Fix NH-3023 test #1501
Fix NH-3023 test #1501
Conversation
|
Some French sneaked into the code and commit messages |
* Cease freezing during 20 minutes in case of winner thread early failure. * Support connection string having a password in it. * Do not silently succeed on failing to setup the winner thread.
fredericDelaporte
force-pushed
the
NH-3023
branch
from
964af09 to
eac3d48
Compare
| @@ -27,7 +29,7 @@ public void ForceDeadlockOnConnection(SqlConnection connection) | |||
| { | |||
| using (var scope = new TransactionScope(TransactionScopeOption.RequiresNew)) | |||
| { | |||
| using (var cxn = new SqlConnection(connection.ConnectionString + ";Pooling=No")) | |||
| using (var cxn = new SqlConnection(connectionString + ";Pooling=No")) | |||
There was a problem hiding this comment.
connection.ConnectionString is purged from password, when there is one. This causes the winner thread to fail in such case.
| @@ -38,6 +40,7 @@ public void ForceDeadlockOnConnection(SqlConnection connection) | |||
| catch (Exception ex) | |||
| { | |||
| winnerEx = ex; | |||
| winnerLock.Release(); | |||
There was a problem hiding this comment.
In case of early failure of the winner thread, the "loser thread" was stuck for 2 minutes. Better release him.
| @@ -57,6 +60,8 @@ public void ForceDeadlockOnConnection(SqlConnection connection) | |||
| if (winnerEx != null) | |||
| _log.Warn("Winner thread failed", winnerEx); | |||
| } | |||
| // If getting here, expected victim has not fail. If expected winner has failed instead, fail the test. | |||
| Assert.That(winnerEx, Is.Null); | |||
There was a problem hiding this comment.
In case of winner thread failure while loser thread does not fail, the test was "succeeding". Prevent this.
|
There are some issues with this PR. NUnit was consuming much more memory than usual. |
|
On some other PR I have seen SQL Server eating 750Mb, which was causing issues but was not related to these PR. It looks to me the build agent has memory troubles these days. The changes from this PR looks to me very unlikely to have any impact on memory consumption. Especially if the NH-3023 tests are still succeeding, since it then means it has not run through the most significant changes (releasing semaphore in case of "winner" failure and asserting its failure). |
I have got annoyed by this test when experimenting with AppVeyor build on my repository.