git-svn-id: https://svn.erp5.org/repos/neo/trunk@2437 71dcc9de-d417-0410-9af5-da40c76e7ee4

This requires explicitly collecting garbage if we get disconnected from
primary master.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2436 71dcc9de-d417-0410-9af5-da40c76e7ee4

For future extension.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2435 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2434 71dcc9de-d417-0410-9af5-da40c76e7ee4

Also, improve test (try to break code in tests, don't test unneeded
properties).

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2433 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2432 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2431 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2430 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2429 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2428 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2427 71dcc9de-d417-0410-9af5-da40c76e7ee4

A storage might notify the master after the cluster as fall down to the
verification state.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2426 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2425 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2424 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2423 71dcc9de-d417-0410-9af5-da40c76e7ee4

This is used by zodbshootout, for example, to only flush storage's cache.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2422 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2421 71dcc9de-d417-0410-9af5-da40c76e7ee4

This should increase process responsiveness when a single _poll call
fetched many packets: as long as packet queue is not empty, no epoll.poll
call was made. This means that outgoing packets were kept in our buffer
until packet queue becomes empty.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2420 71dcc9de-d417-0410-9af5-da40c76e7ee4

Some requests can be safely ignored when received over a closed connection.
This was previously done explicitly in handlers, but it turns out it would
cause a lot of code duplication. Instead, define the policy on a packet
type basis, and apply it to all packets upon reception, before passing it
to handler.
Also, protect request handlers when they respond, as connection might be
closed.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2419 71dcc9de-d417-0410-9af5-da40c76e7ee4

Those responses are implemented using Error packets.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2418 71dcc9de-d417-0410-9af5-da40c76e7ee4

This change has an important effect on "abort" call in client handler: once
primary master node locked a transaction, a client must not be allowed to
abort it (it already asked to commit).
Also, remove parameter value in abortFor (it is the new default behaviour).
Also, document abort a bit more.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2417 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2416 71dcc9de-d417-0410-9af5-da40c76e7ee4

This should improve replication & pack performance.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2415 71dcc9de-d417-0410-9af5-da40c76e7ee4

- set master_conn to None to clarify disconnection
- purge node pool after closing all connections
- allow restarting polling thread after its shutdown
Also, only start polling thread when needed (side-effect of last point).

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2414 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2413 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2412 71dcc9de-d417-0410-9af5-da40c76e7ee4

Otherwise, we would check the same chunk multiple times, if last range
check was successful: at this point, we wouldn't know criterion to find
the first row after replicated chunk.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2411 71dcc9de-d417-0410-9af5-da40c76e7ee4

This caused incorrect object deletions during replication, when a chunk
lacks rows.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2410 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2409 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2408 71dcc9de-d417-0410-9af5-da40c76e7ee4

deleteTransaction is used by replication to remove a single transaction but
not its content.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2407 71dcc9de-d417-0410-9af5-da40c76e7ee4

Otherwise, it might not be called (it connector is not present at
transaction manager's "begin" call time).

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2406 71dcc9de-d417-0410-9af5-da40c76e7ee4

If current chunk size is lesser than maximum, it means latest check for the
whole chunk found a problem. If current smaller check succeeded, move to
the other half and divide size by two.
Also, factorise replication "check...Range" handlers.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2405 71dcc9de-d417-0410-9af5-da40c76e7ee4

This way, it is clear that test doesn't expect replication to end after
current chunk.

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2404 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2403 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2402 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2401 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2400 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2399 71dcc9de-d417-0410-9af5-da40c76e7ee4

git-svn-id: https://svn.erp5.org/repos/neo/trunk@2398 71dcc9de-d417-0410-9af5-da40c76e7ee4