couchbase-index-manager

couchbase-index-manager

Overview

Provides a command-line interface to manage Couchbase indexes, synchronizing them to index definitions provided in files. It is intended to be used as part of a CI/CD pipeline, or to assist with local development.

It also provides an API which may be used by importing a node module.

Common parameters

• -c couchbase://xxx - Couchbase cluster, defaults to localhost
• -u username - Username to connect to the Couchbase cluster
• -p password - Password to connect to the Couchbase cluster
• -q - Quiet output, only prints errors and warnings
• --no-rbac - Disable RBAC, required to connect to 4.x clusters
• --no-color - Suppress color in output

Sync Command

The sync command executes an index synchronization

couchbase-index-manager [common-options] sync [sync-options] <bucketName> <path...>

bucketName should be the name of the bucket to sync, and path is the path to the index definitions. path may be either a single YAML or JSON file, or a directory containing multiple files. Multiple paths may also be provided, they will be processed in order.

Supply "-" as the path to process definitions from stdin. JSON input will be assumed if it starts with a curly brace, otherwise it will be parsed as YAML.

cat definitions.yaml | couchbase-index-manager -c couchbase://node -u Administrator -p password sync beer-sample -

Note: --force is assumed if processing from stdin.

Options

• -f - Skip the configuration prompt, just run the sync (useful for scripting)
• --dry-run - Just output the plan, don't make any changes
• --safe - Don't drop any existing indexes, only create new ones
• -t 30 - Seconds to wait for index build to complete, 0 for infinite (default 5m)
• --bucket-password password - For 4.x clusters, provides the bucket password for secure buckets

Examples

couchbase-index-manager -c couchbase://localhost -u Administrator -p password sync beer-sample ./directory/
couchbase-index-manager -c couchbase://localhost -u Administrator -p password sync beer-sample ./directory/file.yaml
couchbase-index-manager -c couchbase://localhost -u Administrator -p password sync beer-sample ./directory/file.json

Definition Files

Definition files may be written in either JSON or YAML. The define the name of the index, the columns to be index, and may also contain other options.

When YAML is used, multiple definitions may be provided in a single file, separated by a line of dashes.

name: beer_primary
is_primary: true
---
name: BeersByAbv
index_key:
- abv DESC
condition: (`type` = 'beer')
num_replica: 0
---
name: BeersByIbu
index_key:
- ibu
condition: (`type` = 'beer')
num_replica: 0
---
name: OldIndex
lifecycle:
  drop: true

Field	Required	Description
type	N	If present, must be "index"
name	Y	Name of the index.
is_primary	N	True for a primary index.
index_key	N	Array of index keys. May be attributes of documents or deterministic functions.
condition	N	Condition for the WHERE clause of the index.
partition	N	For CB 5.5, object to specify index partitioning
partition.exprs	Y	Required if partition is present, array of strings for attributes used to create partition
partition.stragegy	N	Partition strategy to use, defaults to hash
manual_replica	N	Force manual replica management, rather than using Couchbase 5.X automatic replicas. Automatically set to true for Couchbase 4.X.
num_replica	N	Defaults to 0, number of index replicas to create.
nodes	N	List of nodes for index placement. Automatic placement is used if not present.
lifecycle.drop	N	If true, drops the index if it exists.

A primary index must not have index_key or condition properties. A secondary index must have values in the index_key array. Additionally, there may not be more than one primary index in the set of definitions.

For Couchbase Server 5.0 and later, you may append DESC to the end of an index_key to use a descending collation.

If nodes and num_replica are both present, then num_replica must be the number of nodes minus one.

Overrides

When deploying to multiple environments, there may be variations in index definitions. For example, you may have a different number of replicas or a different list of node assignments. To support this, you may also apply overrides to the index definitions.

Overrides are processed in the order they are found, and can only override index definitions that with the same name. The index definition must also be found before the override. Any field which is not supplied on the override will be skipped, leaving the original value unchanged. The exception is nodes and num_replica, updating one will automatically adjust the other field.

Field	Required	Description
type	Y	Always "override".
name	Y	Name of the index.
is_primary	N	True for a primary index.
index_key	N	Array of index keys. May be attributes of documents or deterministic functions.
condition	N	Condition for the WHERE clause of the index.
partition	N	For CB 5.5, object to specify index partitioning. Use null to remove partition during override.
partition.exprs	N	Array of strings for attributes used to create partition, replaces the existing array.
partition.stragegy	N	Partition strategy to use
manual_replica	N	Force manual replica management, rather than using Couchbase 5.X automatic replicas. Automatically set to true for Couchbase 4.X.
num_replica	N	Number of index replicas to create.
nodes	N	List of nodes for index placement.
lifecycle.drop	N	If true, drops the index if it exists.
post_process	N	Optional Javascript function body which may further alter the index definition. "this" will be the index definition.

Node Maps

When deploying to multiple environments, the names and IPs of nodes in the clusters probably vary. You may use node maps to provide aliases for nodes to support these different environments. Even for a single environment it may be useful to help reduce repetition. For example, you may define several indexes referencing "node1" and "node2" in the nodes attribute. Then add a node map which maps "node1" to "172.21.0.2" and "node2" to "127.21.0.3".

Node maps are processed in the order they are encountered, and values defined in later node maps will override or append to earlier node maps. Any node not found in the node map will be unchanged and treated as a fully qualified name.

Field	Required	Description
type	Y	Always "nodeMap".
map	Y	A hashmap keyed by alias, with the value being the fully qualified node name.

Example

name: BeersByIbu
index_key:
- ibu
condition: (`type` = 'beer')
nodes:
- node1
- node2
---
type: nodeMap
map:
  node1: 172.21.0.2
  node2: 172.21.0.3

Updating Indexes

Updating existing indexes is currently an unsafe operation in most cases. This is because the index must be dropped and recreated, so there is some period of time when the index is unavailable for queries. The exception is if you have at least 1 replica configured and manual_replica is true. In this case, the replicas are replaced one at a time so at least one replica is always available.

Dropping Indexes

If an index is removed from the definition files, it is not dropped. This prevents different CI/CD processes from interfering with each other as they manage different indexes. To drop an index, leave the definition in place but set lifecycle.drop to true.

Manual Index Replica Management

Replicas are emulated on Couchbase Server 4.X by creating multiple indexes. If num_replica is greater than 0, the additional indexes are named with the suffix _replicaN, where N starts at 1. For example, an index with 2 replicas named MyIndex will have 3 indexes, MyIndex, MyIndex_replica1, and MyIndex_replica2.

This approach may also be enabled on Couchbase Server 5.X by settings manual_replica to true on the index definition.

During updates, changes to the nodes list may result in indexes being moved from one node to another. So long as there is at least one replica this is considered a safe operation, as each replica will be moved independently leaving other replicas available to serve requests. Changes to the order of the node list are ignored, only adding or removing nodes results in a change.

Automatic Index Replica Management

On Couchbase Server 5.X, automatic index replica management is the default. In this case, replicas are managed by Couchbase Server directly, rather than by couchbase-index-manager.

Note that for Couchbase Server 5.0 and 5.1, the nodes list is only respected during index creation. Indexes will not be moved between nodes if they already exist. Beginning with Couchbase Server 5.5 an ALTER INDEX command will be used to move replicas between nodes.

Because ALTER INDEX cannot currently change the number of replicas, changes to num_replica or the number of nodes in nodes is an unsafe change that will drop and recreate the index.

Docker Image

A Docker image for running couchbase-index-manager is available at Docker Hub.

docker run --rm -it -v ./:/definitions btburnett3/couchbase-index-manager -c couchbase://cluster -u Administrator -p password sync beer-sample /definitions