[
https://issues.apache.org/jira/browse/ZOOKEEPER-2572?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15488593#comment-15488593
]
Edward Ribeiro edited comment on ZOOKEEPER-2572 at 9/13/16 10:09 PM:
---------------------------------------------------------------------
Hey [~hanm], really good catch! :D
/cc [~fpj], [~breed], [~rakesh_r]
While studying this piece of code, I have seen a curious fact:
*{{FileTxnLog.truncate}} never returns false*. It's either throws IOException
or returns {{true}}.
But then it's call hierarchy eventually calls {{ZkDatabase.truncateLog}} as
below (my comments are in upper case below):
{code}
public boolean truncateLog(long zxid) throws IOException {
clear();
// truncate the log
boolean truncated = snapLog.truncateLog(zxid);
// IT WILL NEVER ENTER IN THIS IF
if (!truncated) {
return false;
}
loadDataBase();
return true;
}
{code}
And the method above is called by {{Learner.syncWithLeader}}, again with a
snippet below that will never be called, plus getting out of the
{{syncWithLeader}} middway.
{quote}
boolean truncated=zk.getZKDatabase().truncateLog(qp.getZxid());
// AGAIN, NEVER CAN WILL BE TRUE
if (!truncated) {
// not able to truncate the log
LOG.error("Not able to truncate the log "
+ Long.toHexString(qp.getZxid()));
System.exit(13);
}
{quote}
Does it make sense what I am saying?
was (Author: eribeiro):
Hey [~hanm], really good catch! :D
/cc [~fpj], [~breed], [~rakesh_r]
While studying this piece of code, I have seen a curious fact: *this method
never returns false*. It's either IOException or {{true}}.
But then it's call hierarchy eventually calls {{ZkDatabase.truncateLog}} as
below (my comments are in upper case below):
{code}
public boolean truncateLog(long zxid) throws IOException {
clear();
// truncate the log
boolean truncated = snapLog.truncateLog(zxid);
// IT WILL NEVER ENTER IN THIS IF
if (!truncated) {
return false;
}
loadDataBase();
return true;
}
{code}
And the method above is called by {{Learner.syncWithLeader}}, again with a
snippet below that will never be called, plus getting out of the
{{syncWithLeader}} middway.
{quote}
boolean truncated=zk.getZKDatabase().truncateLog(qp.getZxid());
// AGAIN, NEVER CAN WILL BE TRUE
if (!truncated) {
// not able to truncate the log
LOG.error("Not able to truncate the log "
+ Long.toHexString(qp.getZxid()));
System.exit(13);
}
{quote}
Does it make sense what I am saying?
> Potential resource leak in FileTxnLog.truncate
> ----------------------------------------------
>
> Key: ZOOKEEPER-2572
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-2572
> Project: ZooKeeper
> Issue Type: Bug
> Components: server
> Affects Versions: 3.4.9, 3.5.2
> Reporter: Michael Han
> Fix For: 3.4.10, 3.5.3
>
>
> In FileTxnLog.truncate, we have:
> {code}
> public boolean truncate(long zxid) throws IOException {
> FileTxnIterator itr = null;
> try {
> itr = new FileTxnIterator(this.logDir, zxid);
> PositionInputStream input = itr.inputStream;
> if(input == null) {
> throw new IOException("No log files found to truncate! This
> could " +
> "happen if you still have snapshots from an old setup
> or " +
> "log files were deleted accidentally or dataLogDir
> was changed in zoo.cfg.");
> }
> long pos = input.getPosition();
> // now, truncate at the current position
> RandomAccessFile raf=new RandomAccessFile(itr.logFile,"rw");
> raf.setLength(pos);
> raf.close();
> while(itr.goToNextLog()) {
> if (!itr.logFile.delete()) {
> LOG.warn("Unable to truncate {}", itr.logFile);
> }
> }
> } finally {
> close(itr);
> }
> return true;
> }
> {code}
> {{raf}} here can be potentially in a state of not closed after leaving the
> method, if there is an (IO) exception thrown from setLength.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
