diff options
| author | Alan Conway <aconway@apache.org> | 2014-12-05 13:39:28 +0000 |
|---|---|---|
| committer | Alan Conway <aconway@apache.org> | 2014-12-05 13:39:28 +0000 |
| commit | 32a91b1044c2fa592cd7e33ea8c54c908b8e04b9 (patch) | |
| tree | 7d9ea55f7562eeda8ce72d86edc46d56deb56503 /qpid/java/broker-codegen | |
| parent | 565b730bf0229bcc2d56b8962e897ac83c585b5d (diff) | |
| download | qpid-python-32a91b1044c2fa592cd7e33ea8c54c908b8e04b9.tar.gz | |
QPID-6252: AMQP 1.0 browsing client generates large number of errors on broker (better fix)
This is a simpler and better fix based on the discussion at:
http://qpid.2158936.n2.nabble.com/Re-svn-commit-r1642720-in-qpid-trunk-qpid-cpp-src-qpid-messaging-amqp-AddressHelper-h-ConnectionConth-td7617083.html
The changes are all client-side:
- A browsing address is unreliable by default. An explicit reliability setting is respected.
- The client session does not record pre-settled deliveries in it's unacked list.
So by default:
- Browsing links are unreliable. Broker sends pre-settled, messages are not recorded in unacked list.
- The user is not required to acknowledge browsed messages for proper clean-up.
- Calling acknowledge() for a browsed message is a no-op, not an error.
If the user explicitly requests a reliable browsing link, then we behave exactly as before.
I can't see any value in doing this with qpidd but maybe with some other broker there might be
a use for being able to control the accept of browsed messages.
This does affect non-browsing, unreliable links but it is an improvement.
Settling a pre-settled messages is a no-op, so there is no point in recording
pre-settled messages in the unacked list since we do nothing when they are
acknoweldged.
git-svn-id: https://svn.apache.org/repos/asf/qpid/trunk@1643276 13f79535-47bb-0310-9956-ffa450edef68
Diffstat (limited to 'qpid/java/broker-codegen')
0 files changed, 0 insertions, 0 deletions
