- 08 Nov, 2010 1 commit
-
-
Grégory Wisniewski authored
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
-
- 05 Nov, 2010 10 commits
-
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2425 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2424 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2423 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
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
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2421 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
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
-
Vincent Pelletier authored
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
-
Vincent Pelletier authored
Those responses are implemented using Error packets. git-svn-id: https://svn.erp5.org/repos/neo/trunk@2418 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
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
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2416 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
- 04 Nov, 2010 4 commits
-
-
Vincent Pelletier authored
This should improve replication & pack performance. git-svn-id: https://svn.erp5.org/repos/neo/trunk@2415 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
- 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
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2413 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2412 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
- 03 Nov, 2010 5 commits
-
-
Vincent Pelletier authored
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
-
Vincent Pelletier authored
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
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2409 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Grégory Wisniewski authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2408 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Grégory Wisniewski authored
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
-
- 02 Nov, 2010 10 commits
-
-
Vincent Pelletier authored
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
-
Vincent Pelletier authored
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
-
Vincent Pelletier authored
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
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2403 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2402 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2401 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2400 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2399 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2398 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2397 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
- 01 Nov, 2010 7 commits
-
-
Vincent Pelletier authored
This prevents lines from next setUp from showing up in test's log. This is not so nice to have to call it here, but it is probably an artificial-enough use case to justify this. git-svn-id: https://svn.erp5.org/repos/neo/trunk@2396 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2395 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2394 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2393 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
There is no reason why we wouldn't see this when runing individual tests, I think. git-svn-id: https://svn.erp5.org/repos/neo/trunk@2392 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
Also, improve exit condition readability. git-svn-id: https://svn.erp5.org/repos/neo/trunk@2391 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
git-svn-id: https://svn.erp5.org/repos/neo/trunk@2390 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
- 30 Oct, 2010 3 commits
-
-
Vincent Pelletier authored
Was enabled accidentally in r2371. git-svn-id: https://svn.erp5.org/repos/neo/trunk@2389 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
But make it optional, as some tests might want to stop everything except clients (stopping single processes is already possible separately). git-svn-id: https://svn.erp5.org/repos/neo/trunk@2388 71dcc9de-d417-0410-9af5-da40c76e7ee4
-
Vincent Pelletier authored
Call it in client, right after starting and stopping poll thread. git-svn-id: https://svn.erp5.org/repos/neo/trunk@2387 71dcc9de-d417-0410-9af5-da40c76e7ee4
-