3.1. Query Server Protocol

A Query Server is an external process that communicates with CouchDB via a simple, custom JSON protocol over stdin/stdout. It is used to processes all design functions calls: views, shows, lists, filters, updates and validate_doc_update.

CouchDB communicates with the Query Server process through stdin/stdout with JSON messages that are terminated by a newline character. Messages that are sent to the Query Server are always array-typed and follow the pattern [<command>, <*arguments>]\n.

3.1.1. reset

Command:reset
Arguments:Query server state (optional)
Returns:true

This resets the state of the Query Server and makes it forget all previous input. If applicable, this is the point to run garbage collection.

CouchDB sends:

  1. ["reset"]

The Query Server answers:

  1. true

To set up new Query Server state, the second argument is used with object data.

CouchDB sends:

  1. ["reset", {"reduce_limit": true, "timeout": 5000}]

The Query Server answers:

  1. true

3.1.2. add_lib

Command:add_lib
Arguments:CommonJS library object by views/lib path
Returns:true

Adds CommonJS library to Query Server state for further usage in map functions.

CouchDB sends:

  1. [
  2.     "add_lib",
  3.     {
  4.         "utils": "exports.MAGIC = 42;"
  5.     }
  6. ]

The Query Server answers:

  1. true

3.1.3. add_fun

Command:add_fun
Arguments:Map function source code.
Returns:true

When creating or updating a view, this is how the Query Server is sent the view function for evaluation. The Query Server should parse, compile, and evaluate the function it receives to make it callable later. If this fails, the Query Server returns an error. CouchDB may store multiple functions before sending any documents.

CouchDB sends:

  1. [
  2.     "add_fun",
  3.     "function(doc) { if(doc.score > 50) emit(null, {'player_name': doc.name}); }"
  4. ]

The Query Server answers:

  1. true

3.1.4. map_doc

Command:map_doc
Arguments:Document object
Returns:Array of key-value pairs per applied function

When the view function is stored in the Query Server, CouchDB starts sending all the documents in the database, one at a time. The Query Server calls the previously stored functions one after another with a document and stores its result. When all functions have been called, the result is returned as a JSON string.

CouchDB sends:

  1. [
  2.     "map_doc",
  3.     {
  4.         "_id": "8877AFF9789988EE",
  5.         "_rev": "3-235256484",
  6.         "name": "John Smith",
  7.         "score": 60
  8.     }
  9. ]

If the function above is the only function stored, the Query Server answers:

  1. [
  2.     [
  3.         [null, {"player_name": "John Smith"}]
  4.     ]
  5. ]

That is, an array with the result for every function for the given document.

If a document is to be excluded from the view, the array should be empty.

CouchDB sends:

  1. [
  2.     "map_doc",
  3.     {
  4.         "_id": "9590AEB4585637FE",
  5.         "_rev": "1-674684684",
  6.         "name": "Jane Parker",
  7.         "score": 43
  8.     }
  9. ]

The Query Server answers:

  1. [[]]

3.1.5. reduce

Command:

reduce

Arguments:
  • Reduce function source
  • Array of map function results where each item represented in format [[key, id-of-doc], value]
Returns:

Array with pair values: true and another array with reduced result

If the view has a reduce function defined, CouchDB will enter into the reduce phase. The Query Server will receive a list of reduce functions and some map results on which it can apply them.

CouchDB sends:

  1. [
  2.     "reduce",
  3.     [
  4.         "function(k, v) { return sum(v); }"
  5.     ],
  6.     [
  7.         [[1, "699b524273605d5d3e9d4fd0ff2cb272"], 10],
  8.         [[2, "c081d0f69c13d2ce2050d684c7ba2843"], 20],
  9.         [[null, "foobar"], 3]
  10.     ]
  11. ]

The Query Server answers:

  1. [
  2.     true,
  3.     [33]
  4. ]

Note that even though the view server receives the map results in the form [[key, id-of-doc], value], the function may receive them in a different form. For example, the JavaScript Query Server applies functions on the list of keys and the list of values.

3.1.6. rereduce

Command:

rereduce

Arguments:
  • Reduce function source
  • List of values

When building a view, CouchDB will apply the reduce step directly to the output of the map step and the rereduce step to the output of a previous reduce step.

CouchDB will send a list of reduce functions and a list of values, with no keys or document ids to the rereduce step.

CouchDB sends:

  1. [
  2.     "rereduce",
  3.     [
  4.         "function(k, v, r) { return sum(v); }"
  5.     ],
  6.     [
  7.         33,
  8.         55,
  9.         66
  10.     ]
  11. ]

The Query Server answers:

  1. [
  2.     true,
  3.     [154]
  4. ]

3.1.7. ddoc

Command:

ddoc

Arguments:

Array of objects.

  • First phase (ddoc initialization):
    • “new”
    • Design document _id
    • Design document object
  • Second phase (design function execution):
    • Design document _id
    • Function path as an array of object keys
    • Array of function arguments
Returns:
  • First phase (ddoc initialization): true
  • Second phase (design function execution): custom object depending on executed function

This command acts in two phases: ddoc registration and design function execution.

In the first phase CouchDB sends a full design document content to the Query Server to let it cache it by _id value for further function execution.

To do this, CouchDB sends:

  1. [
  2.     "ddoc",
  3.     "new",
  4.     "_design/temp",
  5.     {
  6.         "_id": "_design/temp",
  7.         "_rev": "8-d7379de23a751dc2a19e5638a7bbc5cc",
  8.         "language": "javascript",
  9.         "shows": {
  10.             "request": "function(doc,req){ return {json: req}; }",
  11.             "hello": "function(doc,req){ return {body: 'Hello, ' + (doc || {})._id + '!'}; }"
  12.         }
  13.     }
  14. ]

The Query Server answers:

  1. true

After this, the design document will be ready to serve subcommands in the second phase.

3.1.7.1. shows

Command:

ddoc

SubCommand:

shows

Arguments:
  • Document object or null if document id isn’t specified in request
  • Request object
Returns:

Array with two elements:

Executes show function.

Couchdb sends:

  1. [
  2.     "ddoc",
  3.     "_design/temp",
  4.     [
  5.         "shows",
  6.         "doc"
  7.     ],
  8.     [
  9.         null,
  10.         {
  11.             "info": {
  12.                 "db_name": "test",
  13.                 "doc_count": 8,
  14.                 "doc_del_count": 0,
  15.                 "update_seq": 105,
  16.                 "purge_seq": 0,
  17.                 "compact_running": false,
  18.                 "sizes": {
  19.                   "active": 1535048,
  20.                   "disk": 15818856,
  21.                   "external": 15515850
  22.                 },
  23.                 "instance_start_time": "1359952188595857",
  24.                 "disk_format_version": 6,
  25.                 "committed_update_seq": 105
  26.             },
  27.             "id": null,
  28.             "uuid": "169cb4cc82427cc7322cb4463d0021bb",
  29.             "method": "GET",
  30.             "requested_path": [
  31.                 "api",
  32.                 "_design",
  33.                 "temp",
  34.                 "_show",
  35.                 "request"
  36.             ],
  37.             "path": [
  38.                 "api",
  39.                 "_design",
  40.                 "temp",
  41.                 "_show",
  42.                 "request"
  43.             ],
  44.             "raw_path": "/api/_design/temp/_show/request",
  45.             "query": {},
  46.             "headers": {
  47.                 "Accept": "*/*",
  48.                 "Host": "localhost:5984",
  49.                 "User-Agent": "curl/7.26.0"
  50.             },
  51.             "body": "undefined",
  52.             "peer": "127.0.0.1",
  53.             "form": {},
  54.             "cookie": {},
  55.             "userCtx": {
  56.                 "db": "api",
  57.                 "name": null,
  58.                 "roles": [
  59.                     "_admin"
  60.                 ]
  61.             },
  62.             "secObj": {}
  63.         }
  64.     ]
  65. ]

The Query Server sends:

  1. [
  2.     "resp",
  3.     {
  4.         "body": "Hello, undefined!"
  5.     }
  6. ]

3.1.7.2. lists

Command:

ddoc

SubCommand:

lists

Arguments:
Returns:

Array. See below for details.

Executes list function.

The communication protocol for list functions is a bit complex so let’s use an example to illustrate.

Assume we have view a function that emits id-rev pairs:

  1. function(doc) {
  2.     emit(doc._id, doc._rev);
  3. }

And we’d like to emulate _all_docs JSON response with list function. Our first version of the list functions looks like this:

  1. function(head, req){
  2.     start({'headers': {'Content-Type': 'application/json'}});
  3.     var resp = head;
  4.     var rows = [];
  5.     while(row=getRow()){
  6.         rows.push(row);
  7.     }
  8.     resp.rows = rows;
  9.     return toJSON(resp);
  10. }

The whole communication session during list function execution could be divided on three parts:

  1. Initialization

    The first returned object from the list function is an array with the following structure:

    1. ["start", <chunks>, <headers>]

    Where <chunks> is an array of text chunks that will be sent to the client and <headers> is an object with response HTTP headers.

    This message is sent from the Query Server to CouchDB on the start() call which initializes the HTTP response to the client:

    1. [
    2.     "start",
    3.     [],
    4.     {
    5.         "headers": {
    6.             "Content-Type": "application/json"
    7.         }
    8.     }
    9. ]

    After this, the list function may start to process view rows.

  2. View Processing

    Since view results can be extremely large, it is not wise to pass all its rows in a single command. Instead, CouchDB can send view rows one by one to the Query Server allowing view processing and output generation to be processed as a stream.

    CouchDB sends a special array that carries view row data:

    1. [
    2.     "list_row",
    3.     {
    4.         "id": "0cb42c267fe32d4b56b3500bc503e030",
    5.         "key": "0cb42c267fe32d4b56b3500bc503e030",
    6.         "value": "1-967a00dff5e02add41819138abb3284d"
    7.     }
    8. ]

    If the Query Server has something to return on this, it returns an array with a "chunks" item in the head and an array of data in the tail. For this example it has nothing to return, so the response will be:

    1. [
    2.   "chunks",
    3.   []
    4. ]

    When there are no more view rows to process, CouchDB sends a list_end message to signify there is no more data to send:

    1. ["list_end"]

  3. Finalization

    The last stage of the communication process is the returning list tail: the last data chunk. After this, processing of the list function will be complete and the client will receive a complete response.

    For our example the last message is:

    1. [
    2.     "end",
    3.     [
    4.         "{\"total_rows\":2,\"offset\":0,\"rows\":[{\"id\":\"0cb42c267fe32d4b56b3500bc503e030\",\"key\":\"0cb42c267fe32d4b56b3500bc503e030\",\"value\":\"1-967a00dff5e02add41819138abb3284d\"},{\"id\":\"431926a69504bde41851eb3c18a27b1f\",\"key\":\"431926a69504bde41851eb3c18a27b1f\",\"value\":\"1-967a00dff5e02add41819138abb3284d\"}]}"
    5.     ]
    6. ]

In this example, we have returned our result in a single message from the Query Server. This is okay for small numbers of rows, but for large data sets, perhaps with millions of documents or millions of view rows, this would not be acceptable.

Let’s fix our list function and see the changes in communication:

  1. function(head, req){
  2.     start({'headers': {'Content-Type': 'application/json'}});
  3.     send('{');
  4.     send('"total_rows":' + toJSON(head.total_rows) + ',');
  5.     send('"offset":' + toJSON(head.offset) + ',');
  6.     send('"rows":[');
  7.     if (row=getRow()){
  8.         send(toJSON(row));
  9.     }
  10.     while(row=getRow()){
  11.         send(',' + toJSON(row));
  12.     }
  13.     send(']');
  14.     return '}';
  15. }

“Wait, what?” - you’d like to ask. Yes, we’d build JSON response manually by string chunks, but let’s take a look on logs:

  1. [Wed, 24 Jul 2013 05:45:30 GMT] [debug] [<0.19191.1>] OS Process #Port<0.4444> Output :: ["start",["{","\"total_rows\":2,","\"offset\":0,","\"rows\":["],{"headers":{"Content-Type":"application/json"}}]
  2. [Wed, 24 Jul 2013 05:45:30 GMT] [info] [<0.18963.1>] 127.0.0.1 - - GET /blog/_design/post/_list/index/all_docs 200
  3. [Wed, 24 Jul 2013 05:45:30 GMT] [debug] [<0.19191.1>] OS Process #Port<0.4444> Input  :: ["list_row",{"id":"0cb42c267fe32d4b56b3500bc503e030","key":"0cb42c267fe32d4b56b3500bc503e030","value":"1-967a00dff5e02add41819138abb3284d"}]
  4. [Wed, 24 Jul 2013 05:45:30 GMT] [debug] [<0.19191.1>] OS Process #Port<0.4444> Output :: ["chunks",["{\"id\":\"0cb42c267fe32d4b56b3500bc503e030\",\"key\":\"0cb42c267fe32d4b56b3500bc503e030\",\"value\":\"1-967a00dff5e02add41819138abb3284d\"}"]]
  5. [Wed, 24 Jul 2013 05:45:30 GMT] [debug] [<0.19191.1>] OS Process #Port<0.4444> Input  :: ["list_row",{"id":"431926a69504bde41851eb3c18a27b1f","key":"431926a69504bde41851eb3c18a27b1f","value":"1-967a00dff5e02add41819138abb3284d"}]
  6. [Wed, 24 Jul 2013 05:45:30 GMT] [debug] [<0.19191.1>] OS Process #Port<0.4444> Output :: ["chunks",[",{\"id\":\"431926a69504bde41851eb3c18a27b1f\",\"key\":\"431926a69504bde41851eb3c18a27b1f\",\"value\":\"1-967a00dff5e02add41819138abb3284d\"}"]]
  7. [Wed, 24 Jul 2013 05:45:30 GMT] [debug] [<0.19191.1>] OS Process #Port<0.4444> Input  :: ["list_end"]
  8. [Wed, 24 Jul 2013 05:45:30 GMT] [debug] [<0.19191.1>] OS Process #Port<0.4444> Output :: ["end",["]","}"]]

Note, that now the Query Server sends response by lightweight chunks and if our communication process was extremely slow, the client will see how response data appears on their screen. Chunk by chunk, without waiting for the complete result, like they have for our previous list function.

3.1.7.3. updates

Command:

ddoc

SubCommand:

updates

Arguments:
  • Document object or null if document id wasn’t specified in request
  • Request object
Returns:

Array with there elements:

Executes update function.

CouchDB sends:

  1. [
  2.     "ddoc",
  3.     "_design/id",
  4.     [
  5.         "updates",
  6.         "nothing"
  7.     ],
  8.     [
  9.         null,
  10.         {
  11.             "info": {
  12.                 "db_name": "test",
  13.                 "doc_count": 5,
  14.                 "doc_del_count": 0,
  15.                 "update_seq": 16,
  16.                 "purge_seq": 0,
  17.                 "compact_running": false,
  18.                 "sizes": {
  19.                   "active": 7979745,
  20.                   "disk": 8056936,
  21.                   "external": 8024930
  22.                 },
  23.                 "instance_start_time": "1374612186131612",
  24.                 "disk_format_version": 6,
  25.                 "committed_update_seq": 16
  26.             },
  27.             "id": null,
  28.             "uuid": "7b695cb34a03df0316c15ab529002e69",
  29.             "method": "POST",
  30.             "requested_path": [
  31.                 "test",
  32.                 "_design",
  33.                 "1139",
  34.                 "_update",
  35.                 "nothing"
  36.             ],
  37.             "path": [
  38.                 "test",
  39.                 "_design",
  40.                 "1139",
  41.                 "_update",
  42.                 "nothing"
  43.             ],
  44.             "raw_path": "/test/_design/1139/_update/nothing",
  45.             "query": {},
  46.             "headers": {
  47.                 "Accept": "*/*",
  48.                 "Accept-Encoding": "identity, gzip, deflate, compress",
  49.                 "Content-Length": "0",
  50.                 "Host": "localhost:5984"
  51.             },
  52.             "body": "",
  53.             "peer": "127.0.0.1",
  54.             "form": {},
  55.             "cookie": {},
  56.             "userCtx": {
  57.                 "db": "test",
  58.                 "name": null,
  59.                 "roles": [
  60.                     "_admin"
  61.                 ]
  62.             },
  63.             "secObj": {}
  64.         }
  65.     ]
  66. ]

The Query Server answers:

  1. [
  2.     "up",
  3.     null,
  4.     {"body": "document id wasn't provided"}
  5. ]

or in case of successful update:

  1. [
  2.     "up",
  3.     {
  4.         "_id": "7b695cb34a03df0316c15ab529002e69",
  5.         "hello": "world!"
  6.     },
  7.     {"body": "document was updated"}
  8. ]

3.1.7.4. filters

Command:

ddoc

SubCommand:

filters

Arguments:
Returns:

Array of two elements:

  • true
  • Array of booleans in the same order of input documents.

Executes filter function.

CouchDB sends:

  1. [
  2.     "ddoc",
  3.     "_design/test",
  4.     [
  5.         "filters",
  6.         "random"
  7.     ],
  8.     [
  9.         [
  10.             {
  11.                 "_id": "431926a69504bde41851eb3c18a27b1f",
  12.                 "_rev": "1-967a00dff5e02add41819138abb3284d",
  13.                 "_revisions": {
  14.                     "start": 1,
  15.                     "ids": [
  16.                         "967a00dff5e02add41819138abb3284d"
  17.                     ]
  18.                 }
  19.             },
  20.             {
  21.                 "_id": "0cb42c267fe32d4b56b3500bc503e030",
  22.                 "_rev": "1-967a00dff5e02add41819138abb3284d",
  23.                 "_revisions": {
  24.                     "start": 1,
  25.                     "ids": [
  26.                         "967a00dff5e02add41819138abb3284d"
  27.                     ]
  28.                 }
  29.             }
  30.         ],
  31.         {
  32.             "info": {
  33.                 "db_name": "test",
  34.                 "doc_count": 5,
  35.                 "doc_del_count": 0,
  36.                 "update_seq": 19,
  37.                 "purge_seq": 0,
  38.                 "compact_running": false,
  39.                 "sizes": {
  40.                   "active": 7979745,
  41.                   "disk": 8056936,
  42.                   "external": 8024930
  43.                 },
  44.                 "instance_start_time": "1374612186131612",
  45.                 "disk_format_version": 6,
  46.                 "committed_update_seq": 19
  47.             },
  48.             "id": null,
  49.             "uuid": "7b695cb34a03df0316c15ab529023a81",
  50.             "method": "GET",
  51.             "requested_path": [
  52.                 "test",
  53.                 "_changes?filter=test",
  54.                 "random"
  55.             ],
  56.             "path": [
  57.                 "test",
  58.                 "_changes"
  59.             ],
  60.             "raw_path": "/test/_changes?filter=test/random",
  61.             "query": {
  62.                 "filter": "test/random"
  63.             },
  64.             "headers": {
  65.                 "Accept": "application/json",
  66.                 "Accept-Encoding": "identity, gzip, deflate, compress",
  67.                 "Content-Length": "0",
  68.                 "Content-Type": "application/json; charset=utf-8",
  69.                 "Host": "localhost:5984"
  70.             },
  71.             "body": "",
  72.             "peer": "127.0.0.1",
  73.             "form": {},
  74.             "cookie": {},
  75.             "userCtx": {
  76.                 "db": "test",
  77.                 "name": null,
  78.                 "roles": [
  79.                     "_admin"
  80.                 ]
  81.             },
  82.             "secObj": {}
  83.         }
  84.     ]
  85. ]

The Query Server answers:

  1. [
  2.     true,
  3.     [
  4.         true,
  5.         false
  6.     ]
  7. ]

3.1.7.5. views

Command:

ddoc

SubCommand:

views

Arguments:

Array of document objects

Returns:

Array of two elements:

  • true
  • Array of booleans in the same order of input documents.

New in version 1.2.

Executes view function in place of the filter.

Acts in the same way as filters command.

3.1.7.6. validate_doc_update

Command:

ddoc

SubCommand:

validate_doc_update

Arguments:
Returns:

1

Executes validation function.

CouchDB send:

  1. [
  2.     "ddoc",
  3.     "_design/id",
  4.     ["validate_doc_update"],
  5.     [
  6.         {
  7.             "_id": "docid",
  8.             "_rev": "2-e0165f450f6c89dc6b071c075dde3c4d",
  9.             "score": 10
  10.         },
  11.         {
  12.             "_id": "docid",
  13.             "_rev": "1-9f798c6ad72a406afdbf470b9eea8375",
  14.             "score": 4
  15.         },
  16.         {
  17.             "name": "Mike",
  18.             "roles": ["player"]
  19.         },
  20.         {
  21.             "admins": {},
  22.             "members": []
  23.         }
  24.     ]
  25. ]

The Query Server answers:

  1. 1

3.1.7.7. rewrites

Command:

ddoc

SubCommand:

rewrites

Arguments:
Returns:

1

Executes rewrite function.

CouchDB send:

  1. [
  2.     "ddoc",
  3.     "_design/id",
  4.     ["rewrites"],
  5.     [
  6.         {
  7.             "method": "POST",
  8.             "requested_path": [
  9.                 "test",
  10.                 "_design",
  11.                 "1139",
  12.                 "_update",
  13.                 "nothing"
  14.             ],
  15.             "path": [
  16.                 "test",
  17.                 "_design",
  18.                 "1139",
  19.                 "_update",
  20.                 "nothing"
  21.             ],
  22.             "raw_path": "/test/_design/1139/_update/nothing",
  23.             "query": {},
  24.             "headers": {
  25.                 "Accept": "*/*",
  26.                 "Accept-Encoding": "identity, gzip, deflate, compress",
  27.                 "Content-Length": "0",
  28.                 "Host": "localhost:5984"
  29.             },
  30.             "body": "",
  31.             "peer": "127.0.0.1",
  32.             "cookie": {},
  33.             "userCtx": {
  34.                 "db": "test",
  35.                 "name": null,
  36.                 "roles": [
  37.                     "_admin"
  38.                 ]
  39.             },
  40.             "secObj": {}
  41.         }
  42.     ]
  43. ]

The Query Server answers:

  1. [
  2.     "ok",
  3.     {
  4.         "path": "some/path",
  5.         "query": {"key1": "value1", "key2": "value2"},
  6.         "method": "METHOD",
  7.         "headers": {"Header1": "value1", "Header2": "value2"},
  8.         "body": ""
  9.     }
  10. ]

or in case of direct response:

  1. [
  2.     "ok",
  3.     {
  4.         "headers": {"Content-Type": "text/plain"},
  5.         "body": "Welcome!",
  6.         "code": 200
  7.     }
  8. ]

or for immediate redirect:

  1. [
  2.     "ok",
  3.     {
  4.         "headers": {"Location": "http://example.com/path/"},
  5.         "code": 302
  6.     }
  7. ]

3.1.8. Returning errors

When something goes wrong, the Query Server can inform CouchDB by sending a special message in response to the received command.

Error messages prevent further command execution and return an error description to CouchDB. Errors are logically divided into two groups:

  • Common errors. These errors only break the current Query Server command and return the error info to the CouchDB instance without terminating the Query Server process.
  • Fatal errors. Fatal errors signal a condition that cannot be recovered. For instance, if your a design function is unable to import a third party module, it’s better to count such error as fatal and terminate whole process.

3.1.8.1. error

To raise an error, the Query Server should respond with:

  1. ["error", "error_name", "reason why"]

The "error_name" helps to classify problems by their type e.g. if it’s "value_error" to indicate improper data, "not_found" to indicate a missing resource and "type_error" to indicate an improper data type.

The "reason why" explains in human-readable terms what went wrong, and possibly how to resolve it.

For example, calling Update Functions against a non-existent document could produce the error message:

  1. ["error", "not_found", "Update function requires existent document"]

3.1.8.2. forbidden

The forbidden error is widely used by Validate Document Update Functions to stop further function processing and prevent storage of the new document revision. Since this is not actually an error, but an assertion against user actions, CouchDB doesn’t log it at “error” level, but returns HTTP 403 Forbidden response with error information object.

To raise this error, the Query Server should respond with:

  1. {"forbidden": "reason why"}

3.1.8.3. unauthorized

The unauthorized error mostly acts like forbidden one, but with the meaning of please authorize first. This small difference helps end users to understand what they can do to solve the problem. Similar to forbidden, CouchDB doesn’t log it at “error” level, but returns a HTTP 401 Unauthorized response with an error information object.

To raise this error, the Query Server should respond with:

  1. {"unauthorized": "reason why"}

3.1.9. Logging

At any time, the Query Server may send some information that will be saved in CouchDB’s log file. This is done by sending a special log object with a single argument, on a separate line:

  1. ["log", "some message"]

CouchDB does not respond, but writes the received message to the log file:

  1. [Sun, 13 Feb 2009 23:31:30 GMT] [info] [<0.72.0>] Query Server Log Message: some message

These messages are only logged at info level.