[
https://issues.apache.org/jira/browse/COUCHDB-3418?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16012950#comment-16012950
]
Joan Touzet commented on COUCHDB-3418:
--------------------------------------
Hi Paul,
This is expected. Database compression means that only the most recent revision
required to successfully replicate the database is present. In the case of a
deleted document, that means that only the "tombstone" (document ID + revision
+ "_deleted": true) is retained. This is done so that if the document is
re-created later (with a revision 1 greater than the previous revision) we can
successfully detect that this newer revision isn't, for instance, the original
revision of that document. This is the same for all documents - older revisions
will be cleaned away during database compression.
Example: Doc is created (rev 1), modified 2 times (revs 2 and 3), and deleted
(rev 4). You replicate and only see the rev 4 tombstone. Later, someone else
also replicates, but re-creates a new version of that document (rev 5). You
replicate with that person later and retrieve the newer version of the document
(the re-created rev 5).
Does this make sense?
> _changes returns a single item that says a doc is deleted
> ---------------------------------------------------------
>
> Key: COUCHDB-3418
> URL: https://issues.apache.org/jira/browse/COUCHDB-3418
> Project: CouchDB
> Issue Type: Bug
> Components: HTTP Interface
> Reporter: Paul Kuruvilla
>
> I discovered this while working on maintaining a replica of the npm registry
> (https://replicate.npmjs.com) for Gratipay (https://gratipay.com).
> When I hit the {{/registry/_changes}} endpoint with a filter for {{doc_id =
> 'adam-test'}}, I get back a single change which says that the doc was
> deleted. This looks like a bug, because I'd expect that every 'deleted'
> change would have a corresponding change for when the doc was created.
> (Please correct me if my understanding is wrong)
> Here is a cURL command to replicate the issue:
> {code}
> curl -XPOST -H "Content-Type: application/json"
> https://replicate.npmjs.com/registry/_changes\?limit\=10\&filter\=_doc_ids -d
> '{"doc_ids": ["adam-test"]}'
> {code}
> The output I receive is:
> {code}
> {
> "results": [
> {"seq":1908160,"id":"adam-test","changes":[{"rev":"2-a543a8082c03a641a098fbe39e58afed"}],"deleted":true}
> ],
> "last_seq":1916520
> }
> {code}
> This was the first such occurrence we hit in the ~ 4 million changes that
> we've processed.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
