5.1. Compaction

The compaction operation is the way to reduce disk space usage by removing unused and old data from database or view index files. This operation is very similar to the vacuum (SQLite ex.) operation available for other database management systems.

During compaction of the target CouchDB creates new file with the .compact extension and transfers only actual data into. Because of this, CouchDB checks first for the available disk space - it should be twice greater than the compacted file’s data.

When all actual data is successfully transferred to the compacted file CouchDB replaces the target with the compacted file.

5.1.1. Database Compaction

Database compaction compresses the database file by removing unused file sections created during updates. Old documents revisions are replaced with small amount of metadata called tombstone which are used for conflicts resolution during replication. The number of stored revisions (and their tombstones) can be configured by using the _revs_limit URL endpoint.

Compaction can be manually triggered per database and runs as a background task. To start it for specific database there is need to send HTTP POST /{db}/_compact sub-resource of the target database:

curl -H "Content-Type: application/json" -X POST http://localhost:5984/my_db/_compact

On success, HTTP status 202 Accepted is returned immediately:

HTTP/1.1 202 Accepted
Cache-Control: must-revalidate
Content-Length: 12
Content-Type: text/plain; charset=utf-8
Date: Wed, 19 Jun 2013 09:43:52 GMT
Server: CouchDB (Erlang/OTP)

Although the request body is not used you must still specify Content-Type header with application/json value for the request. If you don’t, you will be aware about with HTTP status 415 Unsupported Media Type response:

HTTP/1.1 415 Unsupported Media Type
Cache-Control: must-revalidate
Content-Length: 78
Content-Type: application/json
Date: Wed, 19 Jun 2013 09:43:44 GMT
Server: CouchDB (Erlang/OTP)

{"error":"bad_content_type","reason":"Content-Type must be application/json"}

When the compaction is successful started and running it is possible to get information about it via database information resource:

curl http://localhost:5984/my_db
HTTP/1.1 200 OK
Cache-Control: must-revalidate
Content-Length: 246
Content-Type: application/json
Date: Wed, 19 Jun 2013 16:51:20 GMT
Server: CouchDB (Erlang/OTP)

    "committed_update_seq": 76215,
    "compact_running": true,
    "db_name": "my_db",
    "disk_format_version": 6,
    "doc_count": 5091,
    "doc_del_count": 0,
    "instance_start_time": "0",
    "purge_seq": 0,
    "sizes": {
      "active": 3787996,
      "disk": 17703025,
      "external": 4763321
    "update_seq": 76215

Note that compaction_running field is true indicating that compaction is actually running. To track the compaction progress you may query the _active_tasks resource:

curl http://localhost:5984/_active_tasks
HTTP/1.1 200 OK
Cache-Control: must-revalidate
Content-Length: 175
Content-Type: application/json
Date: Wed, 19 Jun 2013 16:27:23 GMT
Server: CouchDB (Erlang/OTP)

        "changes_done": 44461,
        "database": "my_db",
        "pid": "<0.218.0>",
        "progress": 58,
        "started_on": 1371659228,
        "total_changes": 76215,
        "type": "database_compaction",
        "updated_on": 1371659241

5.1.2. Views Compaction

Views are also need compaction like databases, unlike databases views are compacted by groups per design document. To start their compaction there is need to send HTTP POST /{db}/_compact/{ddoc} request:

curl -H "Content-Type: application/json" -X POST http://localhost:5984/dbname/_compact/designname

This compacts the view index from the current version of the specified design document. The HTTP response code is 202 Accepted (like compaction for databases) and a compaction background task will be created. Views cleanup

View indexes on disk are named after their MD5 hash of the view definition. When you change a view, old indexes remain on disk. To clean up all outdated view indexes (files named after the MD5 representation of views, that does not exist anymore) you can trigger a view cleanup:

curl -H "Content-Type: application/json" -X POST http://localhost:5984/dbname/_view_cleanup

5.1.3. Automatic Compaction

CouchDB’s automatic compaction daemon, internally known as “smoosh”, will trigger compaction jobs for both databases and views based on configurable thresholds for the sparseness of a file and the total amount of space that can be recovered. Channels

Smoosh works using the concept of channels. A channel is essentially a queue of pending compactions. There are separate sets of active channels for databases and views. Each channel is assigned a configuration which defines whether a compaction ends up in the channel’s queue and how compactions are prioritized within that queue.

Smoosh takes each channel and works through the compactions queued in each in priority order. Each channel is processed concurrently, so the priority levels only matter within a given channel. Each channel has an assigned number of active compactions, which defines how many compactions happen for that channel in parallel. For example, a cluster with a lot of database churn but few views might require more active compactions in the database channel(s).

It’s important to remember that a channel is local to a CouchDB node; that is, each node maintains and processes an independent set of compactions. Channels are defined as either “ratio” channels or “slack” channels, depending on the type of algorithm used for prioritization:

  • Ratio: uses the ratio of sizes.file / sizes.active as its driving calculation. The result X must be greater than some configurable value Y for a compaction to be added to the queue. Compactions are then prioritised for higher values of X.
  • Slack: uses the difference of sizes.file - sizes.active as its driving calculation. The result X must be greater than some configurable value Y for a compaction to be added to the queue. Compactions are prioritised for higher values of X.

In both cases, Y is set using the min_priority configuration variable. CouchDB ships with four channels pre-configured: one channel of each type for databases, and another one for views. Channel Configuration

Channels are defined using [smoosh.<channel_name>] configuration blocks, and activated by naming the channel in the db_channels or view_channels configuration setting in the [smoosh] block. The default configuration is

db_channels = upgrade_dbs,ratio_dbs,slack_dbs
view_channels = upgrade_views,ratio_views,slack_views

priority = ratio
min_priority = 2.0

priority = ratio
min_priority = 2.0

priority = slack
min_priority = 16777216

priority = slack
min_priority = 16777216

The “upgrade” channels are a special pair of channels that only check whether the disk_format_version for the file matches the current version, and enqueue the file for compaction (which has the side effect of upgrading the file format) if that’s not the case. There are several additional properties that can be configured for each channel; these are documented in the configuration API Scheduling Windows

Each compaction channel can be configured to run only during certain hours of the day. The channel-specific from, to, and strict_window configuration settings control this behavior. For example

from = 20:00
to = 06:00
strict_window = true

where overnight_channel is the name of the channel you want to configure.

Note: CouchDB determines time via the UTC (GMT) timezone, so these settings must be expressed as UTC (GMT).

The strict_window setting will cause the compaction daemon to suspend all active compactions in this channel when exiting the window, and resume them when re-entering. If strict_window is left at its default of false, the active compactions will be allowed to complete but no new compactions will be started. Migration Guide

Previous versions of CouchDB shipped with a simpler compaction daemon. The configuration system for the new daemon is not backwards-compatible with the old one, so users with customized compaction configurations will need to port them to the new setup. The old daemon’s compaction rules configuration looked like

min_file_size = 131072
check_interval = 3600
snooze_period_ms = 3000

mydb = [{db_fragmentation, "70%"}, {view_fragmentation, "60%"}, {parallel_view_compaction, true}]
_default = [{db_fragmentation, "50%"}, {view_fragmentation, "55%"}, {from, "20:00"}, {to, "06:00"}, {strict_window, true}]

Many of the elements of this configuration can be ported over to the new system. Examining each in detail:

  • min_file_size is now configured on a per-channel basis using the min_size config setting.
  • db_fragmentation is equivalent to configuring a priority = ratio channel with min_priority set to 1.0 / (1 - db_fragmentation/100) and then listing that channel in the [smoosh] db_channels config setting.
  • view_fragmention is likewise equivalent to configuring a priority = ratio channel with min_priority set to 1.0 / (1 - view_fragmentation/100) and then listing that channel in the [smoosh] view_channels config setting.
  • from / to / strict_window: each of these settings can be applied on a per-channel basis in the new daemon. The one behavior change is that the new daemon will suspend compactions upon exiting the allowed window instead of canceling them outright, and resume them when re-entering.
  • parallel_view_compaction: each compaction channel has a concurrency setting that controls how many compactions will execute in parallel in that channel. The total parallelism is the sum of the concurrency settings of all active channels. This is a departure from the previous behavior, in which the daemon would only focus on one database and/or its views (depending on the value of this flag) at a time.

The check_interval and snooze_period_ms settings are obsolete in the event-driven design of the new daemon. The new daemon does not support setting database-specific thresholds as in the mydb setting above. Rather, channels can be configured to focus on specific classes of files: large databases, small view indexes, and so on. Most cases of named database compaction rules can be expressed using properties of those databases and/or their associated views.