couchbase startkey_docid and complex key - couchdb

Question

The call below with the matching result set returns a result that does not start with the doc id a4a6cf44-8a82-494a-a2b9-f6a3ec629f17. As the result set below demonstrates the top 3 keys are identical, but startkey_docid has no effect.
Just a couple questions around this
Will startkey_docid work correctly with a complex key? (This apparently works for couchdb)
If yes to the above, is the source of the issue the use of dateToArray(doc.created) in our view?
Is this a bug?
View:
function (doc, meta)
{
if(meta.type == "json" && doc.type == "POST")
{
emit([doc.category, dateToArray(doc.created), doc.visibility], null);
}
}
Call:
?startkey=["auto",[2013,10,10,23,12,0],"EVERYONE"]&endkey=["auto",[2013,12,11,23,12,0],"EVERYONE"]&startkey_docid=a4a6cf44-8a82-494a-a2b9-f6a3ec629f17
Result:
{
total_rows: 20,
rows: [{
id: "a4a6cf44-8a82",
key: ["auto", [2013, 11, 8, 1, 17, 46], "EVERYONE"],
value: null
}, {
id: "a4a6cf44-8a82-494a-a2b9",
key: ["auto", [2013, 11, 8, 1, 17, 46], "EVERYONE"],
value: null
}, {
id: "a4a6cf44-8a82-494a-a2b9-f6a3ec629f17",
key: ["auto", [2013, 11, 8, 1, 17, 46], "EVERYONE"],
value: null
}, {
id: "41070cfc-a85c-424c-9b87-fce0616c77c1",
key: ["auto", [
2013, 11, 11, 20, 28, 21], "EVERYONE"],
value: null
}, {
id: "bb275e3c-54da-4e85-8cc3-21defff4e278",
key: ["auto", [
2013, 11, 13, 1, 41, 7], "EVERYONE"],
value: null
}]
}
Example data set. post_id is the document id.
{
type: "POST",
post_id: "a4a6cf44-8a82",
visibility: "EVERYONE",
userID: "<user_id>",
title: "Some title 1",
category: "auto",
description: "",
created: "2013-11-07 17:17:46 -0800",
modified: "2013-11-07 17:17:46 -0800"
}, {
type: "POST",
post_id: "a4a6cf44-8a82-494a-a2b9",
visibility: "EVERYONE",
userID: "<user_id>",
title: "Some title 2",
category: "auto",
description: "",
created: "2013-11-07 17:17:46 -0800",
modified: "2013-11-07 17:17:46 -0800"
}, {
type: "POST",
post_id: "a4a6cf44-8a82-494a-a2b9-f6a3ec629f17",
visibility: "EVERYONE",
userID: "<user_id>",
title: "Some title 3",
category: "auto",
description: "",
created: "2013-11-07 17:17:46 -0800",
modified: "2013-11-07 17:17:46 -0800"
}, {
type: "POST",
post_id: "41070cfc-a85c-424c-9b87-fce0616c77c1",
visibility: "EVERYONE",
userID: "<user_id>",
title: "Some title 4",
category: "auto",
description: "",
created: "2013-11-11 12:28:21 -0800",
modified: "2013-11-11 12:28:21 -0800"
}, {
type: "POST",
post_id: "bb275e3c-54da-4e85-8cc3-21defff4e278",
visibility: "EVERYONE",
userID: "<user_id>",
title: "Some title 5",
category: "auto",
description: "",
created: "2013-11-12 17:41:07 -0800",
modified: "2013-11-12 17:41:07 -0800"
}

Answer 1

Okay, this is an old thread, but for reference sake, I'll post my findings and try to explain this functionality the best I can.
Example
Given a database of over 9,000 movie documents, each containing attributes like title, year, rating, genre, etc., define a view MovieIndexView with index name MovieIndex for this database that emits the following complex (but rather simple) key for filtering movies by rating, and sorting by year:
function (doc) {
emit([doc.rating, doc.year], doc);
}
The view can be queried with sorted rows using techniques similar to the way the OP has shown:
- Request -
GET /movies_db/_design/MovieIndex/_view/MovieIndexView?
startkey=["R",{}]
&endkey=["R",0]
&descending=true
&limit=6
- Response -
{
"total_rows": 9411,
"offset": 1,
"rows": [
{
"id": "2802144",
"key": [
"R",
2014
],
"value": {
"_id": "2802144",
"_rev": "1-924e12ba1f1144e3a453bbd8978acc5c",
"title": "Kingsman: The Secret Service",
"year": 2014,
"rating": "R",
"runtime": "129 min",
"genre": [
"Action",
"Adventure",
"Comedy"
],
"director": "Matthew Vaughn"
}
},
<...>
{
"id": "2278388",
"key": [
"R",
2014
],
"value": {
"_id": "2278388",
"_rev": "1-c38b7f5eb43abfd59fb8514277290e46",
"title": "The Grand Budapest Hotel",
"year": 2014,
"rating": "R",
"runtime": "99 min",
"genre": [
"Adventure",
"Comedy",
"Drama"
],
"director": "Wes Anderson"
}
}
]
}
Note we only pulled the first 6 documents by setting the limit variable (I pasted only the first and last relevant items).
Pagination
A first shot at pagination for the next sequence of rows could be straight forward. Simply set the startkey of the next request to the key value of the last item in the previous set, while keeping the value for endkey unchanged:
startkey=["R",2014]
&endkey=["R",0]
&descending=true
&limit=6
Do you see a problem with this approach? This new start key fails to specify which of the multiple movie items with the same value for startkey (movies with the same rating, that came out the same year) the response should start with. While there is nothing syntactically wrong with the request, you will notice that the returned rows will not start with the next sequential item we want, but rather with another arbitrary item in the database with a matching value for key.
This is where startkey_docid comes in handy:
- Request -
GET /movies_db/_design/MovieIndexView/_view/MovieIndex?
startkey=["R",2014]
&endkey=["R",0]
&descending=true
&limit=6
&startkey_docid=2278388
This request will return rows starting with the correct item in the sequence, providing the best means of paginating the rows from the view.
The Takeaway
A startkey_docid must be used together with a relevant startkey for your pagination to work.
I hope this helps anyone who may be busting their heads on this subtle feature.
Sources
CouchDB Official Docs: http://docs.couchdb.org/en/latest/couchapp/views/collation.html
Interesting thread exchange: http://grokbase.com/t/couchdb/user/091defx51x/sort-by-date-and-find-by-key
Similar SO topic: https://stackoverflow.com/questions/10142850/need-a-couchdb-trick-to-sort-by-date-and-filter-by-group#=

Answer 2

You can find more information on how to use startkey_docid properly here: http://blog.couchbase.com/startkeydocid-behaviour