couchdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Simon Eisenmann (JIRA)" <j...@apache.org>
Subject [jira] Created: (COUCHDB-691) CouchDB pull replication from HTTPS does not recover after disconnect (hangs)
Date Fri, 12 Mar 2010 11:43:27 GMT
CouchDB pull replication from HTTPS does not recover after disconnect (hangs)
-----------------------------------------------------------------------------

                 Key: COUCHDB-691
                 URL: https://issues.apache.org/jira/browse/COUCHDB-691
             Project: CouchDB
          Issue Type: Bug
          Components: Replication
    Affects Versions: 0.10.1
         Environment: CouchDB 0.10.1 on Ubuntu 9.10 64bit with Nginx 0.7.62. 
            Reporter: Simon Eisenmann


have several CouchDB instances replicating through untrusted network
space. Thus these instances are behind a Nginx SSL-Proxy. Everything
works fine though when for whatever reason one of the connection breaks
then this pull replication never recovers. Even restarting the
replication job does not have any effect despite not giving an error.
Also in Futon the replication jobs are still reported as running (they
never go away).

I just have set up a local test environment with just two nodes
replicating to each other. One of the nodes is behind Nginx with SSL,
and the other is directly reachable unencrypted. When restarting the
unencrypted instance the pull replication on the other Couch recovers
like a charm and and things are in sync quickly again. Not so when i
restart the instance behind HTTPS. This replication never results in any
action again until the instance doing the pull replication is restarted.

After a couple bit of debugging i found that it seems like the _changes
feed is never again requested from the just restarted instance. As soon
as i restart the instance i get the following entry in the Nginx log:

10.1.1.201 - - [10/Mar/2010:17:40:50 +0100]
"GET /database_1/_changes?style=all_docs&heartbeat=10000&since=3135&feed=continuous
HTTP/1.1" 200 408 "-" "CouchDB/0.10.1"

This means the long running connection has just finished (this was the
former working replication request). Afterwards i would suspect the
Couch to start up such a request again, though this never happens.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message