Strange delay in objectgrid server   

We took a look at the log that was provided. Not knowing your configuration we had to make some assumptions. With additional information some things may be more clear. Here is what to look at:

Do you have replication enabled? If you have replication enabled, it is one of the operations that happens during the time between when the map being updated on the primary and afterCompletion being called. From looking at the trace it appears that replication is not enabled, but I could be mistaken.

The other major operation called during this time is TransactionCallback.commit(). I assume that you are using a TransactionCallback since you are using the write behind logic. From looking at the trace it appears you are calling find or get on one entry and updating it. According to the trace the get is a cache hit so the Loader should never be invoked. The TransactionCallback.begin will be called when the transaction is started, but it should not create a database connection at that time. Only if a Loader is invoked should it lazily create a database connection. When the TransactionCallback.commit is called it should look to see if a connection exists in the slot on the TxID that is reserved for connections and if it doesn't it should no-op otherwise it should close the connection and commit the database transaction. I am assuming the problem is that TransactionCallback.begin() is always creating a database connection and TransactionCallback.commit() is closing it. We had a similar issue with another customer where this type of issue was seen where cache hits were taking as long as cache misses and this was the issue.

If this is not the issue please let us know.

Jared Anderson