Story #2052
Metacat Create locks SystemMetadata Map as well as Synchronization on the same PID
100%
Description
Synchronization will lock the SystemMetadata PID before create in Metacat. Metacat in create will lock as well.
Does put on the Hazelcast SystemMetadata Map block on a locked pid?
Are we certain all updates to the SystemMetadata Map will not cause a deadlock?
History
#1 Updated by Dave Vieglais about 13 years ago
- Position set to 1
- Target version changed from Sprint-2011.48-Block.6 to Sprint-2011.49-Block.6
#2 Updated by Dave Vieglais about 13 years ago
- Target version changed from Sprint-2011.49-Block.6 to Sprint-2011.51-Block.6
- Position deleted (
2) - Position set to 1
#3 Updated by Chris Jones almost 13 years ago
- Assignee changed from Chris Jones to Ben Leinfelder
Assigning to Ben for expediency.
#4 Updated by Matthew Jones almost 13 years ago
- Target version changed from Sprint-2011.51-Block.6 to Sprint-2012.01-Block.1.1
- Position deleted (
14) - Position set to 320
#5 Updated by Ben Leinfelder almost 13 years ago
A few things:
1) I don't see a call to "lock" a PID on the Hazelcast SystemMetadata map in d1_synchronization.
2) I'm not sure why d1_synchronization needs to interact directly with the Hazelcast SystemMetadata map. There are methods to get(), create() and registerSystemMetadata() on the CN and it seems reasonable for this d1_synchronization process to use those API methods instead of going through the "backdoor" of the HZ map.
(Perhaps there is a reason for calling the shared map directly and I am missing the point?)
If all interactions with the SystemMetadata map were limited to the CN implementation, I think we would not be at risk for deadlock behavior.
#6 Updated by Ben Leinfelder almost 13 years ago
- Status changed from New to Closed
chris and robert say:
The synchronization process gets a system metadata lock from ILock in the processing cluster rather than the "storeage" cluster (that is shared with Metacat)