[jira] [Commented] (GEODE-10327) Tests that use GfshRule leave behind orphaned processes and do not save artifacts for debugging failures

["ASF subversion and git services (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/GEODE-10327?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17542170#comment-17542170 ] 

ASF subversion and git services commented on GEODE-10327:
---------------------------------------------------------

Commit 774505e7c74cff8c572be1ec4f4bb2b0f3e1a091 in geode's branch refs/heads/develop from Kirk Lund
[ https://gitbox.apache.org/repos/asf?p=geode.git;h=774505e7c7 ]

GEODE-10327: Overhaul GfshRule to kill processes and save artifacts for failures (#7571)

PROBLEM

Tests that use GfshRule leave behind orphaned processes and do not save artifacts for debugging failures.

SOLUTION

GfshRule needs to cleanup all processes it forks. It also needs to save off all runtime artifacts such as logging, stats, pid files, diskstores to enable debugging of test failures.

DETAILS

Enhance GfshRule and modify all tests using it for proper debugging and to prevent test pollution.

Overhaul of GfshRule:

* kill ALL geode processes during cleanup
* use FolderRule to ensure all logs and files are properly saved off when a test fails
* extract GfshExecutor from JUnit rule code
* GfshExecutor allows a test to use any number of Geode versions with just one GfshRule
* add Gfsh log level support for easier debugging
* add support for new VmConfiguration to allow control over Geode and Java versions
* overhaul API of GfshRule and companion classes for better consistency and design

New FolderRule:

* replaces TemporaryFolder and saves off all content when a test fails
* creates root directory under the gradle worker instead of under temp

Update HTTP session caching module tests:

* use new FolderRule to save all artifacts when a test fails
* use nio Paths for filesystem variables

Update acceptance and upgrade tests that use GfshRule:

* use new improved GfshRule and GfshExecutor
* use new FolderRule instead of TemporaryFolder to save all artifacts when a test fails
* use --disable-default-server in tests with no clients
* fix flakiness of many tests by using random ports instead of default or hardcoded port values
* reformat GfshRule API usage in tests to improve readability and consistency
* add GfshStopper to provide common place to await process stop (stop locator/server is async so restarting with same ports is very prone to hitting BindExceptions)

Update ProcessUtils:

* extract NativeProcessUtils and make it public for direct use
* rename InternalProcessUtils as ProcessUtilsProvider and move to its own class
* rethrow IOExceptions as UncheckedIOExceptions
* fix flakiness in NativeProcessUtilsTest by moving findAvailablePid into test method

Minor changes:

* improve code formatting and readability
* convert from old io File to nio Path APIs as much as possible
* close output streams to fix filesystem issues on Windows

Fixes flaky test tickets:

* DeployJarAcceptanceTest GEODE-9615
* possibly other tests that uses GfshRule

NOTES

The jdk8, jdk17 and windows labels were used to run tests on more environments.

This PR contains mostly test and framework changes. The only product code altered is ServerLauncher and several classes in org.apache.geode.internal.process, all of which is in geode-core.

> Tests that use GfshRule leave behind orphaned processes and do not save artifacts for debugging failures
> --------------------------------------------------------------------------------------------------------
>
>                 Key: GEODE-10327
>                 URL: https://issues.apache.org/jira/browse/GEODE-10327
>             Project: Geode
>          Issue Type: Bug
>          Components: tests
>    Affects Versions: 1.15.0, 1.16.0
>            Reporter: Kirk Lund
>            Assignee: Kirk Lund
>            Priority: Major
>              Labels: Java17, pull-request-available
>
> GfshRule needs to cleanup all processes it forks. It also needs to save off all runtime artifacts such as logging, stats, pid files, diskstores to enable debugging of test failures.



--
This message was sent by Atlassian Jira
(v8.20.7#820007)