Within the MapReduce implementation, are reduce functions indexed similarly to map functions? - view

Question

If I have a couple docs in Couch that look like this:
{
"_id": "be890e3ee1457e920f12722c44001b0e", // Or whatever auto ID
"_rev": "7-74d1787aa3ca6d2526c4436577da660f", // Or whatever auto rev
"type_": "count",
"value": -1,
"time": 1485759832925 // This is an Epoch time, the result of this JavaScript: var x = (new Date()).getTime(), that I calculate in the console just before saving the doc
}
And then I create a map function to retrieve these docs like so (that I run directly after creating a few docs):
function(doc) {
if (doc.type_) {
if (doc.time) {
var datetime = (new Date()).getTime();
var docTime = doc.time;
var docAge = datetime - docTime;
// Only emit docs younger than 1 minute
if (docAge / 1000 <= 60) {
emit(doc.time, docAge);
};
};
};
};
I found that once the view is calculated, that the docAge will never change and that the docs will always be emitted despite being 'too old'.
If you open a doc and re-save it, then the view will NOT emit that doc (because it reflects as a CouchDB update and now the time value is too old), but other docs will not have been recalculated (i.e. the docAge for those docs is still the same).
So by this I can see that views are incrementally updated to reflect changed docs. And as I understand, they are cached.
Question:
Where are these cached views stored?
Are Group and reduce output recalculated from scratch everytime the map
function incrementally updates?

Answer 1

Your views are not being "cached" per-se. The idea behind CouchDB views is that they are deterministic, and thus should not be influenced by anything beyond the document in question.
Using new Date() in your view means that you are bringing in an external resource (the clock) which means your view index will be computed in a way you aren't intending based on your question.
Your map function must deal in absolutes, so it should output the timestamp irregardless of the time that your view index is rebuilt. From your application, you'll pass the time you want to query as a parameter to the view query.
For example, consider this view function:
function (doc) {
if (doc.type_ && doc.time) {
emit(doc.time);
}
}
It will output the time for all your documents. Then, you will query the view passing in the expected timeframe.
?start_key=<timestamp from 1 minute ago>
Then you will get the documents whose timestamp falls in the last minute. You can include end_key to specify an upper-limit.
There's a bit of a mental hurdle to overcome with how MapReduce views in CouchDB are designed to work, so I would highly recommend their Guide to Views to get started. (in fact, their newest documentation is quite good and I would highly recommend reading through all of it)