

None

Guest Agent

The work will require some preliminary refactoring of the guest manager. This may also include tasks such that:

  • providing a common framework for executing Couchbase CLI calls The guest-agent currently executes only two CLI commands hardcoded in ‘system.py’ file. We will need to execute a greater variety of commands. This would include CouchbaseAdmin to execute CLI calls and handle results in a generic way. It will also facilitate reducing ‘system.py’ which we have been moving away from in other guests as well.
  • making the ramsize quota relative to the node’s total RAM Couchbase 3 requires a parameter to cluster-init is RAMSIZE quota. The valid value is between 256MB and 80% of available RAM. Make the default % a Trove variable.
  • using appropriate guestagent models for the Couchbase administrative user Couchbase should be using it’s own User model rather than relying on generic MySQL-object.

The following methods will be implemented in the CouchbaseGuestAgentAPI:

def initialize_cluster(self):
    LOG.debug("Configuring cluster parameters via node: %s" % self.id)
    self._call("initialize_cluster",
               guest_api.AGENT_HIGH_TIMEOUT, self.version_cap)

def get_cluster_password(self):
    LOG.debug("Retrieving cluster password from node: %s" % self.id)
    return self._call("get_cluster_password",
                      guest_api.AGENT_LOW_TIMEOUT, self.version_cap)

def get_cluster_rebalance_status(self):
    LOG.debug("Retrieving status of current cluster rebalance via node: %s"
              % self.id)
    return self._call("get_cluster_rebalance_status",
                      guest_api.AGENT_LOW_TIMEOUT, self.version_cap)

def add_nodes(self, nodes):
    LOG.debug("Adding nodes to the cluster: %s" % self.id)
    self._cast('add_nodes', self.version_cap, nodes=nodes)

def remove_nodes(self, nodes):
    LOG.debug("Removing nodes from the cluster: %s" % self.id)
    self._cast('remove_nodes', self.version_cap, nodes=nodes)

def cluster_complete(self):
    LOG.debug("Sending a setup completion notification for node: %s"
              % self.id)
    return self._call("cluster_complete", guest_api.AGENT_HIGH_TIMEOUT,
                      self.version_cap)

Alternatives

None

Dashboard Impact (UX)

Will need to enable Couchbase as a clustering datastore.

Implementation

Assignee(s)

Primary assignee:
Petr Malik <pmalik@tesora.com>

Milestones

Ocata

Work Items

The preliminary work in the guest manager and cluster strategy itself may be delivered in two separate commits if it facilitates smoother review process.

Upgrade Implications

None

Dependencies

None

Testing

  • Manager unittests will be added where appropriate.
  • The scenario tests already cover implemented functionality.

Documentation Impact

Datastore documentation for Couchbase will need to be updated to reflect clustering support.

Appendix

None