Could be, but I don't think I'm seeing that in the test failure cases
I've seen lately. In some cases the VM itself was crashing! Notice
ZOOKEEPER-2 was seeing NPEs in the QuorumPeer.
I did notice in AsyncTest that it's closing the server(s) before closing
the clients, which is causing alot of noise in the logs (planning to
What OS/cpu type configuration are you on? I'm on Ubuntu on 1core cpu.
Hiram Chirino wrote:
Lots of times when a test runs it seems like the ports are already
bound. Anybody else see this? Is this the cause of the intermittent
failures you see?
On Wed, Jul 30, 2008 at 12:32 PM, Patrick Hunt <[EMAIL PROTECTED]> wrote:
I'm soliciting ideas on how to improve the reliability and usability
(creating new tests in particular) of our current test harness.
Hudson and some users are seeing intermittent failures, primarily due to
timing related issue in test setup; the test starts a server, runs some
tests, then shuts down the server, loop to the next test. There is some code
in ClientBase that's supposed to provide a latch for the server startup, but
we also have a number of "sleeps" in the test setup, without which the tests
fail more frequently (so something is still busted). In particular I want to
make it easier to write server tests and to remove the need for sleeps as
this causes the unit tests to run slowly.
Additionally we are seeing a need to tests clients in addition to the server
(much of our current testing is related to verifying the correct function of
the zk server). In this case we are not currently able to test any client
failure handling cases (such as disconnect handling) as we are running
against a fully functional zk server.
I'm thinking we should do two things:
1) create a better test harness for the server
2) implement a mock ZooKeeper.java that has similar semantics as zk server
proper but has the capability to inject/reproduce/verify various error
scenarios for client testing.
If you have any ideas/suggestions/comments/etc.. or would like to work
on/with please let me know.