Node+Quota creation in transaction log can crash leader startup
---------------------------------------------------------------
Key: ZOOKEEPER-1448
URL: https://issues.apache.org/jira/browse/ZOOKEEPER-1448
Project: ZooKeeper
Issue Type: Bug
Components: server
Affects Versions: 3.3.5
Reporter: Botond Hejj
Fix For: 3.3.6
Hi,
I've found a bug in zookeeper related to quota creation which can shutdown
zookeeper leader on startup.
Steps to reproduce:
1. create /quota_bug
2. setquota -n 10000 /quota_bug
3. stop the whole ensemble (the previous operations should be in the
transaction log)
4. start all the servers
5. the elected leader will shutdown with an exception (Missing stat node for
count /zookeeper/quota/quota_bug/zookeeper_
stats)
I've debugged a bit what happening and I found the following problem:
On startup each server loads the last snapshot and replays the last transaction
log. While doing this it fills up the pTrie variable of the DataTree with the
path of the nodes which have quota.
After the leader is elected the leader servers loads the snapshot and last
transaction log but it doesn't clean up the pTrie variable. This means it still
contains the "/quota_bug" path. Now when the "create /quota_bug" is processed
from the transaction log the DataTree already thinks that the quota nodes
("/zookeeper/quota/quota_bug/zookeeper_limits" and
"/zookeeper/quota/quota_bug/zookeeper_stats") are created but those node
creation actually comes later in the transaction log. This leads to the missing
stat node exception.
I think clearing the pTrie should solve this problem.
--
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
