opendev/gerrit
Code Issues Proposed changes

Merge "Reenable ProjectWatchIT"

Browse Source
This commit is contained in:
Edwin Kempin
2016-07-04 14:36:50 +00:00
committed by Gerrit Code Review
parent e10d200258 f21f18fa70
commit ea204d6eea
2 changed files with 1 additions and 67 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
gerrit-acceptance-tests/src/test/java/com/google/gerrit/acceptance/server/project/BUCK
View File

@@ -1,15 +1,7 @@
include_defs('//gerrit-acceptance-tests/tests.defs')
FLAKY_TEST_CASES=['ProjectWatchIT.java']
acceptance_tests(
group = 'server_project',
srcs = glob(['*IT.java'], excludes=FLAKY_TEST_CASES),
srcs = glob(['*IT.java']),
labels = ['server'],
)
acceptance_tests(
group = 'server_project_flaky',
srcs = FLAKY_TEST_CASES,
labels = ['server', 'flaky'],
)

58
gerrit-acceptance-tests/src/test/java/com/google/gerrit/acceptance/server/project/ProjectWatchIT.java
View File

@@ -32,64 +32,6 @@ import java.util.List;
@NoHttpd
public class ProjectWatchIT extends AbstractDaemonTest {
/**
* Tests message project watches on new patch sets
* <p>
* As of 2015-06-21 this test is marked flaky for triggering race
* conditions between indexing and project watches filters as
* of 2015-06-21.
* <p>
* The test $SOMETIMES fails, stating that 2 emails instead of only
* 1 got sent. The root issue is the inserting of two patch sets
* (one shortly after the other), where the first patch set would
* not match a user's filter while the second one would.
* <p>
* The test basically:
* <ol>
* <li>Sets up a watch on the text 'sekret' in the commit message.</li>
* <li>Pushes a change without sekret in the commit message (no
* email is expected). (We'll refer to this as PS1)</li>
* <li>Push another patch set to the same change with sekret in the
* commit message (1 email is expected). (We'll refer to this as PS2)</li>
* <li>[...]</li>
* </ol>
* <p>The expected flow of actions for step 2+3 is:
* <pre>
* (i) Write PS1 to the index
* (ii) Send out emails for PS1 after checking project watches from
* fresh ChangeData
* (iii) Write PS2 to the index
* (iv) Send out emails for PS2 after checking project watches from
* fresh ChangeData
* </pre>
* <p>
* But as step (ii) and step (iv) happen on separate threads, steps
* (ii) and (iii) might get turned around and become:
* <pre>
* * Write PS1 to the index
* * Write PS2 to the index
* * Send out emails for PS1 after checking project watches from
* fresh ChangeData
* * Send out emails for PS2 after checking project watches from
* fresh ChangeData
* </pre>
* <p>
* Hence, the filters for project watches for the emails for PS1 query
* the index after PS2 has already been written there. Hence, the
* filters for PS1 use the commit message of PS2 when filtering on
* 'message:sekret'.
* <p>
* Since in the ProjectWatchIT test, PS2 contains 'sekret', the filters
* for sending out emails for PS1 see a commit message containing
* 'sekret', and the watches match for both PS1 and PS2, although they
* should only match for PS2.
* <p>
* This explains why the test is only failing sometimes, and also why it
* is more likely to occur when the system is under load.
* <p>
* A demo exposing the race condition is available at
* <a href="https://gerrit-review.googlesource.com/#/c/68719/1">https://gerrit-review.googlesource.com/#/c/68719/1</a>.
*/
@Test
public void newPatchSetsNotifyConfig() throws Exception {
Address addr = new Address("Watcher", "watcher@example.com");