5:16 p.m.
I extended the test a bit by adding transactions.
A minimal cache.xml:
| <?xml version="1.0" encoding="UTF-8"?>
| <jbosscache xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
| xmlns="urn:jboss:jbosscache-core:config:3.0">
|
| <!-- Configure the TransactionManager -->
| <transaction
transactionManagerLookupClass="org.jboss.cache.transaction.GenericTransactionManagerLookup
| "/>
|
| </jbosscache>
|
| TransactionManager txmanager = new
DummyTransactionManagerLookup().getTransactionManager();
|
Then starting and ending a transaction around the put in the write thread:
| for (int id=1000*iteration + 100*writer; id<1000*iteration +
100*(writer+1); id++)
| {
| String key = "Key" + id;
| txmanager.begin();
| cache.put(FQN, key, VALUE);
| LOG.info("Putting " + key);
| Object result = cache.get(FQN, key);
| if (result == null)
| {
| LOG.error("Missing value for immediate get of " + key
+ ".");
| }
| txmanager.commit();
| result = cache.get(FQN, key);
| if (result == null)
| {
| LOG.error("Missing value for post transaction get of "
+ key + ".");
| }
| }
|
This produced some interesting results:
The immediate get never failed.
The post commit get failed sometimes.
The readthread get failed sometimes.
The readthread failed on items that the post commit thread successfully returned.
For example
anonymous wrote :
| $ grep "post transaction" errors.20091201.log | wc -l
| 228
| $ grep "Missing value for Key" errors.20091201.log | wc -l
| 3660
|
Next, I synchronized a section of the write thread:
| for (int id=1000*iteration + 100*writer; id<1000*iteration +
100*(writer+1); id++)
| {
| String key = "Key" + id;
| synchronized (LOCK_HOLDER)
| {
| txmanager.begin();
| cache.put(FQN, key, VALUE);
| LOG.info("Putting " + key);
| Object result = cache.get(FQN, key);
| if (result == null)
| {
| LOG.error("Missing value for immediate get of " +
key + ".");
| }
| txmanager.commit();
| result = cache.get(FQN, key);
| if (result == null)
| {
| LOG.error("Missing value for post transaction get of
" + key + ".");
| }
| }
| }
|
With this, I have no missed gets.
What happens if I move the commit outside of the synchronization block?
| for (int id=1000*iteration + 100*writer; id<1000*iteration +
100*(writer+1); id++)
| {
| String key = "Key" + id;
| synchronized (LOCK_HOLDER)
| {
| txmanager.begin();
| cache.put(FQN, key, VALUE);
| LOG.info("Putting " + key);
| Object result = cache.get(FQN, key);
| if (result == null)
| {
| LOG.error("Missing value for immediate get of " +
key + ".");
| }
| }
| txmanager.commit();
| Object result = cache.get(FQN, key);
| if (result == null)
| {
| LOG.error("Missing value for post transaction get of "
+ key + ".");
| }
| }
|
Sure enough, the missing values issues return when the commit is not synchronized.
In my production code I have placed the cache writes (puts, remove, etc) inside of
synchronized blocks using a common lock object, and my data integrity issues
have vanished. However, the need for the synchronization in these cases was
definitely unexpected.
Alex
View the original post :
http://www.jboss.org/index.html?module=bb&op=viewtopic&p=4268403#...
Reply to the post :
http://www.jboss.org/index.html?module=bb&op=posting&mode=reply&a...