isomarcte / http4sfilehandle

Demonstrates the file handle resource bug with http4s 0.14.2a.

Each folder client and server is an sbt project.

Running the Server

Ther server will write files sent to it into an out folder which it creates relative to the path where you start sbt, so usually server/out.

Beware the server will attempt to delete any out folder that is already there assuming it is from another run.

Running the Client

The client will create about 100,000 files in a folder called in relative to the working directory you start sbt from, so normally client. These files will contain the System.currentTimeMillis() just to give them some bytes.

Beware if there is already an in folder the client will assume it is from another run and attempt to delete it.

After this is done it attempts to connect to the server on localhost:8080 over http.

What will happen

The default setup sends the files from the client using the PooledHttp1Client, although the behavior is the same with the SimpleHttp1Client. It uses a Process.constant(100000).toSource.through(nio.file.chunkR(fileName.toPath())) to represent the EntityBody.

This will fail, for me at least, 100% of the time with the system running out of file handles.

You will seen another Process.constant(100000).toSource.through(nio.file.chunkR(fc)) commented out right below the Path based version. This version uses the lazy val fc instead, which is a java.nio.file.channels.AsynchronousFileChannel. You will also need to uncommment the fc.close() line in the response handler. This works for me 100% of the time with no errors, with either client.