[ https://issues.apache.org/jira/browse/COUCHDB-2535?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14266682#comment-14266682 ]
Robert Newson commented on COUCHDB-2535: ---------------------------------------- Obviously we should do better than crash. The question for me is what we do for partial replications. This has bothered me for a while as it occurs in other circumstances (design docs silently not copied if not admin, things that fail the filter or vdu, etc). I suppose as long as we bump the docs_failed count (or whatever it's called) then it's no worse that we have done in the past, but who sees that count? Can you see it for feed=continuous at all? (it's in the http response to _replicate calls, in case no one has any idea what I'm referring to). > Crash when replicating doc that exceeds max_document_size > --------------------------------------------------------- > > Key: COUCHDB-2535 > URL: https://issues.apache.org/jira/browse/COUCHDB-2535 > Project: CouchDB > Issue Type: Bug > Security Level: public(Regular issues) > Components: Replication > Affects Versions: 1.5.1 > Reporter: Rami Alia > > DB SOURCE has a max_document_size of 400MB, DB TARGET has a max_document_size > of 40MB. Attempt to replicate a doc greater than 40MB from SOURCE to TARGET. > Observed result: > SOURCE replicator crashes followed by SOURCE couchdb crashing > Expected result: > SOURCE/TARGET handle this as gracefully as a validation fail and not crash > replication or couchdb -- This message was sent by Atlassian JIRA (v6.3.4#6332)