Dan,
On 7/12/23 22:58, Dan McLaughlin wrote:
Well, the deeper I get into the problem, the more complicated it gets. I
thought I was onto something, thinking the size of the JSON might have
something to do with it, so I created a Python script to call curl POSTs
with increasingly larger JSON
13 Jul 2023 13:31:29 Dan McLaughlin :
Mark,
Never mind the last message, I thought it was working, but I had
looked at my test balancer config and forgot I had left the
application config using http instead of h2. So it's still busted. I
am running out of time on my end to spend on this at
Mark,
Never mind the last message, I thought it was working, but I had
looked at my test balancer config and forgot I had left the
application config using http instead of h2. So it's still busted. I
am running out of time on my end to spend on this at the moment.
Since it seems to be a
Hey Mark,
I found a workaround/fix. On the Tomcat Connector, instead of using
protocol=HTTP/1.1, I changed it to
protocol="org.apache.coyote.http11.Http11Nio2Protocol," I haven't had
a single failure since. Not only that, but our application response
times are noticeably faster.
--
Thanks,
Well, the deeper I get into the problem, the more complicated it gets. I
thought I was onto something, thinking the size of the JSON might have
something to do with it, so I created a Python script to call curl POSTs
with increasingly larger JSON thinking I would eventually hit some size
limit,
Mark,
I'm working on a test case. I've built a simple spring boot war with a rest
API "jsonInput" that accepts any JSON and responds with {"message":"OK"}.
What I've determined so far is that it only happens when you are proxying
the request through Apache using mod_proxy_http2, and the size of
12 Jul 2023 13:40:18 Dan McLaughlin :
I can confirm that if I switch h2 to http, everything works as
expected, change it back to h2 or h2c, and it breaks.
That makes me think it is an h2 bug in Tomcat.
Mark, Please let me know if the http2 logs weren't enough to tell you
what's happening;
I can confirm that if I switch h2 to http, everything works as
expected, change it back to h2 or h2c, and it breaks.
Mark, Please let me know if the http2 logs weren't enough to tell you
what's happening; if not, I'll work on creating a simple standalone
reproduction using docker.
--
Thanks,
Hi Mark,
I already provided the output from org.apache.coyote.http2.level =
FINE in the very first post to this thread. I didn't include
everything because all the header information includes things I don't
necessarily want to post publicly and because it would take a while
for me to obfuscate.
On 11/07/2023 19:10, Dan McLaughlin wrote:
One other note, is I can switch to h2c, and it still fails, and a packet
capture shows the entire JSON is delivered to Tomcat, and when I put the
JSON from the packet inspection together (Packets 10199 --> 10208) and
compare it to what the browser says
Dan,
On 7/11/23 13:10, Dan McLaughlin wrote:
One other note, is I can switch to h2c, and it still fails, and a packet
capture shows the entire JSON is delivered to Tomcat, and when I put the
JSON from the packet inspection together (Packets 10199 --> 10208) and
compare it to what the browser
One other note, is I can switch to h2c, and it still fails, and a packet
capture shows the entire JSON is delivered to Tomcat, and when I put the
JSON from the packet inspection together (Packets 10199 --> 10208) and
compare it to what the browser says was sent, they are identical. There are
no
I was already using the latest Amazon Corretto 20, so I tried moving to the
latest OpenJDK 20, which made no difference. So I'm now using the latest
Apache HTTPD 2.4, Tomcat 10.1, and JDK20.
So it's either some configuration in Apache or Tomcat that needs to be
tweaked, an issue with our app
I just upgraded to Tomcat 10.1.11, and it still fails. A comparison of the
logs between a failed POST on 10.1.10 and 10.1.11 look slightly different,
but it still fails for the same reason. I'm going to try updating the JDK
and see if that helps.
--
Thanks,
Dan
On Tue, Jul 11, 2023 at 10:32
@markt
I was looking over the latest release notes for 10.1.11. Any chance either
of these changes could be related...
- [image: Fix:] Refactor blocking reads and writes for the NIO connector
to remove code paths that could allow a notification from the Poller to be
missed resuting in
15 matches
Mail list logo