Hi Binti,
Probably the issue is related to the fact that you’re working with the cache
inside of CacheStore.writeAll method. Have you tried to use
PESSIMISTIC/REPEATABLE_READ mode? Is the situation the same? Please attach
thread dumps from all the nodes after the hang is reproduced one more time.
In the meanwhile, Alex G. or Sam as I see in the code of
IgniteTxManager.commitTx we remove timeout object after a transaction is moved
into a COMMITTING state
if (tx.timeout() > 0) {
cctx.time().removeTimeoutObject(tx);
if (log.isDebugEnabled())
log.debug("Unregistered transaction with timeout processor: " + tx);
}
What is the reason of that behavior? Am I missing something?
—
Denis
> On Jun 17, 2016, at 6:24 PM, bintisepaha <[email protected]> wrote:
>
> Denis, the txn timeout works in the code as long as it happens before commit
> is called.
> I am thinking if its in committing state, it does not honor timeout.
>
> When the framework calls the checkValid() method in IgniteTxLocalAdapter, it
> makes a call to the below
>
> if (remainingTime() == -1 && setRollbackOnly())
> throw new IgniteTxTimeoutCheckedException("Cache transaction
> timed out " +
> "(was rolled back automatically): " + this);
>
> setRollbackOnly() calls state(MARKED_ROLLBACK)
> and here I see that it will only return true if it was in one of the below
> states
>
>
> case MARKED_ROLLBACK: {
> valid = prev == ACTIVE || prev == PREPARING || prev ==
> PREPARED;
>
> break;
> }
>
> So if it was in COMMITTING, we would not throw a Txn Timeout exception. Is
> this correct?
>
> here is the sample code from our project, we maintain an
> orderUpdateLossPrevention cache in case of node failures when primary node
> fails before it write behinds the entry to the DB Store. if the write behind
> finishes, we remove the entry.
>
> First TXN that updates a transaction cache
> (orderUpdateLossPrevention.put(order.getOrderKey(), order)) along with other
> caches (all transactional)
>
> Transaction tx = null;
> try {
> if (txns == null)
> txns = ignite.transactions();
>
> tx=ignite.transactions().txStart(TransactionConcurrency.OPTIMISTIC,
> TransactionIsolation.SERIALIZABLE);
> tx.timeout(10*1000);
> for (OrderHolder oh : orderHolderList) {
> Order order = oh.getOrder();
> if(order!=null){
> OrderHelper.saveOrder(order);
>
> orderUpdateLossPrevention.put(order.getOrderKey(), order);
>
> }
> }
> // more code
> tx.commit(); // the thread hangs here forever, because some other thread on
> some other node is doing the below...stuck on write behind.
>
> Second Transaction where the same entry is removed when the writebehind is
> finished. its a part of the method call from CacheStore writeAll().
>
> IgniteTransactions txns = null;
> Transaction tx = null;
> Ignite ignite = Ignition.ignite(Global.DATA_GRID_NAME);
> try {
> if (txns == null)
> txns = ignite.transactions();
> tx=ignite.transactions().txStart(TransactionConcurrency.OPTIMISTIC,
> TransactionIsolation.SERIALIZABLE);
> tx.timeout(2*1000);
> orderUpdateLossPreventionCache.remove(entry.getKey());
> tx.commit();
> } catch (IgniteException e) {
> logger.error("Received IgniteException - rolling back transaction", e);
> tx.rollback();
> throw e;
> }
>
>
>
>
>
> --
> View this message in context:
> http://apache-ignite-users.70518.x6.nabble.com/transaction-not-timing-out-tp5540p5725.html
> Sent from the Apache Ignite Users mailing list archive at Nabble.com.
