Unix Domain Socket - ensure that interest in READ is removed upon EOF #1056
Conversation
| @@ -312,7 +312,7 @@ object Dependencies { | |||
|
|
|||
| val UnixDomainSocket = Seq( | |||
| libraryDependencies ++= Seq( | |||
| "com.github.jnr" % "jnr-unixsocket" % "0.18" // BSD/ApacheV2/CPL/MIT as per https://github.com/akka/alpakka/issues/620#issuecomment-348727265 | |||
| "com.github.jnr" % "jnr-unixsocket" % "0.19" // BSD/ApacheV2/CPL/MIT as per https://github.com/akka/alpakka/issues/620#issuecomment-348727265 | |||
There was a problem hiding this comment.
What’s the fix they made? I’m wondering if they fixed a native buffer issue, in which case perhaps an alternate PR for this with consideration of using native buffers. There’s a FixMe somewhere in this code with a ref to the bug...
There was a problem hiding this comment.
Socket options weren't set correctly -- i.e. big endian was always used. This would cause buffer sizes to be quite wrong on intel arch: jnr/jnr-unixsocket@aa2c1e4
jnr/jnr-unixsocket#48 is still open though.
| @@ -170,6 +170,7 @@ object UnixDomainSocket extends ExtensionId[UnixDomainSocket] with ExtensionIdPr | |||
| sendReceiveContext.receive = PendingReceiveAck(queue, buffer, pendingResult) | |||
| key.interestOps(key.interestOps() & ~SelectionKey.OP_READ) | |||
| } else { | |||
| key.interestOps(key.interestOps() & ~SelectionKey.OP_READ) | |||
There was a problem hiding this comment.
This can now move outside the conditional block given that both conditions remove the subscription.
There was a problem hiding this comment.
Yup, moving it outside the conditional would be good.
|
Got it. Thanks.
… On 30 Jun 2018, at 07:11, Jason Longshore ***@***.***> wrote:
@longshorej commented on this pull request.
In project/Dependencies.scala:
> @@ -312,7 +312,7 @@ object Dependencies {
val UnixDomainSocket = Seq(
libraryDependencies ++= Seq(
- "com.github.jnr" % "jnr-unixsocket" % "0.18" // BSD/ApacheV2/CPL/MIT as per #620 (comment)
+ "com.github.jnr" % "jnr-unixsocket" % "0.19" // BSD/ApacheV2/CPL/MIT as per #620 (comment)
Socket options weren't set correctly -- i.e. big endian was always used. This would cause buffer sizes to be quite wrong on intel arch: ***@***.***
jnr/jnr-unixsocket#48 is still open though.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub, or mute the thread.
|
longshorej
force-pushed
the
uds-deregister-read
branch
from
235f238 to
5932b0c
Compare
longshorej
force-pushed
the
uds-deregister-read
branch
from
5932b0c to
6977572
Compare
|
Thank you, good to see the nifty details fixed. |
Fixes an issue observed with unix domain socket connector where when the read side was closed, the NIO event loop would go into an infinite loop with
channel.readreturning-1.We should have be deregistering interest in
READwhen completing the queue.Related PR: #878
/cc @huntc