On Wed, 22 Jun 2022 06:09:39 GMT, Jaikiran Pai <j...@openjdk.org> wrote:
> Can I please get a review for this change which adds a utility to the JDK > test library to help launch the JWebServer? As noted in the JBS issue, this > utility does the necessary work to make sure when the `launch()` method > returns, the jwebserver is ready to receive requests. This helps remove a lot > of boilerplate code from individual tests. > > As part of this commit, the existing `MaxRequestTimeTest` has been migrated > to use this new utility. This existing test continues to pass with this > change. > > tier1, tier2 and tier3 testing passed without any related issues. Hello Alan, > If a test needs an "in-process" HTTP server then it is easy to use the > com.sun.net.httpserver API directly or use ToolProvider to find jwebserver. > So I think the use-case you are looking at is JDK tests that want a HTTP > server in a separate process, is that right? The usage of this new utility will be for those tests that test the `jwebserver` tool itself. In other words, tests that would want to run the `jwebserver` command from within the test and then run some tests against that launched process. So you are right, this utility is for launching a separate process. On a related note, jwebserver doesn't implement the `ToolProvider` SPI, so (for in-process usages), it cannot be found using the `ToolProvider`. My understanding, based on some previous discussions I found around this, is that it was intentional not to implement the `ToolProvider` interface for `jwebserver`. > It's important that such tests need to be able to clean up. If I read the > proposal correctly then they must call Process:destroy, is that right? That is correct. The test which uses this utility to launch the server will be returned by a `record` which will provide the test with the access to the launched `Process`. The test is then responsible for managing that Process including destroying it. The `MaxRequestTimeTest` which uses this new utility, in this PR, has the code which does this destruction. Additionally, the utility itself takes the responsibility of destroying the `Process`, if the `launch()` itself failed. The `launch()` is considered as failed if after the Process has been started, the utility cannot find the server's host/port from the Process's output. In cases like that, this utility will destroy the `Process` and then throw an `IOException` from `launch()` ------------- PR: https://git.openjdk.org/jdk/pull/9232
