5.2. CouchDB Replication Protocol

Version:
3

The CouchDB Replication Protocol is a protocol for synchronising JSONdocuments between 2 peers over HTTP/1.1 by using the public CouchDB RESTAPI and is based on the Apache CouchDB MVCC Data model.

5.2.1. Preface

5.2.1.1. Language

The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”,“SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “MAY”, and “OPTIONAL” in thisdocument are to be interpreted as described in RFC 2119.

5.2.1.2. Goals

The primary goal of this specification is to describe the CouchDB ReplicationProtocol under the hood.

The secondary goal is to provide enough detailed information about the protocolto make it easy to build tools on any language and platform that can synchronizedata with CouchDB.

5.2.1.3. Definitions

JSON:
JSON is a text format for theserialization of structured data. It is described in ECMA-262 andRFC 4627.
URI:
A URI is defined by RFC 3986. It can be a URL as definedin RFC 1738.
ID:
An identifier (could be a UUID) as described in RFC 4122.
Revision:
A MVCC token value of following pattern: N-sig where N is ALWAYSa positive integer and sig is the Document signature (custom).Don’t mix it up with the revision in version control systems!
Leaf Revision:
The last Document Revision in a series of changes. Documents may havemultiple Leaf Revisions (aka Conflict Revisions) due to concurrent updates.
Document:
A document is a JSON object with an ID and Revision defined in _id and_rev fields respectively. A Document’s ID MUST be unique withinthe Database where it is stored.
Database:
A collection of Documents with a unique URI.
Changes Feed:
A stream of Document-changing events (create, update, delete) forthe specified Database.
Sequence ID:
An ID provided by the Changes Feed. It MUST be incremental,but MAY NOT always be an integer.
Source:
Database from where the Documents are replicated.
Target:
Database where the Documents are replicated to.
Replication:
The one-way directed synchronization process of Source and Target endpoints.
Checkpoint:
Intermediate Recorded Sequence ID used for Replication recovery.
Replicator:
A service or an application which initiates and runs Replication.
Filter Function:
A special function of any programming language that is used to filterDocuments during Replication (see Filter Functions)
Filter Function Name:
An ID of a Filter Function that may be used as a symbolic reference (akacallback function) to apply the related Filter Function to Replication.
Filtered Replication:
Replication of Documents from Source to Target using a Filter Function.
Full Replication:
Replication of all Documents from Source to Target.
Push Replication:
Replication process where Source is a local endpoint and Target is remote.
Pull Replication:
Replication process where Source is a remote endpoint and Target is local.
Continuous Replication:
Replication that “never stops”: after processing all events from theChanges Feed, the Replicator doesn’t close the connection, but awaits newchange events from the Source. The connection is kept alive by periodicheartbeats.
Replication Log:
A special Document that holds Replication history (recorded Checkpointsand a few more statistics) between Source and Target.
Replication ID:
A unique value that unambiguously identifies the Replication Log.

5.2.2. Replication Protocol Algorithm

The CouchDB Replication Protocol is not magical, butan agreement on usage of the public CouchDB HTTP REST API toenable Documents to be replicated from Source to Target.

The reference implementation, written in Erlang, is provided by thecouch_replicator module in Apache CouchDB.

It is RECOMMENDED that one follow this algorithm specification, use the sameHTTP endpoints, and run requests with the same parameters to provide acompletely compatible implementation. Custom Replicator implementations MAY usedifferent HTTP API endpoints and request parameters depending on their localspecifics and they MAY implement only part of the Replication Protocol to runonly Push or Pull Replication. However, while such solutions could also run theReplication process, they loose compatibility with the CouchDB Replicator.

5.2.2.1. Verify Peers

  1. + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - +
  2. ' Verify Peers: '
  3. ' '
  4. ' 404 Not Found +--------------------------------+ '
  5. ' +----------------------- | Check Source Existence | '
  6. ' | +--------------------------------+ '
  7. ' | | HEAD /source | '
  8. ' | +--------------------------------+ '
  9. ' | | '
  10. ' | | 200 OK '
  11. ' | v '
  12. ' | +--------------------------------+ '
  13. ' | | Check Target Existence | ----+ '
  14. ' | +--------------------------------+ | '
  15. ' | | HEAD /target | | '
  16. ' | +--------------------------------+ | '
  17. ' | | | '
  18. ' | | 404 Not Found | '
  19. ' v v | '
  20. ' +-------+ No +--------------------------------+ | '
  21. ' | Abort | <----------------- | Create Target? | | '
  22. ' +-------+ +--------------------------------+ | '
  23. ' ^ | | '
  24. ' | | Yes | '
  25. ' | v | '
  26. ' | Failure +--------------------------------+ | '
  27. ' +----------------------- | Create Target | | '
  28. ' +--------------------------------+ | '
  29. ' | PUT /target | | '
  30. ' +--------------------------------+ | '
  31. ' | | '
  32. ' | 201 Created 200 OK | '
  33. ' | | '
  34. + - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - - - - | - +
  35. | |
  36. + - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - - - - | - +
  37. ' Get Peers Information: | | '
  38. ' +------------------------------------+ '
  39. ' | '
  40. ' v '
  41. ' +--------------------------------+ '
  42. ' | Get Source Information | '
  43. ' +--------------------------------+ '
  44. ' '
  45. + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - +

The Replicator MUST ensure that both Source and Target existby using HEAD /{db} requests.

5.2.2.1.1. Check Source Existence

Request:

  1. HEAD /source HTTP/1.1
    Host: localhost:5984
    User-Agent: CouchDB


Response:

  1. HTTP/1.1 200 OK
    Cache-Control: must-revalidate
    Content-Type: application/json
    Date: Sat, 05 Oct 2013 08:50:39 GMT
    Server: CouchDB (Erlang/OTP)


5.2.2.1.2. Check Target Existence

Request:

  1. HEAD /target HTTP/1.1
    Host: localhost:5984
    User-Agent: CouchDB


Response:

  1. HTTP/1.1 200 OK
    Cache-Control: must-revalidate
    Content-Type: application/json
    Date: Sat, 05 Oct 2013 08:51:11 GMT
    Server: CouchDB (Erlang/OTP)


5.2.2.1.3. Create Target?

In case of a non-existent Target, the Replicator MAY make a PUT /{db}request to create the Target:

Request:

  1. PUT /target HTTP/1.1
    Accept: application/json
    Host: localhost:5984
    User-Agent: CouchDB


Response:

  1. HTTP/1.1 201 Created
    Content-Length: 12
    Content-Type: application/json
    Date: Sat, 05 Oct 2013 08:58:41 GMT
    Server: CouchDB (Erlang/OTP)

    {
    "ok": true
    }


However, the Replicator’s PUT request MAY NOT succeeded due to insufficientprivileges (which are granted by the provided credential) and so receive a401 Unauthorized or a 403 Forbidden error. Such errors SHOULD be expectedand well handled:



  1. HTTP/1.1 500 Internal Server Error
    Cache-Control: must-revalidate
    Content-Length: 108
    Content-Type: application/json
    Date: Fri, 09 May 2014 13:50:32 GMT
    Server: CouchDB (Erlang OTP)

    {
    "error": "unauthorized",
    "reason": "unauthorized to access or create database http://localhost:5984/target&#34;
    }


5.2.2.1.4. Abort

In case of a non-existent Source or Target, Replication SHOULD be aborted withan HTTP error response:



  1. HTTP/1.1 500 Internal Server Error
    Cache-Control: must-revalidate
    Content-Length: 56
    Content-Type: application/json
    Date: Sat, 05 Oct 2013 08:55:29 GMT
    Server: CouchDB (Erlang OTP)

    {
    "error": "db_not_found",
    "reason": "could not open source"
    }


5.2.2.2. Get Peers Information

  1. + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -+
  2. ' Verify Peers: '
  3. ' +------------------------+ '
  4. ' | Check Target Existence | '
  5. ' +------------------------+ '
  6. ' | '
  7. ' | 200 OK '
  8. ' | '
  9. + - - - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - -+
  10. |
  11. + - - - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - -+
  12. ' Get Peers Information: | '
  13. ' v '
  14. ' +------------------------+ '
  15. ' | Get Source Information | '
  16. ' +------------------------+ '
  17. ' | GET /source | '
  18. ' +------------------------+ '
  19. ' | '
  20. ' | 200 OK '
  21. ' v '
  22. ' +------------------------+ '
  23. ' | Get Target Information | '
  24. ' +------------------------+ '
  25. ' | GET /target | '
  26. ' +------------------------+ '
  27. ' | '
  28. ' | 200 OK '
  29. ' | '
  30. + - - - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - -+
  31. |
  32. + - - - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - -+
  33. ' Find Common Ancestry: | '
  34. ' | '
  35. ' v '
  36. ' +-------------------------+ '
  37. ' | Generate Replication ID | '
  38. ' +-------------------------+ '
  39. ' '
  40. + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -+

The Replicator retrieves basic information both from Source and Target usingGET /{db} requests. The GET response MUST contain JSON objects withthe following mandatory fields:

  • instance_start_time (string): Always "0". (Returned for legacyreasons.)
  • update_seq (number / string): The current database Sequence ID.
    Any other fields are optional. The information that the Replicator needsis the updateseq field: this value will be used to define a _temporary(because Database data is subject to change) upper bound for changes feedlistening and statistic calculating to show proper Replication progress.

5.2.2.2.1. Get Source Information

Request:

  1. GET /source HTTP/1.1
    Accept: application/json
    Host: localhost:5984
    User-Agent: CouchDB


Response:

  1. HTTP/1.1 200 OK
    Cache-Control: must-revalidate
    Content-Length: 256
    Content-Type: application/json
    Date: Tue, 08 Oct 2013 07:53:08 GMT
    Server: CouchDB (Erlang OTP)

    {
    "committed_update_seq": 61772,
    "compact_running": false,
    "data_size": 70781613961,
    "db_name": "source",
    "disk_format_version": 6,
    "disk_size": 79132913799,
    "doc_count": 41961,
    "doc_del_count": 3807,
    "instance_start_time": "0",
    "purge_seq": 0,
    "update_seq": 61772
    }


5.2.2.2.2. Get Target Information

Request:

  1. GET /target/ HTTP/1.1
    Accept: application/json
    Host: localhost:5984
    User-Agent: CouchDB


Response:

  1. HTTP/1.1 200 OK
    Content-Length: 363
    Content-Type: application/json
    Date: Tue, 08 Oct 2013 12:37:01 GMT
    Server: CouchDB (Erlang/OTP)

    {
    "compact_running": false,
    "db_name": "target",
    "disk_format_version": 5,
    "disk_size": 77001455,
    "doc_count": 1832,
    "doc_del_count": 1,
    "instance_start_time": "0",
    "other": {
    "data_size": 50829452
    },
    "purge_seq": 0,
    "update_seq": "1841-g1AAAADveJzLYWBgYMlgTmGQT0lKzi9KdUhJMtbLSs1LLUst0k"
    }


5.2.2.3. Find Common Ancestry

  1. + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - +
  2. ' Get Peers Information: '
  3. ' '
  4. ' +-------------------------------------------+ '
  5. ' | Get Target Information | '
  6. ' +-------------------------------------------+ '
  7. ' | '
  8. + - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - - - - - - - - +
  9. |
  10. + - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - - - - - - - - +
  11. ' Find Common Ancestry: v '
  12. ' +-------------------------------------------+ '
  13. ' | Generate Replication ID | '
  14. ' +-------------------------------------------+ '
  15. ' | '
  16. ' | '
  17. ' v '
  18. ' +-------------------------------------------+ '
  19. ' | Get Replication Log from Source | '
  20. ' +-------------------------------------------+ '
  21. ' | GET /source/_local/replication-id | '
  22. ' +-------------------------------------------+ '
  23. ' | '
  24. ' | 200 OK '
  25. ' | 404 Not Found '
  26. ' v '
  27. ' +-------------------------------------------+ '
  28. ' | Get Replication Log from Target | '
  29. ' +-------------------------------------------+ '
  30. ' | GET /target/_local/replication-id | '
  31. ' +-------------------------------------------+ '
  32. ' | '
  33. ' | 200 OK '
  34. ' | 404 Not Found '
  35. ' v '
  36. ' +-------------------------------------------+ '
  37. ' | Compare Replication Logs | '
  38. ' +-------------------------------------------+ '
  39. ' | '
  40. ' | Use latest common sequence as start point '
  41. ' | '
  42. + - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - - - - - - - - +
  43. |
  44. |
  45. + - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - - - - - - - - +
  46. ' Locate Changed Documents: | '
  47. ' | '
  48. ' v '
  49. ' +-------------------------------------------+ '
  50. ' | Listen Source Changes Feed | '
  51. ' +-------------------------------------------+ '
  52. ' '
  53. + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - +

5.2.2.3.1. Generate Replication ID

Before Replication is started, the Replicator MUST generate a Replication ID.This value is used to track Replication History, resume and continue previouslyinterrupted Replication process.

The Replication ID generation algorithm is implementation specific. Whateveralgorithm is used it MUST uniquely identify the Replication process. CouchDB’sReplicator, for example, uses the following factors in generating a ReplicationID:

  • Persistent Peer UUID value. For CouchDB, the localServer UUID is used
  • Source and Target URI and if Source or Target are local or remote Databases
  • If Target needed to be created
  • If Replication is Continuous
  • Any custom headers
  • Filter function code if used
  • Changes Feed query parameters, if any

Note

See couch_replicator_ids.erl for an example of a Replication ID generationimplementation.

5.2.2.3.2. Retrieve Replication Logs from Source and Target

Once the Replication ID has been generated, the Replicator SHOULD retrievethe Replication Log from both Source and Target usingGET /{db}/_local/{docid}:

Request:

  1. GET /source/_local/b3e44b920ee2951cb2e123b63044427a HTTP/1.1
    Accept: application/json
    Host: localhost:5984
    User-Agent: CouchDB


Response:

  1. HTTP/1.1 200 OK
    Cache-Control: must-revalidate
    Content-Length: 1019
    Content-Type: application/json
    Date: Thu, 10 Oct 2013 06:18:56 GMT
    ETag: "0-8"
    Server: CouchDB (Erlang OTP)

    {
    "_id": "_local/b3e44b920ee2951cb2e123b63044427a",
    "_rev": "0-8",
    "history": [
    {
    "doc_write_failures": 0,
    "docs_read": 2,
    "docs_written": 2,
    "end_last_seq": 5,
    "end_time": "Thu, 10 Oct 2013 05:56:38 GMT",
    "missing_checked": 2,
    "missing_found": 2,
    "recorded_seq": 5,
    "session_id": "d5a34cbbdafa70e0db5cb57d02a6b955",
    "start_last_seq": 3,
    "start_time": "Thu, 10 Oct 2013 05:56:38 GMT"
    },
    {
    "doc_write_failures": 0,
    "docs_read": 1,
    "docs_written": 1,
    "end_last_seq": 3,
    "end_time": "Thu, 10 Oct 2013 05:56:12 GMT",
    "missing_checked": 1,
    "missing_found": 1,
    "recorded_seq": 3,
    "session_id": "11a79cdae1719c362e9857cd1ddff09d",
    "start_last_seq": 2,
    "start_time": "Thu, 10 Oct 2013 05:56:12 GMT"
    },
    {
    "doc_write_failures": 0,
    "docs_read": 2,
    "docs_written": 2,
    "end_last_seq": 2,
    "end_time": "Thu, 10 Oct 2013 05:56:04 GMT",
    "missing_checked": 2,
    "missing_found": 2,
    "recorded_seq": 2,
    "session_id": "77cdf93cde05f15fcb710f320c37c155",
    "start_last_seq": 0,
    "start_time": "Thu, 10 Oct 2013 05:56:04 GMT"
    }
    ],
    "replication_id_version": 3,
    "session_id": "d5a34cbbdafa70e0db5cb57d02a6b955",
    "source_last_seq": 5
    }


The Replication Log SHOULD contain the following fields:

  • history (array of object): Replication history. Required
    • doc_write_failures (number): Number of failed writes
    • docs_read (number): Number of read documents
    • docs_written (number): Number of written documents
    • end_last_seq (number): Last processed Update Sequence ID
    • end_time (string): Replication completion timestamp in RFC 5322format
    • missing_checked (number): Number of checked revisions on Source
    • missing_found (number): Number of missing revisions found on Target
    • recorded_seq (number): Recorded intermediate Checkpoint. Required
    • session_id (string): Unique session ID. Commonly, a random UUID valueis used. Required
    • start_last_seq (number): Start update Sequence ID
    • start_time (string): Replication start timestamp in RFC 5322 format
  • replication_id_version (number): Replication protocol version. DefinesReplication ID calculation algorithm, HTTP API calls and the othersroutines. Required
  • session_id (string): Unique ID of the last session. Shortcut tothe session_id field of the latest history object. Required
  • source_last_seq (number): Last processed Checkpoint. Shortcut tothe recorded_seq field of the latest history object. Required
    This request MAY fall with a 404 Not Found response:
Request:

  1. GET /source/_local/b6cef528f67aa1a8a014dd1144b10e09 HTTP/1.1
    Accept: application/json
    Host: localhost:5984
    User-Agent: CouchDB


Response:

  1. HTTP/1.1 404 Object Not Found
    Cache-Control: must-revalidate
    Content-Length: 41
    Content-Type: application/json
    Date: Tue, 08 Oct 2013 13:31:10 GMT
    Server: CouchDB (Erlang OTP)

    {
    "error": "not_found",
    "reason": "missing"
    }


That’s OK. This means that there is no information about the current Replicationso it must not have been run previously and as such the Replicator MUST runa Full Replication.

5.2.2.3.3. Compare Replication Logs

If the Replication Logs are successfully retrieved from both Source and Targetthen the Replicator MUST determine their common ancestry by following the nextalgorithm:

  • Compare session_id values for the chronological last session - if theymatch both Source and Target have a common Replication history and it seemsto be valid. Use source_last_seq value for the startup Checkpoint
  • In case of mismatch, iterate over the history collection to search forthe latest (chronologically) common session_id for Source and Target.Use value of recorded_seq field as startup Checkpoint
    If Source and Target has no common ancestry, the Replicator MUST runFull Replication.

5.2.2.4. Locate Changed Documents

  1. + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - +
  2. ' Find Common Ancestry: '
  3. ' '
  4. ' +------------------------------+ '
  5. ' | Compare Replication Logs | '
  6. ' +------------------------------+ '
  7. ' | '
  8. ' | '
  9. + - - - - - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - - +
  10. |
  11. + - - - - - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - - +
  12. ' Locate Changed Documents: | '
  13. ' | '
  14. ' | '
  15. ' v '
  16. ' +-------------------------------+ '
  17. ' +------> | Listen to Changes Feed | -----+ '
  18. ' | +-------------------------------+ | '
  19. ' | | GET /source/_changes | | '
  20. ' | | POST /source/_changes | | '
  21. ' | +-------------------------------+ | '
  22. ' | | | '
  23. ' | | | '
  24. ' | There are new changes | | No more changes '
  25. ' | | | '
  26. ' | v v '
  27. ' | +-------------------------------+ +-----------------------+ '
  28. ' | | Read Batch of Changes | | Replication Completed | '
  29. ' | +-------------------------------+ +-----------------------+ '
  30. ' | | '
  31. ' | No | '
  32. ' | v '
  33. ' | +-------------------------------+ '
  34. ' | | Compare Documents Revisions | '
  35. ' | +-------------------------------+ '
  36. ' | | POST /target/_revs_diff | '
  37. ' | +-------------------------------+ '
  38. ' | | '
  39. ' | 200 OK | '
  40. ' | v '
  41. ' | +-------------------------------+ '
  42. ' +------- | Any Differences Found? | '
  43. ' +-------------------------------+ '
  44. ' | '
  45. ' Yes | '
  46. ' | '
  47. + - - - - - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - - +
  48. |
  49. + - - - - - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - - +
  50. ' Replicate Changes: | '
  51. ' v '
  52. ' +-------------------------------+ '
  53. ' | Fetch Next Changed Document | '
  54. ' +-------------------------------+ '
  55. ' '
  56. + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - +

5.2.2.4.1. Listen to Changes Feed

When the start up Checkpoint has been defined, the Replicator SHOULD readthe Source’s Changes Feed by using a GET /{db}/_changesrequest. This request MUST be made with the following query parameters:

  • feed parameter defines the Changes Feed response style: for ContinuousReplication the continuous value SHOULD be used, otherwise - normal.
  • style=all_docs query parameter tells the Source that it MUST includeall Revision leaves for each document’s event in output.
  • For Continuous Replication the heartbeat parameter defines the heartbeatperiod in milliseconds. The RECOMMENDED value by default is 10000(10 seconds).
  • If a startup Checkpoint was found during the Replication Logs comparison,the since query parameter MUST be passed with this value.In case of Full Replication it MAY be 0 (number zero) orbe omitted.
    Additionally, the filter query parameter MAY be specified to enable afilter function on Source side. Othercustom parameters MAY also be provided.

5.2.2.4.2. Read Batch of Changes

Reading the whole feed in a single shot may not be an optimal use of resources.It is RECOMMENDED to process the feed in small chunks. However, there isno specific recommendation on chunk size since it is heavily dependent onavailable resources: large chunks requires more memory while they reduceI/O operations and vice versa.

Note, that Changes Feed output format is different for a request withfeed=normal and withfeed=continuous query parameter.

Normal Feed:

Request:

  1. GET /source/_changes?feed=normal&style=all_docs&heartbeat=10000 HTTP/1.1
    Accept: application/json
    Host: localhost:5984
    User-Agent: CouchDB


Response:

  1. HTTP/1.1 200 OK
    Cache-Control: must-revalidate
    Content-Type: application/json
    Date: Fri, 09 May 2014 16:20:41 GMT
    Server: CouchDB (Erlang OTP)
    Transfer-Encoding: chunked

    {"results":[
    {"seq":14,"id":"f957f41e","changes":[{"rev":"3-46a3"}],"deleted":true}
    {"seq":29,"id":"ddf339dd","changes":[{"rev":"10-304b"}]}
    {"seq":37,"id":"d3cc62f5","changes":[{"rev":"2-eec2"}],"deleted":true}
    {"seq":39,"id":"f13bd08b","changes":[{"rev":"1-b35d"}]}
    {"seq":41,"id":"e0a99867","changes":[{"rev":"2-c1c6"}]}
    {"seq":42,"id":"a75bdfc5","changes":[{"rev":"1-967a"}]}
    {"seq":43,"id":"a5f467a0","changes":[{"rev":"1-5575"}]}
    {"seq":45,"id":"470c3004","changes":[{"rev":"11-c292"}]}
    {"seq":46,"id":"b1cb8508","changes":[{"rev":"10-ABC"}]}
    {"seq":47,"id":"49ec0489","changes":[{"rev":"157-b01f"},{"rev":"123-6f7c"}]}
    {"seq":49,"id":"dad10379","changes":[{"rev":"1-9346"},{"rev":"6-5b8a"}]}
    {"seq":50,"id":"73464877","changes":[{"rev":"1-9f08"}]}
    {"seq":51,"id":"7ae19302","changes":[{"rev":"1-57bf"}]}
    {"seq":63,"id":"6a7a6c86","changes":[{"rev":"5-acf6"}],"deleted":true}
    {"seq":64,"id":"dfb9850a","changes":[{"rev":"1-102f"}]}
    {"seq":65,"id":"c532afa7","changes":[{"rev":"1-6491"}]}
    {"seq":66,"id":"af8a9508","changes":[{"rev":"1-3db2"}]}
    {"seq":67,"id":"caa3dded","changes":[{"rev":"1-6491"}]}
    {"seq":68,"id":"79f3b4e9","changes":[{"rev":"1-102f"}]}
    {"seq":69,"id":"1d89d16f","changes":[{"rev":"1-3db2"}]}
    {"seq":71,"id":"abae7348","changes":[{"rev":"2-7051"}]}
    {"seq":77,"id":"6c25534f","changes":[{"rev":"9-CDE"},{"rev":"3-00e7"},{"rev":"1-ABC"}]}
    {"seq":78,"id":"SpaghettiWithMeatballs","changes":[{"rev":"22-5f95"}]}
    ],
    "last_seq":78}


Continuous Feed:

Request:

  1. GET /source/_changes?feed=continuous&style=all_docs&heartbeat=10000 HTTP/1.1
    Accept: application/json
    Host: localhost:5984
    User-Agent: CouchDB


Response:

  1. HTTP/1.1 200 OK
    Cache-Control: must-revalidate
    Content-Type: application/json
    Date: Fri, 09 May 2014 16:22:22 GMT
    Server: CouchDB (Erlang OTP)
    Transfer-Encoding: chunked

    {"seq":14,"id":"f957f41e","changes":[{"rev":"3-46a3"}],"deleted":true}
    {"seq":29,"id":"ddf339dd","changes":[{"rev":"10-304b"}]}
    {"seq":37,"id":"d3cc62f5","changes":[{"rev":"2-eec2"}],"deleted":true}
    {"seq":39,"id":"f13bd08b","changes":[{"rev":"1-b35d"}]}
    {"seq":41,"id":"e0a99867","changes":[{"rev":"2-c1c6"}]}
    {"seq":42,"id":"a75bdfc5","changes":[{"rev":"1-967a"}]}
    {"seq":43,"id":"a5f467a0","changes":[{"rev":"1-5575"}]}
    {"seq":45,"id":"470c3004","changes":[{"rev":"11-c292"}]}
    {"seq":46,"id":"b1cb8508","changes":[{"rev":"10-ABC"}]}
    {"seq":47,"id":"49ec0489","changes":[{"rev":"157-b01f"},{"rev":"123-6f7c"}]}
    {"seq":49,"id":"dad10379","changes":[{"rev":"1-9346"},{"rev":"6-5b8a"}]}
    {"seq":50,"id":"73464877","changes":[{"rev":"1-9f08"}]}
    {"seq":51,"id":"7ae19302","changes":[{"rev":"1-57bf"}]}
    {"seq":63,"id":"6a7a6c86","changes":[{"rev":"5-acf6"}],"deleted":true}
    {"seq":64,"id":"dfb9850a","changes":[{"rev":"1-102f"}]}
    {"seq":65,"id":"c532afa7","changes":[{"rev":"1-6491"}]}
    {"seq":66,"id":"af8a9508","changes":[{"rev":"1-3db2"}]}
    {"seq":67,"id":"caa3dded","changes":[{"rev":"1-6491"}]}
    {"seq":68,"id":"79f3b4e9","changes":[{"rev":"1-102f"}]}
    {"seq":69,"id":"1d89d16f","changes":[{"rev":"1-3db2"}]}
    {"seq":71,"id":"abae7348","changes":[{"rev":"2-7051"}]}
    {"seq":75,"id":"SpaghettiWithMeatballs","changes":[{"rev":"21-5949"}]}
    {"seq":77,"id":"6c255","changes":[{"rev":"9-CDE"},{"rev":"3-00e7"},{"rev":"1-ABC"}]}
    {"seq":78,"id":"SpaghettiWithMeatballs","changes":[{"rev":"22-5f95"}]}


For both Changes Feed formats record-per-line style is preserved to simplifyiterative fetching and decoding JSON objects with less memory footprint.

5.2.2.4.3. Calculate Revision Difference

After reading the batch of changes from the Changes Feed, the Replicator forms aJSON mapping object for Document ID and related leaf Revisions and sendsthe result to Target via a POST /{db}/_revs_diff request:

Request:

  1. POST /target/_revs_diff HTTP/1.1
    Accept: application/json
    Content-Length: 287
    Content-Type: application/json
    Host: localhost:5984
    User-Agent: CouchDB

    {
    "baz": [
    "2-7051cbe5c8faecd085a3fa619e6e6337"
    ],
    "foo": [
    "3-6a540f3d701ac518d3b9733d673c5484"
    ],
    "bar": [
    "1-d4e501ab47de6b2000fc8a02f84a0c77",
    "1-967a00dff5e02add41819138abb3284d"
    ]
    }


Response:

  1. HTTP/1.1 200 OK
    Cache-Control: must-revalidate
    Content-Length: 88
    Content-Type: application/json
    Date: Fri, 25 Oct 2013 14:44:41 GMT
    Server: CouchDB (Erlang/OTP)

    {
    "baz": {
    "missing": [
    "2-7051cbe5c8faecd085a3fa619e6e6337"
    ]
    },
    "bar": {
    "missing": [
    "1-d4e501ab47de6b2000fc8a02f84a0c77"
    ]
    }
    }


In the response the Replicator receives a Document ID – Revisions mapping,but only for Revisions that do not exist in Target and are REQUIRED to betransferred from Source.

If all Revisions in the request match the current state of the Documents thenthe response will contain an empty JSON object:

Request

  1. POST /target/_revs_diff HTTP/1.1
    Accept: application/json
    Content-Length: 160
    Content-Type: application/json
    Host: localhost:5984
    User-Agent: CouchDB

    {
    "foo": [
    "3-6a540f3d701ac518d3b9733d673c5484"
    ],
    "bar": [
    "1-967a00dff5e02add41819138abb3284d"
    ]
    }


Response:

  1. HTTP/1.1 200 OK
    Cache-Control: must-revalidate
    Content-Length: 2
    Content-Type: application/json
    Date: Fri, 25 Oct 2013 14:45:00 GMT
    Server: CouchDB (Erlang/OTP)

    {}


5.2.2.4.4. Replication Completed

When there are no more changes left to process and no more Documents left toreplicate, the Replicator finishes the Replication process. If Replicationwasn’t Continuous, the Replicator MAY return a response to client withstatistics about the process.



  1. HTTP/1.1 200 OK
    Cache-Control: must-revalidate
    Content-Length: 414
    Content-Type: application/json
    Date: Fri, 09 May 2014 15:14:19 GMT
    Server: CouchDB (Erlang OTP)

    {
    "history": [
    {
    "doc_write_failures": 2,
    "docs_read": 2,
    "docs_written": 0,
    "end_last_seq": 2939,
    "end_time": "Fri, 09 May 2014 15:14:19 GMT",
    "missing_checked": 1835,
    "missing_found": 2,
    "recorded_seq": 2939,
    "session_id": "05918159f64842f1fe73e9e2157b2112",
    "start_last_seq": 0,
    "start_time": "Fri, 09 May 2014 15:14:18 GMT"
    }
    ],
    "ok": true,
    "replication_id_version": 3,
    "session_id": "05918159f64842f1fe73e9e2157b2112",
    "source_last_seq": 2939
    }


5.2.2.5. Replicate Changes

  1. + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - +
  2. ' Locate Changed Documents: '
  3. ' '
  4. ' +-------------------------------------+ '
  5. ' | Any Differences Found? | '
  6. ' +-------------------------------------+ '
  7. ' | '
  8. ' | '
  9. ' | '
  10. + - - - - - - - - - - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - +
  11. |
  12. + - - - - - - - - - - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - +
  13. ' Replicate Changes: | '
  14. ' v '
  15. ' +-------------------------------------+ '
  16. ' +---------> | Fetch Next Changed Document | <---------------------+ '
  17. ' | +-------------------------------------+ | '
  18. ' | | GET /source/docid | | '
  19. ' | +-------------------------------------+ | '
  20. ' | | | '
  21. ' | | | '
  22. ' | | 201 Created | '
  23. ' | | 200 OK 401 Unauthorized | '
  24. ' | | 403 Forbidden | '
  25. ' | | | '
  26. ' | v | '
  27. ' | +-------------------------------------+ | '
  28. ' | +------ | Document Has Changed Attachments? | | '
  29. ' | | +-------------------------------------+ | '
  30. ' | | | | '
  31. ' | | | | '
  32. ' | | | Yes | '
  33. ' | | | | '
  34. ' | | v | '
  35. ' | | +------------------------+ Yes +---------------------------+ '
  36. ' | | No | Are They Big Enough? | -------> | Update Document on Target | '
  37. ' | | +------------------------+ +---------------------------+ '
  38. ' | | | | PUT /target/docid | '
  39. ' | | | +---------------------------+ '
  40. ' | | | '
  41. ' | | | No '
  42. ' | | | '
  43. ' | | v '
  44. ' | | +-------------------------------------+ '
  45. ' | +-----> | Put Document Into the Stack | '
  46. ' | +-------------------------------------+ '
  47. ' | | '
  48. ' | | '
  49. ' | v '
  50. ' | No +-------------------------------------+ '
  51. ' +---------- | Stack is Full? | '
  52. ' | +-------------------------------------+ '
  53. ' | | '
  54. ' | | Yes '
  55. ' | | '
  56. ' | v '
  57. ' | +-------------------------------------+ '
  58. ' | | Upload Stack of Documents to Target | '
  59. ' | +-------------------------------------+ '
  60. ' | | POST /target/_bulk_docs | '
  61. ' | +-------------------------------------+ '
  62. ' | | '
  63. ' | | 201 Created '
  64. ' | v '
  65. ' | +-------------------------------------+ '
  66. ' | | Ensure in Commit | '
  67. ' | +-------------------------------------+ '
  68. ' | | POST /target/_ensure_full_commit | '
  69. ' | +-------------------------------------+ '
  70. ' | | '
  71. ' | | 201 Created '
  72. ' | v '
  73. ' | +-------------------------------------+ '
  74. ' | | Record Replication Checkpoint | '
  75. ' | +-------------------------------------+ '
  76. ' | | PUT /source/_local/replication-id | '
  77. ' | | PUT /target/_local/replication-id | '
  78. ' | +-------------------------------------+ '
  79. ' | | '
  80. ' | | 201 Created '
  81. ' | v '
  82. ' | No +-------------------------------------+ '
  83. ' +---------- | All Documents from Batch Processed? | '
  84. ' +-------------------------------------+ '
  85. ' | '
  86. ' Yes | '
  87. ' | '
  88. + - - - - - - - - - - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - +
  89. |
  90. + - - - - - - - - - - - - - - - - - - - - - - - - - | - - - - - - - - - - - - - - +
  91. ' Locate Changed Documents: | '
  92. ' v '
  93. ' +-------------------------------------+ '
  94. ' | Listen to Changes Feed | '
  95. ' +-------------------------------------+ '
  96. ' '
  97. + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - +

5.2.2.5.1. Fetch Changed Documents

At this step the Replicator MUST fetch all Document Leaf Revisions from Sourcethat are missed at Target. This operation is effective if Replication WILLuse previously calculated Revision differences since they definemissing Documents and their Revisions.

To fetch the Document the Replicator will make a GET /{db}/{docid} requestwith the following query parameters:

  • revs=true: Instructs the Source to include the list of all known revisionsinto the Document in the _revisions field. This information is needed tosynchronize the Document’s ancestors history between Source and Target
  • The open_revs query parameter contains a JSON array with a list ofLeaf Revisions that are needed to be fetched. If the specified Revisionexists then the Document MUST be returned for this Revision. Otherwise,Source MUST return an object with the single field missing with themissed Revision as the value. In case the Document contains attachments,Source MUST return information only for those ones that had been changed(added or updated) since the specified Revision values. If an attachmentwas deleted, the Document MUST NOT have stub information for it
  • latest=true: Ensures, that Source will return the latest Document Revisionregardless of which one was specified in the openrevs query parameter.This parameter solves a race condition problem where the requested Documentmay be changed in between this step and handling related events on theChanges Feed
    In the response Source SHOULD return _multipart/mixed
    or respondinstead with application/json unless the Accept headerspecifies a different mime type. The multipart/mixed content typeallows handling the response data as a stream, since there could be multipledocuments (one per each Leaf Revision) plus several attachments. Theseattachments are mostly binary and JSON has no way to handle such data except asbase64 encoded strings which are very ineffective for transfer and processingoperations.

With a multipart/mixed response the Replicator handles multipleDocument Leaf Revisions and their attachments one by one as raw data withoutany additional encoding applied. There is also one agreement to make dataprocessing more effective: the Document ALWAYS goes before its attachments, sothe Replicator has no need to process all the data to map relatedDocuments-Attachments and may handle it as stream with lesser memory footprint.

Request:

  1. GET /source/SpaghettiWithMeatballs?revs=true&open_revs=[%225-00ecbbc%22,%221-917fa23%22,%223-6bcedf1%22]&latest=true HTTP/1.1
    Accept: multipart/mixed
    Host: localhost:5984
    User-Agent: CouchDB


Response:

  1. HTTP/1.1 200 OK
    Content-Type: multipart/mixed; boundary="7b1596fc4940bc1be725ad67f11ec1c4"
    Date: Thu, 07 Nov 2013 15:10:16 GMT
    Server: CouchDB (Erlang OTP)
    Transfer-Encoding: chunked

    7b1596fc4940bc1be725ad67f11ec1c4
    Content-Type: application/json

    {
    "_id": "SpaghettiWithMeatballs",
    "_rev": "1-917fa23",
    "_revisions": {
    "ids": [
    "917fa23"
    ],
    "start": 1
    },
    "description": "An Italian-American delicious dish",
    "ingredients": [
    "spaghetti",
    "tomato sauce",
    "meatballs"
    ],
    "name": "Spaghetti with meatballs"
    }
    7b1596fc4940bc1be725ad67f11ec1c4
    Content-Type: multipart/related; boundary="a81a77b0ca68389dda3243a43ca946f2"

    a81a77b0ca68389dda3243a43ca946f2
    Content-Type: application/json

    {
    "_attachments": {
    "recipe.txt": {
    "content_type": "text/plain",
    "digest": "md5-R5CrCb6fX10Y46AqtNn0oQ==",
    "follows": true,
    "length": 87,
    "revpos": 7
    }
    },
    "_id": "SpaghettiWithMeatballs",
    "_rev": "7-474f12e",
    "_revisions": {
    "ids": [
    "474f12e",
    "5949cfc",
    "00ecbbc",
    "fc997b6",
    "3552c87",
    "404838b",
    "5defd9d",
    "dc1e4be"
    ],
    "start": 7
    },
    "description": "An Italian-American delicious dish",
    "ingredients": [
    "spaghetti",
    "tomato sauce",
    "meatballs",
    "love"
    ],
    "name": "Spaghetti with meatballs"
    }
    a81a77b0ca68389dda3243a43ca946f2
    Content-Disposition: attachment; filename="recipe.txt"
    Content-Type: text/plain
    Content-Length: 87

    1. Cook spaghetti
    2. Cook meetballs
    3. Mix them
    4. Add tomato sauce
    5.
    6. PROFIT!

    a81a77b0ca68389dda3243a43ca946f2
    7b1596fc4940bc1be725ad67f11ec1c4
    Content-Type: application/json; error="true"

    {"missing":"3-6bcedf1"}
    7b1596fc4940bc1be725ad67f11ec1c4


After receiving the response, the Replicator puts all the received data into alocal stack for further bulk upload to utilize network bandwidth effectively.The local stack size could be limited by number of Documents or bytes ofhandled JSON data. When the stack is full the Replicator uploads all thehandled Document in bulk mode to the Target. While bulk operations are highlyRECOMMENDED to be used, in certain cases the Replicator MAY upload Documents toTarget one by one.

Note

Alternative Replicator implementations MAY use alternative ways to retrieveDocuments from Source. For instance, PouchDB doesn’t use the MultipartAPIand fetches only the latest Document Revision with inline attachments as asingleJSON object. While this is still valid CouchDB HTTP API usage, suchsolutions MAY require a different API implementation for non-CouchDBPeers.

5.2.2.5.2. Upload Batch of Changed Documents

To upload multiple Documents in a single shot the Replicator sends aPOST /{db}/_bulk_docs request to Target with payload containing a JSON objectwith the following mandatory fields:

  • docs (array of objects): List of Document objects to update on Target.These Documents MUST contain the _revisions field that holds a list of thefull Revision history to let Target create Leaf Revisions that correctlypreserve ancestry
  • new_edits (boolean): Special flag that instructs Target to storeDocuments with the specified Revision (field _rev) value as-is withoutgenerating a new revision. Always false
    The request also MAY contain X-Couch-Full-Commit that controlsCouchDB commit policy. Other PeersMAY ignore this header or use it to control similar localfeature.
Request:

  1. POST /target/_bulk_docs HTTP/1.1
    Accept: application/json
    Content-Length: 826
    Content-Type:application/json
    Host: localhost:5984
    User-Agent: CouchDB
    X-Couch-Full-Commit: false

    {
    "docs": [
    {
    "_id": "SpaghettiWithMeatballs",
    "_rev": "1-917fa2381192822767f010b95b45325b",
    "_revisions": {
    "ids": [
    "917fa2381192822767f010b95b45325b"
    ],
    "start": 1
    },
    "description": "An Italian-American delicious dish",
    "ingredients": [
    "spaghetti",
    "tomato sauce",
    "meatballs"
    ],
    "name": "Spaghetti with meatballs"
    },
    {
    "_id": "LambStew",
    "_rev": "1-34c318924a8f327223eed702ddfdc66d",
    "_revisions": {
    "ids": [
    "34c318924a8f327223eed702ddfdc66d"
    ],
    "start": 1
    },
    "servings": 6,
    "subtitle": "Delicious with scone topping",
    "title": "Lamb Stew"
    },
    {
    "_id": "FishStew",
    "_rev": "1-9c65296036141e575d32ba9c034dd3ee",
    "_revisions": {
    "ids": [
    "9c65296036141e575d32ba9c034dd3ee"
    ],
    "start": 1
    },
    "servings": 4,
    "subtitle": "Delicious with fresh bread",
    "title": "Fish Stew"
    }
    ],
    "new_edits": false
    }


In its response Target MUST return a JSON array with a list of Document updatestatuses. If the Document has been stored successfully, the list item MUSTcontain the field ok with true value. Otherwise it MUST containerror and reason fields with error type and a human-friendly reasondescription.

Document updating failure isn’t fatal as Target MAY reject the update for itsown reasons. It’s RECOMMENDED to use error type forbidden for rejections,but other error types can also be used (like invalid field name etc.). TheReplicator SHOULD NOT retry uploading rejected documents unless there aregood reasons for doing so (e.g. there is special error type for that).

Note that while a update may fail for one Document in the response,Target can still return a 201 Created response. Same will be true if allupdates fail for all uploaded Documents.

Response:

  1. HTTP/1.1 201 Created
    Cache-Control: must-revalidate
    Content-Length: 246
    Content-Type: application/json
    Date: Sun, 10 Nov 2013 19:02:26 GMT
    Server: CouchDB (Erlang/OTP)

    [
    {
    "ok": true,
    "id": "SpaghettiWithMeatballs",
    "rev":" 1-917fa2381192822767f010b95b45325b"
    },
    {
    "ok": true,
    "id": "FishStew",
    "rev": "1-9c65296036141e575d32ba9c034dd3ee"
    },
    {
    "error": "forbidden",
    "id": "LambStew",
    "reason": "sorry",
    "rev": "1-34c318924a8f327223eed702ddfdc66d"
    }
    ]


5.2.2.5.3. Upload Document with Attachments

There is a special optimization case when then Replicator WILL NOT use bulkupload of changed Documents. This case is applied when Documents contain alot of attached files or the files are too big to be efficiently encoded withBase64.

Note

CouchDB defines a limit of 8 attachments per Document and each attachedfile size should not be greater than 64 KiB. While this is a RECOMMENDEDlimitation, other Replicator implementations MAY have their own values.

For this case the Replicator issues a /{db}/{docid}?new_edits=false request with multipart/related content type. Sucha request allows one to easily stream the Document and all its attachmentsone by one without any serialization overhead.

Request:

  1. PUT /target/SpaghettiWithMeatballs?new_edits=false HTTP/1.1
    Accept: application/json
    Content-Length: 1030
    Content-Type: multipart/related; boundary="864d690aeb91f25d469dec6851fb57f2"
    Host: localhost:5984
    User-Agent: CouchDB

    2fa48cba80d0cdba7829931fe8acce9d
    Content-Type: application/json

    {
    "_attachments": {
    "recipe.txt": {
    "content_type": "text/plain",
    "digest": "md5-R5CrCb6fX10Y46AqtNn0oQ==",
    "follows": true,
    "length": 87,
    "revpos": 7
    }
    },
    "_id": "SpaghettiWithMeatballs",
    "_rev": "7-474f12eb068c717243487a9505f6123b",
    "_revisions": {
    "ids": [
    "474f12eb068c717243487a9505f6123b",
    "5949cfcd437e3ee22d2d98a26d1a83bf",
    "00ecbbc54e2a171156ec345b77dfdf59",
    "fc997b62794a6268f2636a4a176efcd6",
    "3552c87351aadc1e4bea2461a1e8113a",
    "404838bc2862ce76c6ebed046f9eb542",
    "5defd9d813628cea6e98196eb0ee8594"
    ],
    "start": 7
    },
    "description": "An Italian-American delicious dish",
    "ingredients": [
    "spaghetti",
    "tomato sauce",
    "meatballs",
    "love"
    ],
    "name": "Spaghetti with meatballs"
    }
    2fa48cba80d0cdba7829931fe8acce9d
    Content-Disposition: attachment; filename="recipe.txt"
    Content-Type: text/plain
    Content-Length: 87

    1. Cook spaghetti
    2. Cook meetballs
    3. Mix them
    4. Add tomato sauce
    5.
    6. PROFIT!

    2fa48cba80d0cdba7829931fe8acce9d


Response:

  1. HTTP/1.1 201 Created
    Cache-Control: must-revalidate
    Content-Length: 105
    Content-Type: application/json
    Date: Fri, 08 Nov 2013 16:35:27 GMT
    Server: CouchDB (Erlang/OTP)

    {
    "ok": true,
    "id": "SpaghettiWithMeatballs",
    "rev": "7-474f12eb068c717243487a9505f6123b"
    }


Unlike bulk updating via POST /{db}/_bulk_docs endpoint, the response MAYcome with a different status code. For instance, in the case when the Documentis rejected, Target SHOULD respond with a 403 Forbidden:

Response:

  1. HTTP/1.1 403 Forbidden
    Cache-Control: must-revalidate
    Content-Length: 39
    Content-Type: application/json
    Date: Fri, 08 Nov 2013 16:35:27 GMT
    Server: CouchDB (Erlang/OTP)

    {
    "error": "forbidden",
    "reason": "sorry"
    }


Replicator SHOULD NOT retry requests in case of a 401 Unauthorized,403 Forbidden, 409 Conflict or 412 Precondition Failed since repeatingthe request couldn’t solve the issue with user credentials or uploaded data.

5.2.2.5.4. Ensure In Commit

Once a batch of changes has been successfully uploaded to Target, theReplicator issues a POST /{db}/_ensure_full_commit request to ensure thatevery transferred bit is laid down on disk or other persistent storage place.Target MUST return 201 Created response with a JSON object containing thefollowing mandatory fields:

  • instance_start_time (string): Timestamp of when the database wasopened, expressed in microseconds since the epoch

  • ok (boolean): Operation status. Constantly true

Request:

  1. POST /target/_ensure_full_commit HTTP/1.1
  2. Accept: application/json
  3. Content-Type: application/json
  4. Host: localhost:5984

Response:

  1. HTTP/1.1 201 Created
  2. Cache-Control: must-revalidate
  3. Content-Length: 53
  4. Content-Type: application/json
  5. Date: Web, 06 Nov 2013 18:20:43 GMT
  6. Server: CouchDB (Erlang/OTP)
  7.  
  8. {
  9. "instance_start_time": "0",
  10. "ok": true
  11. }

5.2.2.5.5. Record Replication Checkpoint

Since batches of changes were uploaded and committed successfully, theReplicator updates the Replication Log both on Source and Target recordingthe current Replication state. This operation is REQUIRED so that in the caseof Replication failure the replication can resume from last point of success,not from the very beginning.

Replicator updates Replication Log on Source:

Request:

  1. PUT /source/_local/afa899a9e59589c3d4ce5668e3218aef HTTP/1.1
    Accept: application/json
    Content-Length: 591
    Content-Type: application/json
    Host: localhost:5984
    User-Agent: CouchDB

    {
    "_id": "_local/afa899a9e59589c3d4ce5668e3218aef",
    "_rev": "0-1",
    "_revisions": {
    "ids": [
    "31f36e40158e717fbe9842e227b389df"
    ],
    "start": 1
    },
    "history": [
    {
    "doc_write_failures": 0,
    "docs_read": 6,
    "docs_written": 6,
    "end_last_seq": 26,
    "end_time": "Thu, 07 Nov 2013 09:42:17 GMT",
    "missing_checked": 6,
    "missing_found": 6,
    "recorded_seq": 26,
    "session_id": "04bf15bf1d9fa8ac1abc67d0c3e04f07",
    "start_last_seq": 0,
    "start_time": "Thu, 07 Nov 2013 09:41:43 GMT"
    }
    ],
    "replication_id_version": 3,
    "session_id": "04bf15bf1d9fa8ac1abc67d0c3e04f07",
    "source_last_seq": 26
    }


Response:

  1. HTTP/1.1 201 Created
    Cache-Control: must-revalidate
    Content-Length: 75
    Content-Type: application/json
    Date: Thu, 07 Nov 2013 09:42:17 GMT
    Server: CouchDB (Erlang/OTP)

    {
    "id": "_local/afa899a9e59589c3d4ce5668e3218aef",
    "ok": true,
    "rev": "0-2"
    }


…and on Target too:

Request:

  1. PUT /target/_local/afa899a9e59589c3d4ce5668e3218aef HTTP/1.1
    Accept: application/json
    Content-Length: 591
    Content-Type: application/json
    Host: localhost:5984
    User-Agent: CouchDB

    {
    "_id": "_local/afa899a9e59589c3d4ce5668e3218aef",
    "_rev": "1-31f36e40158e717fbe9842e227b389df",
    "_revisions": {
    "ids": [
    "31f36e40158e717fbe9842e227b389df"
    ],
    "start": 1
    },
    "history": [
    {
    "doc_write_failures": 0,
    "docs_read": 6,
    "docs_written": 6,
    "end_last_seq": 26,
    "end_time": "Thu, 07 Nov 2013 09:42:17 GMT",
    "missing_checked": 6,
    "missing_found": 6,
    "recorded_seq": 26,
    "session_id": "04bf15bf1d9fa8ac1abc67d0c3e04f07",
    "start_last_seq": 0,
    "start_time": "Thu, 07 Nov 2013 09:41:43 GMT"
    }
    ],
    "replication_id_version": 3,
    "session_id": "04bf15bf1d9fa8ac1abc67d0c3e04f07",
    "source_last_seq": 26
    }


Response:

  1. HTTP/1.1 201 Created
    Cache-Control: must-revalidate
    Content-Length: 106
    Content-Type: application/json
    Date: Thu, 07 Nov 2013 09:42:17 GMT
    Server: CouchDB (Erlang/OTP)

    {
    "id": "_local/afa899a9e59589c3d4ce5668e3218aef",
    "ok": true,
    "rev": "2-9b5d1e36bed6ae08611466e30af1259a"
    }


5.2.2.6. Continue Reading Changes

Once a batch of changes had been processed and transferred to Targetsuccessfully, the Replicator can continue to listen to the Changes Feed for newchanges. If there are no new changes to process the Replication is consideredto be done.

For Continuous Replication, the Replicator MUST continue to wait for new changesfrom Source.

5.2.3. Protocol Robustness

Since the CouchDB Replication Protocol works on top of HTTP, which is based onTCP/IP, the Replicator SHOULD expect to be working within an unstableenvironment with delays, losses and other bad surprises that might eventuallyoccur. The Replicator SHOULD NOT count every HTTP request failure as a fatalerror. It SHOULD be smart enough to detect timeouts, repeat failed requests,be ready to process incomplete or malformed data and so on. Data must flow- that’s the rule.

5.2.4. Error Responses

In case something goes wrong the Peer MUST respond with a JSON object withthe following REQUIRED fields:

  • error (string): Error type for programs and developers
  • reason (string): Error description for humans

5.2.4.1. Bad Request

If a request contains malformed data (like invalid JSON) the Peer MUST respondwith a HTTP 400 Bad Request and bad_request as error type:

  1. {
  2. "error": "bad_request",
  3. "reason": "invalid json"
  4. }

5.2.4.2. Unauthorized

If a Peer REQUIRES credentials be included with the request and the requestdoes not contain acceptable credentials then the Peer MUST respond with theHTTP 401 Unauthorized and unauthorized as error type:

  1. {
  2. "error": "unauthorized",
  3. "reason": "Name or password is incorrect"
  4. }

5.2.4.3. Forbidden

If a Peer receives valid user credentials, but the requester does not havesufficient permissions to perform the operation then the PeerMUST respond with a HTTP 403 Forbidden and forbidden as error type:

  1. {
  2. "error": "forbidden",
  3. "reason": "You may only update your own user document."
  4. }

5.2.4.4. Resource Not Found

If the requested resource, Database or Document wasn’t found on a Peer, the PeerMUST respond with a HTTP 404 Not Found and not_found as error type:

  1. {
  2. "error": "not_found",
  3. "reason": "database \"target\" does not exists"
  4. }

5.2.4.5. Method Not Allowed

If an unsupported method was used then the Peer MUST respond with aHTTP 405 Method Not Allowed and method_not_allowed as error type:

  1. {
  2. "error": "method_not_allowed",
  3. "reason": "Only GET, PUT, DELETE allowed"
  4. }

5.2.4.6. Resource Conflict

A resource conflict error occurs when there are concurrent updates of the sameresource by multiple clients. In this case the Peer MUST respond with a HTTP409 Conflict and conflict as error type:

  1. {
  2. "error": "conflict",
  3. "reason": "document update conflict"
  4. }

5.2.4.7. Precondition Failed

The HTTP 412 Precondition Failed response may be sent in case of an attempt tocreate a Database (error type db_exists) that already existsor some attachment information is missing (error type missing_stub).There is no explicit error type restrictions, but it is RECOMMEND to use errortypes that are previously mentioned:

  1. {
  2. "error": "db_exists",
  3. "reason": "database \"target\" exists"
  4. }

5.2.4.8. Server Error

Raised in case an error is fatal and the Replicator cannot do anything tocontinue Replication. In this case the Replicator MUST return a HTTP500 Internal Server Error response with an error description (no restrictions on errortype applied):

  1. {
  2. "error": "worker_died",
  3. "reason": "kaboom!"
  4. }

5.2.5. Optimisations

There are RECOMMENDED approaches to optimize the Replication process:

  • Keep the number of HTTP requests at a reasonable minimum
  • Try to work with a connection pool and make parallel/multiple requestswhenever possible
  • Don’t close sockets after each request: respect the keep-alive option
  • Use continuous sessions (cookies, etc.) to reduce authentication overhead
  • Try to use bulk requests for every operations with Documents
  • Find out optimal batch size for Changes feed processing
  • Preserve Replication Logs and resume Replication from the last Checkpointwhenever possible
  • Optimize filter functions: let them run as fast as possible
  • Get ready for surprises: networks are very unstable environments

5.2.6. API Reference

5.2.6.1. Common Methods

5.2.6.2. For Target

5.2.6.3. For Source

5.2.7. Reference

原文: http://docs.couchdb.org/en/stable/replication/protocol.html