Github user davisp commented on the pull request:
https://github.com/apache/couchdb-couch-replicator/pull/29#issuecomment-192382324
The underlying cause here is that the replicator now uses the http
connection pool for the changes feed on the source database. This was done to
handle cases where there are many long lived replications. The changes feed
reconnecting was causing significant CPU usage due to having to renegotiate TLS.
However, using a worker from the pool means that if the pool only has one
worker, then everything trying to read from the database (get_pending_changes
*or* reading documents) will have to wait until the changes feed to exit (ie,
its dead locked). Alternatively, previously we *did* use more HTTP connections
that requested, it was just that one of them wasn't part of the pool. So this
isn't a change in behavior.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
