You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hbase.apache.org by Guilherme Germoglio <ge...@gmail.com> on 2010/09/08 16:33:52 UTC
missing sync in HTablePool.getTable()
Hello,
I think I came across a small (and mostly harmless) problem in
HTablePool.getTable() code:
1 public HTableInterface getTable(String tableName) {
2 LinkedList<HTableInterface> queue = tables.get(tableName);
3 if(queue == null) {
4 queue = new LinkedList<HTableInterface>();
5 tables.put(tableName, queue);
6 return createHTable(tableName);
7 }
8 HTableInterface table;
9 synchronized(queue) {
10 table = queue.poll();
11 }
12 if(table == null) {
13 return createHTable(tableName);
14 }
15 return table;
16 }
Considering that this method is invoked by multiple threads, it may happen
that while 'queue == null' is true in line 3, it is possible to have a queue
mapped to that name into the tables map when line 5 is executed.
However, I don't think it will cause any harm because the overwritten queue
will eventually be garbage-collected. :-)
Should I write a patch for it?
If so, I considered two approaches:
1. sync the *tables* attribute from lines 3 to 7.
2. or consider the interface ConcurrentMap so we could use putIfAbsent()
I prefer the second. What do you think?
Thanks,
Guilherme
Re: missing sync in HTablePool.getTable()
Posted by Stack <st...@duboce.net>.
On Wed, Sep 8, 2010 at 7:33 AM, Guilherme Germoglio <ge...@gmail.com> wrote:
> However, I don't think it will cause any harm because the overwritten queue
> will eventually be garbage-collected. :-)
>
> Should I write a patch for it?
>
I agree its harmless looking but if you have a moment, please write a patch.
> If so, I considered two approaches:
>
> 1. sync the *tables* attribute from lines 3 to 7.
> 2. or consider the interface ConcurrentMap so we could use putIfAbsent()
>
> I prefer the second. What do you think?
>
Yes. Me too. Change tables to be a concurrent Map type instead of
the wrap to make it synchronized.
Thanks G,
St.Ack