Shawn, The bugs kept me up, so I was pondering why I didn't notice them myself.
While writing the blog the RequestDispatcher.push method got deprecated and I switched to using the PushBuilder. As it turns out this caused of both issues. When using RequestDispatcher.push you need to added the context root yourself and because you're creating the complete request path I didn't add the query parameters and everything worked just fine :) Apparently the PushBuilder takes care of adding the context root which is nice but I certainly didn't expect this. I don't really understand why the PushBuilder passes on the query parameters by default. I expect most resources that will be pushed to be static in nature and the query parameters are not needed for these resources. Nice that it is possible but would have expected this to be an opt-in type of feature rather than a default. Great that you get the same results, always good to have a repeatable case. I'll try to confirm your findings with the latest snapshot. On 19 January 2015 at 07:06, Shawn Bissell <[email protected]> wrote: > Tom, I pulled your changes and I reverted back to the Dec 22 snapshot (git > e8c88cfd9cf3cab89788cd530838314089ce9b23) for Jetty you are using in your > Docker image, and I got the same results as you. Those timeout errors went > away, and yes pushing the full page (402 requests) causes > java.lang.IllegalStateExceptions you saw. So I believe the latest snapshot > probably fixed that error, but introduced an incompatibility with Chrome > Canary 42. > > > > > On 2015-Jan-18, at 1:32 PM, Tom Eyckmans <[email protected]> wrote: > > Hi Shaw, > > Thanks for taking the time to look at this and the great feedback, to bad > for me it is not really working the way I thought is was, but thats the > only way you really learn right :) > > I didn't know about the chrome://net-internals/#events thanks for > pointing me to it. Looks like a great resource. > > I changed the push code (also added a default(true) to the push parameter) > and now I also see the SPDY_STREAM_ADOPTED_PUSH_STREAM events. Thanks for > pointing me to the problems. > > Here are some additional test findings (mentioned log files can be found > here https://github.com/teyckmans/http2-push/tree/master/logs): > > I didn't see the following thread deaths in the Jetty output previously: > > 2015-01-18 19:50:50.505:WARN:oejut.QueuedThreadPool:qtp396180261-188: > Unexpected thread death: > org.eclipse.jetty.util.thread.QueuedThreadPool$3@b6df857 in > qtp396180261{STARTED,10<=200<=200,i=129,q=0} > 2015-01-18 19:51:50.363:WARN:oejut.QueuedThreadPool:qtp396180261-219: > Unexpected thread death: > org.eclipse.jetty.util.thread.QueuedThreadPool$3@b6df857 in > qtp396180261{STARTED,10<=199<=200,i=195,q=0} > > I also had a case when the page kept on loading and there was no active > SPDY session listed on the net-internals page in Chrome. See > chrome_spdy_session_hangs.log in github project, spdy session just stopped > fetching, without timeout I kept waiting for a while but it didn't time > out. This was in combination with the thread deaths on the server side. > Would have expected Chrome to timeout at some point. > > After some more testing without rows and column restrictions (pushing 400 > resources) I got the following IllegalStateExceptions in > HttpTransportOverHTTP2.send(HttpTransportOverHTTP2.java:100), in this case > server is taking up 100% cpu as it is logging like crazy. > Looks like the HTTP2 transport code got stuck: > > 2015-01-18 20:08:07.833:WARN:oejs.HttpChannel:qtp396180261-107: Commit > failed > java.lang.IllegalStateException: committed > at > org.eclipse.jetty.http2.server.HttpTransportOverHTTP2.send(HttpTransportOverHTTP2.java:100) > at > org.eclipse.jetty.server.HttpChannel.sendResponse(HttpChannel.java:591) > at > org.eclipse.jetty.server.HttpChannel$CommitCallback.failed(HttpChannel.java:712) > at > org.eclipse.jetty.http2.server.HttpTransportOverHTTP2.send(HttpTransportOverHTTP2.java:100) > > The last 3 lines of the stack are repeated 131 times! > > I've pushed out a new version of the teyckmans/blog-http2-push docker > image and installed it here: > > https://146.148.90.85:8443/blog-http2-push/push?push=true&rows=1&columns=1 > > > Sometimes the page loads fast (1.15 - 1.20 seconds) but sometimes the page > takes (+/-4.5 seconds) when using > https://146.148.90.85:8443/blog-http2-push/push?rows=5 > Haven't found the cause of this yet. > > I haven't tested with a fresh snapshot build from the latest sources, I'll > try and get to that somewhere this week. > > On 18 January 2015 at 19:24, Shawn Bissell <[email protected]> > wrote: > >> I posted this on Tom Eyckmans’ blog ( >> http://blog.iadvise.eu/2015/01/12/http2-server-push/), but I figure this >> is a better place for the discussion since there seems to be a problem with >> the push mechanism itself… >> >> First of all Tom, great work on making this example. I tried creating a >> similar jetty push example and failed miserably :) I hate to break it to >> you, but the http2-push site is pushing a different url from the requested >> one so the pushes are wasted. Hitting the url >> https://localhost:8443/blog-http2-push/push?push=true&rows=0&columns=1 >> (I had to look at the source to determine the ?push=true was required for >> push) >> >> If you look in the Chrome (Canary build 42) >> chrome://net-internals/#events screen and find your SPDY_SESSION you can >> see that the push promise has a url of >> >> >> /blog-http2-push/blog-http2-push/images/slice_0_0.jpg?push=true&rows=0&columns=1 >> >> where as the url requested in the page is just >> >> /blog-http2-push/images/slice_0_0.jpg >> >> So there are 2 problems there … the pushed url path has an extra >> blog-http2-push in it and the pushed url has the querystring in it. >> >> I tried fixing the servlet code but not calling the absoluteResourcePath >> method and by setting the query tring to null. >> pushBuilder.setQueryString(null); >> And then I could see the SPDY_STREAM_ADOPTED_PUSH_STREAM events happening >> in Chrome, but there was some sort of timeout and the client closes the >> streams and the pushed resources were not loaded at all. >> >> Here is what I see in the debug log >> >> 2015-01-18 10:11:58.898:DBUG:oejhs.HttpChannelOverHTTP2: >> qtp565760380-27: HTTP2 PUSH Request #240/798f5a73: >> GET https://localhost:8443/blog-http2-push/images/slice_5_19.jpg HTTP/2 >> accept: >> text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8 >> accept-encoding: gzip, deflate, sdch >> accept-language: en-US,en;q=0.8 >> cache-control: public, max-age=777 >> pragma: no-cache >> user-agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_10_1) >> AppleWebKit/537.36 (KHTML, like Gecko) Chrome/42.0.2278.0 Safari/537.36 >> referer: https://localhost:8443/blog-http2-push/push >> …. >> 2015-01-18 10:11:58.899:DBUG:oejhs.HttpChannelOverHTTP2: >> qtp565760380-27: HTTP2 Commit Response #1/798f5a73: >> HTTP/2 200 null >> Server: Jetty(9.3.0-SNAPSHOT) >> Content-Type: text/html;charset=iso-8859-1 >> 2015-01-18 10:11:58.899:DBUG:oejhs.HttpTransportOverHTTP2: >> qtp565760380-27: HTTP2 Response #1: >> HTTP/2 200 >> Server: Jetty(9.3.0-SNAPSHOT) >> Content-Type: text/html;charset=iso-8859-1 >> …. >> 2015-01-18 10:11:58.900:DBUG:oejhs.HttpTransportOverHTTP2: >> qtp565760380-27: HTTP2 Response #1 committed >> … >> *15 seconds later* >> ... >> 2015-01-18 10:12:13.801:DBUG:oeji.IdleTimeout: >> Scheduler-1530388690: >> HTTP2Stream@48dd8f83{id=2,sendWindow=10485760,recvWindow=65535,reset=false,REMOTELY_CLOSED} >> idle timeout check, elapsed: 15004 ms, remaining: -4 ms >> 2015-01-18 10:12:13.801:DBUG:oeji.IdleTimeout: >> Scheduler-1530388690: >> HTTP2Stream@48dd8f83{id=2,sendWindow=10485760,recvWindow=65535,reset=false,REMOTELY_CLOSED} >> idle timeout expired >> 2015-01-18 10:12:13.801:DBUG:oejh.HTTP2Stream: >> Scheduler-1530388690: Idle timeout 15000ms expired on HTTP2Stream@48dd8f83 >> {id=2,sendWindow=10485760,recvWindow=65535,reset=false,REMOTELY_CLOSED} >> >> _______________________________________________ >> jetty-users mailing list >> [email protected] >> To change your delivery options, retrieve your password, or unsubscribe >> from this list, visit >> https://dev.eclipse.org/mailman/listinfo/jetty-users >> > > _______________________________________________ > jetty-users mailing list > [email protected] > To change your delivery options, retrieve your password, or unsubscribe > from this list, visit > https://dev.eclipse.org/mailman/listinfo/jetty-users > > > > _______________________________________________ > jetty-users mailing list > [email protected] > To change your delivery options, retrieve your password, or unsubscribe > from this list, visit > https://dev.eclipse.org/mailman/listinfo/jetty-users >
_______________________________________________ jetty-users mailing list [email protected] To change your delivery options, retrieve your password, or unsubscribe from this list, visit https://dev.eclipse.org/mailman/listinfo/jetty-users
