[
https://issues.apache.org/jira/browse/BOOKKEEPER-140?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13172875#comment-13172875
]
Sijie Guo commented on BOOKKEEPER-140:
--------------------------------------
> the second scenario
As Utkarsh commented in the email,
{quote}
If it's not the first local subscribe, then region A won't be contacted, and
the subscription request will succeed even if region A is down.
{quote}
so if messages are published in region A, subscriber in region B would not
receive.
refine the second scenario as below:
{code}
2. region shuts down when attaches existing subscriptions.
1) In region A, there is a local subscriber a for topic T. in region B, local
subscriber b for topic T. in region B, local subscribe c for topic T.
2) servers are all restarted in all three regions. But region A is
network-partitioned (or shuts down) from region B and region C.
3) subscriber b attach to Topic T in region B. subscriber c attach to Topic T
in region C.
4) hub server owns Topic T in region B, would do remote subscription to region
A and C. remote subscribe to region A succeed, while remote subscribe to region
C would failed. hub server responds SUCCESS to subscriber b. subscriber c did
the same thing in region C.
5) region A (restarted / network is connected again) became connected with
region B and region C. messages published in region A. But subscribe b and c
would never receive any messages.
{code}
> Hub server doesn't subscribe remote region correctly when a region is down.
> ---------------------------------------------------------------------------
>
> Key: BOOKKEEPER-140
> URL: https://issues.apache.org/jira/browse/BOOKKEEPER-140
> Project: Bookkeeper
> Issue Type: Bug
> Components: hedwig-server
> Affects Versions: 4.0.0
> Reporter: Sijie Guo
> Assignee: Sijie Guo
> Fix For: 4.1.0
>
> Attachments: BOOKKEEPER-140.patch, BOOKKEEPER-140.patch
>
>
> Hub server doesn't subscribe remote region correctly in following cases:
> (assume there is 3 regions, A, B, C)
> 1. region shuts down before first subscribe.
> 1) region C is down.
> 2) subscribe-a subscribe a topic in region A. a subscription state is created
> in region A's zookeeper. but remote subscribe to region C would fail since
> region C is down. hub server will respond client that subscribe failed
> without deleting subscription state. The following subscriptions using same
> subscribe id and same topic would failed due to NodeExists.
> 2. region shuts down when attaches existing subscriptions.
> 1) In region A, there is a local subscriber a for topic T. in region B,
> subscriber b for topic T. in region B, subscribe c for topic T.
> 2) servers are all restarted in all three regions. But region C is
> network-partitioned (or shuts down) from region A and region B.
> 3) subscriber b and subscribe c try to subscribe T again. hub servers in
> region B, C will try to remote subscribe region A, but should failed. There
> is no mechanism to retry remote subscribe. so if messages are published to
> topic T in region A, subscribe b and subscribe c would receive any message.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
