7:18 p.m.
Hello,
I'm trying out HTTP/2 with undertow-core in my application, and I see
that I'm able to get the server into a bad state pretty reliably by
refreshing quickly in my browser. It stops responding to requests and
the CPU usage of the Java process starts to climb towards 100%.
I was able to replicate this in the HTTP/2 example in the latest from
the master branch, by changing the main HttpHandler. This is what I
put in Http2Server#main(String[]):
Undertow server = Undertow.builder()
.setServerOption(UndertowOptions.ENABLE_HTTP2, true)
.setServerOption(UndertowOptions.ENABLE_SPDY, true)
.addHttpListener(8080, bindAddress)
.addHttpsListener(8443, bindAddress, sslContext)
.setHandler(new HttpHandler() {
@Override
public void handleRequest(HttpServerExchange exchange) {
if (exchange.isInIoThread()) {
exchange.dispatch(this);
return;
}
exchange.startBlocking();
exchange.getResponseHeaders().put(
new HttpString("Content-Type"), "text/html");
try (BufferedWriter writer = new BufferedWriter(
new OutputStreamWriter(
exchange.getOutputStream(),
StandardCharsets.UTF_8))) {
writer.append("<!DOCTYPE html><html>\n");
for (int i = 0; i < 100; i++) {
writer.append(
"<link rel=\"stylesheet\" href=\"/" + i +
".css\">\n");
}
writer.append("hello");
} catch (IOException e) {
e.printStackTrace();
throw new RuntimeException(e);
}
}
}).build();
Then I go to https://localhost:8443/ in a browser and hold down F5 for
a bit. I start seeing errors like this:
java.nio.channels.ClosedChannelException
at
io.undertow.server.protocol.framed.AbstractFramedStreamSinkChannel.write(AbstractFramedStreamSinkChannel.java:396)
at org.xnio.channels.Channels.writeFinalBasic(Channels.java:961)
at
io.undertow.server.protocol.framed.AbstractFramedStreamSinkChannel.writeFinal(AbstractFramedStreamSinkChannel.java:420)
at
org.xnio.conduits.StreamSinkChannelWrappingConduit.writeFinal(StreamSinkChannelWrappingConduit.java:66)
at
org.xnio.conduits.ConduitStreamSinkChannel.writeFinal(ConduitStreamSinkChannel.java:104)
at
io.undertow.channels.DetachableStreamSinkChannel.writeFinal(DetachableStreamSinkChannel.java:195)
at
io.undertow.server.HttpServerExchange$WriteDispatchChannel.writeFinal(HttpServerExchange.java:1882)
at
io.undertow.io.UndertowOutputStream.writeBufferBlocking(UndertowOutputStream.java:288)
at io.undertow.io.UndertowOutputStream.close(UndertowOutputStream.java:331)
at sun.nio.cs.StreamEncoder.implClose(StreamEncoder.java:320)
at sun.nio.cs.StreamEncoder.close(StreamEncoder.java:149)
at java.io.OutputStreamWriter.close(OutputStreamWriter.java:233)
at java.io.BufferedWriter.close(BufferedWriter.java:266)
at io.undertow.examples.http2.Http2Server$1.handleRequest(Http2Server.java:94)
at io.undertow.server.Connectors.executeRootHandler(Connectors.java:199)
at io.undertow.server.HttpServerExchange$1.run(HttpServerExchange.java:774)
at
java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
at
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
at java.lang.Thread.run(Thread.java:745)
In case it helps, I saw a different error in my real (not undertow
example) application when I started seeing the same effects visibly
(site down and CPU climbing to 100%). Maybe it is the same root cause
but just being caught/logged differently:
ERROR [2015-05-28 14:23:59,804] org.xnio.nio: XNIO000011: Task
io.undertow.server.protocol.framed.AbstractFramedChannel$2@308b5f1f
failed with an exception
java.lang.IllegalStateException: null
at
io.undertow.server.protocol.framed.AbstractFramedStreamSinkChannel.getBuffer(AbstractFramedStreamSinkChannel.java:523)
~[undertow-core-1.3.0.Beta1-SNAPSHOT.jar:1.3.0.Beta1-SNAPSHOT]
at
io.undertow.server.protocol.framed.AbstractFramedChannel.flushSenders(AbstractFramedChannel.java:496)
~[undertow-core-1.3.0.Beta1-SNAPSHOT.jar:1.3.0.Beta1-SNAPSHOT]
at
io.undertow.server.protocol.framed.AbstractFramedChannel$2.run(AbstractFramedChannel.java:581)
~[undertow-core-1.3.0.Beta1-SNAPSHOT.jar:1.3.0.Beta1-SNAPSHOT]
at org.xnio.nio.WorkerThread.safeRun(WorkerThread.java:560)
~[xnio-nio-3.3.1.Final.jar:3.3.1.Final]
at org.xnio.nio.WorkerThread.run(WorkerThread.java:462)
~[xnio-nio-3.3.1.Final.jar:3.3.1.Final]
I see the same behavior in my real application with both undertow-core
1.2.6.Final and 1.3.0.Beta1-SNAPSHOT (the latter installed locally
from source). I only tried the Http2Server example in the latest
version though.
-Michael
12:28 a.m.
Hi Michael,
That stack trace seems to show a close() call to the writer in your handler, but I
don't see that call in the code?
Note that getting I/o exceptions during write and close calls is normal after a client
abnormally terminates. You typically just drop the exception (perhaps trace logging).
Could you get a jstack output under the scenario where you see high CPU load?
1:04 a.m.
Hi Jason,
I believe the close() call is coming from me declaring the writer in a
try-with-resources statement.
Here's a jstack from just now, where my CPU is sitting at 96%:
https://gist.github.com/michaelhixson/8a6f2ebf4b167fb1b67e
-Michael
On Fri, May 29, 2015 at 9:28 PM, Jason T. Greene
<jason.greene(a)redhat.com> wrote:
1:04 a.m.
Hi Jason,
I believe the close() call is coming from me declaring the writer in a
try-with-resources statement.
Here's a jstack from just now, where my CPU is sitting at 96%:
https://gist.github.com/michaelhixson/8a6f2ebf4b167fb1b67e
-Michael
On Fri, May 29, 2015 at 9:28 PM, Jason T. Greene
<jason.greene(a)redhat.com> wrote:
1:26 a.m.
Also here is a jmap -histo from a little later:
https://gist.github.com/michaelhixson/de6ede25fdb79a4f25b4
(apologies in advance if gmail sends this email twice like my last one)
-Michael
On Fri, May 29, 2015 at 10:04 PM, Michael Hixson
<michael.hixson(a)gmail.com> wrote:
7:13 p.m.
Was this Jira issue 461? I'm seeing promising results with this
commit, while if I revert the commit I run into the issue again.
https://github.com/undertow-io/undertow/commit/8f8e1658c5aface7f65a351e98...
If so, thanks!
-Michael
On Fri, May 29, 2015 at 10:26 PM, Michael Hixson
<michael.hixson(a)gmail.com> wrote:
3:36 a.m.
https://github.com/undertow-io/undertow/commit/4cd914008fc7fddca97af53005...
Is the one.
Stuart
----- Original Message -----
3585
days inactive
3591
days old
6 comments
3 participants
participants (3)
-
Jason T. Greene -
Michael Hixson -
Stuart Douglas