[
https://issues.apache.org/jira/browse/BOOKKEEPER-294?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13295504#comment-13295504
]
Flavio Junqueira edited comment on BOOKKEEPER-294 at 6/15/12 7:39 AM:
----------------------------------------------------------------------
I'm not convinced that we need a separate lock mechanism for directories. Why
can't we use the zookeeper lock to imply that the directories are locked? In my
view, the zookeeper lock implies that a bookie with a given identity is
available and there is some process responsible for that identity. That
includes the directories it is using.
bq. A better sequence to start the bookie is first start bookie, start
NIOServer and register bookie.
If my rationale is correct, I think we can achieve this, without a race, by
making this simple change in BookieServer:
{code}
public void start() throws IOException {
nioServerFactory = new NIOServerFactory(conf, this);
this.bookie.start();
nioServerFactory.start();
running = true;
deathWatcher = new DeathWatcher(conf);
deathWatcher.start();
// register jmx
registerJMX();
}
{code}
I simply swapped the first two statements. My rationale is: we create the
socket and start/register the bookie. We may receive a request before starting
the NIOServerFactory, but the socket is there already.
was (Author: fpj):
I'm not convinced that we need a separate lock mechanism for directories.
Why can't we use the zookeeper lock to imply that the directories are locked?
In my view, the zookeeper lock implies that a bookie with a given identity is
available and there is some process responsible for that identity. That
includes the directories it is using.
bq. A better sequence to start the bookie is first start bookie, start
NIOServer and register bookie.
If my rationale is correct, I think we can achieve this, without a race, by
making this simple change in BookieServer:
{code}
public void start() throws IOException {
nioServerFactory = new NIOServerFactory(conf, this);
this.bookie.start();
nioServerFactory.start();
running = true;
deathWatcher = new DeathWatcher(conf);
deathWatcher.start();
// register jmx
registerJMX();
}
{code}
I simply swapped the first two statements. My rational is: we create the socket
and start/register the bookie. We may receive a request before starting the
NIOServerFactory, but the socket is there already.
> Not able to start the bookkeeper before the ZK session timeout.
> ---------------------------------------------------------------
>
> Key: BOOKKEEPER-294
> URL: https://issues.apache.org/jira/browse/BOOKKEEPER-294
> Project: Bookkeeper
> Issue Type: Bug
> Components: bookkeeper-server
> Affects Versions: 4.1.0
> Reporter: Gopinathan A
> Assignee: Rakesh R
> Fix For: 4.2.0, 4.1.1
>
> Attachments: BOOKKEEPER-294.1.patch, BOOKKEEPER-294.2.patch,
> BOOKKEEPER-294.3.patch, BOOKKEEPER-294.4.patch, BOOKKEEPER-294.patch,
> BOOKKEEPER-294.patch
>
>
> Not able to start the bookkeeper before the ZK session timeout.
> Here i killed the bookie and started again.
> {noformat}
> 2012-06-12 20:00:25,220 - INFO [main:LedgerCache@65] - openFileLimit is 900,
> pageSize is 8192, pageLimit is 456781
> 2012-06-12 20:00:25,238 - ERROR [main:Bookie@453] - ZK exception registering
> ephemeral Znode for Bookie!
> org.apache.zookeeper.KeeperException$NodeExistsException: KeeperErrorCode =
> NodeExists for /ledgers/available/10.18.40.216:3181
> at org.apache.zookeeper.KeeperException.create(KeeperException.java:119)
> at org.apache.zookeeper.KeeperException.create(KeeperException.java:51)
> at org.apache.zookeeper.ZooKeeper.create(ZooKeeper.java:778)
> at org.apache.bookkeeper.bookie.Bookie.registerBookie(Bookie.java:450)
> at org.apache.bookkeeper.bookie.Bookie.<init>(Bookie.java:348)
> at org.apache.bookkeeper.proto.BookieServer.<init>(BookieServer.java:64)
> at org.apache.bookkeeper.proto.BookieServer.main(BookieServer.java:249)
> {noformat}
--
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
