[COUCHDB-793] replication hangs (recent @dev thread)

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.10.2, 0.11
• Fix Version/s: 0.11.1, 1.0, 1.0.3
• Component/s: Replication

Environment

trunk and 0.11 partially

• Assignee: Unassigned
• Reporter: Filipe David Borba Manana
• Created: 10/Jun/10 18:16
• Updated: 07/Apr/12 22:12
• Resolved: 12/Jun/10 18:01

Description

Following the recent @dev thread about the replication.js test hanging, the following patch fixes 2 causes for this:

1) In couch_rep_reader, if the reader_loop process finishes before all monitored processes (which read docs from source db) finish, the couch_db_writer process will never receive the message

{complete, HighSeq}

. This happens more frequently for replication by doc_ids

2) For trunk only, in couch_rep_writer, if we replicate a document with attachments and the first couch_rep_httpc upload (remote target db case) try doesn't succeed, subsequent tries will always fail because the couch_work_queue used was closed after the first try and the streaming function passed to ibrowse will always returns eof.

The following patch fixes both problems. With it, running the replication test several times in a row (20+) succeeds, as well as all other JS and Etap tests.

Reason 1) might be the cause for COUCHDB-596 as well.

Issue Links

• is related to: COUCHDB-596 Replication tasks sometimes never complete