This behaviour was probably lost with the master's handler split and was found by fixing unit tests.
During recovery, if a node is lost it's state is updated but not propagated to other (storages) nodes because after the recovery they will all retreive the whole node list.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1475 71dcc9de-d417-0410-9af5-da40c76e7ee4

Send an error to the storage node if the partition table sent is invalid instead of killing the master.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1474 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1473 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1472 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1471 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1470 71dcc9de-d417-0410-9af5-da40c76e7ee4

KeyError may be raised from many statements here, the try/except give no indications.
Also reduce block indentation for readability and add some comments.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1469 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1468 71dcc9de-d417-0410-9af5-da40c76e7ee4

This packet replace notifyPartitionChange when used by storage's replicator at the end of a single partition replication.
In those case, only one partition was involved and the node concerned must match with the sender identify, causing many useless on the master side.
This packet contains only the partition number that is replicated and implies that the node's cell is now up to date.
It should be sent only by the storage node from replicator and received by master node.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1467 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1466 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1465 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1464 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1463 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1462 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1461 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1460 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1459 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1458 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1457 71dcc9de-d417-0410-9af5-da40c76e7ee4

- testUtil module is now empty, but kept for futur tests.
- NeoTestBase got a new getNextTID() method with a copy/paste from master's transactions manager to generate TIDs.
- testProtocol module no more import getNextTID() from neo.util module
- testTransactions test module is updated to test new features provides by transactions manager.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1456 71dcc9de-d417-0410-9af5-da40c76e7ee4

Instead of let the ltid in the master app module, the transaction manager keep it and compute the next each time a transaction begins.
The getNextTID() method has moved from util to transactions module.
Last TID collected during recovery is also handled by the transaction manager through setLastTID() method.
Any reference to app.ltid are updated,

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1455 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1454 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1453 71dcc9de-d417-0410-9af5-da40c76e7ee4

The manager kept track of pending (unfinished) transactions initiated by the client nodes and locked by storage nodes.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1452 71dcc9de-d417-0410-9af5-da40c76e7ee4

- Handle a node list instead of a single node
- Send only one packet per peer, reduce bandwidth and latency
- Remove the related XXX

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1451 71dcc9de-d417-0410-9af5-da40c76e7ee4

MySQLManager.close() is only used in tests where many connections are opened.
In a storage process, the connection is opened only once and kept until it shutdown.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1450 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1449 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1448 71dcc9de-d417-0410-9af5-da40c76e7ee4

Now the broadcastPartitionChanges() method check itself the cell_list content.
- If empty, it does nothing.
- If not empty, it generate itself the next PTID and broadcast the changes.

Remove the related XXX.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1447 71dcc9de-d417-0410-9af5-da40c76e7ee4

- An update means that the item must be already known
- The __call__ replacement for __str__ might be a false good idea
- Module neo.master.test.connector no more exists
- XXX in client/app.py is related to deleted code.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1446 71dcc9de-d417-0410-9af5-da40c76e7ee4

- Rollback out of a transaction is not an issue with MySQL.
- Any unimplemented packet type must fallback to default handler and raise UnExpectedPacketError exception.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1445 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1444 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1443 71dcc9de-d417-0410-9af5-da40c76e7ee4

- Delayed node acceptation
- Verification process review
- Partition table tweaking improvement required
- Node/Connection state review
- Error packet review

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1442 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1441 71dcc9de-d417-0410-9af5-da40c76e7ee4

Objects no more related to currently assigned partitions are flushed from relationnal database since implementation of PartitionTable.dropPartition() in r1228.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1440 71dcc9de-d417-0410-9af5-da40c76e7ee4

See r1356 and followings.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1439 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1438 71dcc9de-d417-0410-9af5-da40c76e7ee4

- Use the client's service handler because of handler split
- Fix use of identifyToMasterNode()
- Set IDs on packets
- Use new-style packet decoding

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1437 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@1436 71dcc9de-d417-0410-9af5-da40c76e7ee4