Add withReadTimeout(Integer) method to NettyDockerCmdExecFactory #982
Conversation
This enhances our Netty support so that, like JerseyDockerCmdExecFactory, we have a withReadTimeout(milliseconds) method that configures the factory to implement a read timeout. This allows clients to cope with recalcitrant docker daemons - if the daemon stops responding then the client code will get a SocketTimeoutException (much like it would if it was using Jersey instead of Netty) instead of just hanging forever.
| */ | ||
| @Override | ||
| protected synchronized void channelIdle(ChannelHandlerContext ctx, IdleStateEvent evt) throws Exception { | ||
| assert evt.state() == IdleState.READER_IDLE; |
There was a problem hiding this comment.
did you copy it from somewhere?
There was a problem hiding this comment.
Not "copied". It was coded by initially using the ReadTimeoutHandler class as specified by its Javadoc, but then developed from there.
Using it as-documented was a long way from ideal because Netty's ReadTimeoutHandler throws a peculiar exception (with no stacktrace) when it triggers, with no proper error message, which is why its javadoc suggests that you then add a second handler class to turn the thrown exception into the exception you really wanted. Ideally it'd be possible to extend/configure their class with your own functionality to control the exception (or for their class to throw a decent exception to start with!), but they don't provide for that.
So what I ended up doing instead was to throw the exception that I really wanted from the outset, using Netty's more generic IdleStateHandler as the parent class (as Netty's ReadTimeoutHandler does) and going from there.
It's not the most original bit of coding ever because of the constraints (it needs to fit within Netty's APIs), but it's not a copy - there's a finite amount of difference that's possible while still having code that works.
This enhances our Netty support so that, like
JerseyDockerCmdExecFactory, we have a withReadTimeout(milliseconds)
method that configures the factory to implement a read timeout.
This allows clients to cope with recalcitrant docker daemons - if the
daemon stops responding then the client code will get a
SocketTimeoutException (much like it would if it was using Jersey
instead of Netty) instead of just hanging forever.
See issue #588
This change is