Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Interrupt flag is not always cleared in between requests #7548

Closed
arhimondr opened this issue Feb 8, 2022 · 8 comments · Fixed by #7563 or #7773
Closed

Interrupt flag is not always cleared in between requests #7548

arhimondr opened this issue Feb 8, 2022 · 8 comments · Fixed by #7563 or #7773
Assignees
Labels
Bug For general bugs on Jetty side

Comments

@arhimondr
Copy link
Contributor

arhimondr commented Feb 8, 2022

Jetty version(s)

9.4.44.v20210927

Java version/vendor (use: java -version)

openjdk version "11.0.14" 2022-01-18 LTS
OpenJDK Runtime Environment Zulu11.54+23-CA (build 11.0.14+9-LTS)
OpenJDK 64-Bit Server VM Zulu11.54+23-CA (build 11.0.14+9-LTS, mixed mode)

OS type/version

Linux fedora 5.16.5-200.fc35.x86_64 #1 SMP PREEMPT Tue Feb 1 21:37:11 UTC 2022 x86_64 x86_64 x86_64 GNU/Linux

Description

Sometimes when a thread processing a request is interrupted the "interrupted" flag is carried over to the next request handled by the same thread.

How to reproduce?

  1. Interrupt a thread while it is processing Request 1
  2. Send Request 2. When Request 2 is processed by the same thread check if the current thread is interrupted.

Unfortunately the issue doesn't reproduce all the time.

@arhimondr arhimondr added the Bug For general bugs on Jetty side label Feb 8, 2022
@arhimondr
Copy link
Contributor Author

Example stack trace: https://gist.github.com/arhimondr/81d80f27722d5429bf5e210034660369

In QueuedThreadPool the interrupt flag seem to be properly cleared: https://github.com/eclipse/jetty.project/blob/6c79be91b74ccb5e9fa7fecc7adb1daff9219dc4/jetty-util/src/main/java/org/eclipse/jetty/util/thread/QueuedThreadPool.java#L1055

I wonder if a single job in the QueuedThreadPool can be processing more than a single HTTP request?

@sbordet
Copy link
Contributor

sbordet commented Feb 8, 2022

@arhimondr like you noticed, we reset the interrupted flag for every job we run.

Jetty calls Thread.interrupt() only when it's being stopped, so could this be a race of Jetty being stopped while the application is still processing a request?

I wonder if a single job in the QueuedThreadPool can be processing more than a single HTTP request?

That is not the case.

@arhimondr
Copy link
Contributor Author

so could this be a race of Jetty being stopped while the application is still processing a request?

It shouldn't be the case.

That is not the case.

@sbordet Does anything change for non blocking (async) requests?

It looks super weird. I will try to create a simple reproducer that I can share.

@arhimondr
Copy link
Contributor Author

arhimondr commented Feb 8, 2022

@sbordet

Actually I'm able to reproduce the problem with something as simple as

public class TestingServlet
            extends HttpServlet
    {
        @Override
        protected void doGet(HttpServletRequest req, HttpServletResponse resp)
                throws IOException
        {
            if (Thread.currentThread().isInterrupted()) {
                resp.sendError(555);
                return;
            }
            Thread.currentThread().interrupt();
        }
    }

When sending requests repeatedly (even with a simple single threaded loop) some of them return status 555.

@joakime
Copy link
Contributor

joakime commented Feb 8, 2022

Need more details on how you are using/deploying this servlet.
It matters.

Example:
If you are using a ServletContextHandler or embedded mode.

First time through on a REQUEST dispatch you hit Thread.currentThread().interrupt().
Which causes an error.
Now you have an ERROR dispatch to the same servlet.
You now have the flow through Thread.currentThread.isInterrupted(),
But you call resp.sendError() on a an ERROR dispatch. (that's not supported).
Thread comes back to QTP and is cleaned up properly.

@arhimondr
Copy link
Contributor Author

@joakime Here is a different reproducer without sendError involved:

private static class TestingServlet
            extends HttpServlet
    {
        @Override
        protected void doGet(HttpServletRequest req, HttpServletResponse resp)
                throws IOException
        {
            try (Writer writer = resp.getWriter()) {
                boolean interrupted = Thread.currentThread().isInterrupted();
                writer.write(interrupted ? "interrupted" : "not interrupted");
            }
            Thread.currentThread().interrupt();
        }
    }

The reproducer is based on Airlift that uses Jetty. Here is the full implementation: https://gist.github.com/arhimondr/873aee23135e538c0a4d5f7735db1b98

This class initializes the server: https://github.com/airlift/airlift/blob/master/http-server/src/main/java/io/airlift/http/server/HttpServer.java

@sbordet
Copy link
Contributor

sbordet commented Feb 8, 2022

I think the problem is that we clear the interrupted flag in QueuedThreadPool, but not in ReservedThread.

@gregw
Copy link
Contributor

gregw commented Feb 9, 2022

I'm wondering if we should also clear in HttpChannel.unhandle, as we can iterate over several calls to the application there ( eg error dispatch) and we should probably not prepared interrupt status.... or should we... as it is for the same request???

arhimondr added a commit to arhimondr/trino that referenced this issue Feb 9, 2022
Driver methods such as Driver#updateSplitAssignment can be ran from a
thread other than a thread created by TaskExecutor. Interrupting a
thread that is not managed by the engine is potentially unsafe.

For example the Jetty server may preserve the "interrupted" flag in between
requests, potentially impacting a different task or a query:
jetty/jetty.project#7548

While the issue in Jetty is likely a bug that will be fixed there is no
real reason why to interrupt a method other than Driver#process (which
is only invoked by the TaskExecutor).

Other driver methods (such as Driver#updateSplitAssignment) are
lightweight and shouldn't take any significant time to finish while the
interrupt mechanism is rather intended to terminate a processing
that could potentially take a significant amount of time.
@sbordet sbordet self-assigned this Feb 10, 2022
sbordet added a commit that referenced this issue Feb 10, 2022
Now clearing the interrupt flag.

Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
martint pushed a commit to trinodb/trino that referenced this issue Feb 11, 2022
Driver methods such as Driver#updateSplitAssignment can be ran from a
thread other than a thread created by TaskExecutor. Interrupting a
thread that is not managed by the engine is potentially unsafe.

For example the Jetty server may preserve the "interrupted" flag in between
requests, potentially impacting a different task or a query:
jetty/jetty.project#7548

While the issue in Jetty is likely a bug that will be fixed there is no
real reason why to interrupt a method other than Driver#process (which
is only invoked by the TaskExecutor).

Other driver methods (such as Driver#updateSplitAssignment) are
lightweight and shouldn't take any significant time to finish while the
interrupt mechanism is rather intended to terminate a processing
that could potentially take a significant amount of time.
sbordet added a commit that referenced this issue Mar 23, 2022
…s. (#7563)

Now clearing the interrupt flag.

Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
sbordet added a commit that referenced this issue Mar 23, 2022
…s. (#7563)

Now clearing the interrupt flag.

Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
(cherry picked from commit 7b648f6)
sbordet added a commit that referenced this issue Mar 23, 2022
…s. (#7563) (#7773)

Now clearing the interrupt flag.

Signed-off-by: Simone Bordet <simone.bordet@gmail.com>
(cherry picked from commit 7b648f6)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Bug For general bugs on Jetty side
Projects
None yet
4 participants