Am 13.10.2018 um 11:46 schrieb Rainer Jung:
Am 11.10.2018 um 20:55 schrieb Ruediger Pluem:
The test fails for me as well for 2.4.36 on SLES12. Small bodies are OK, large not. The limit is somewhere between 1.3 and 1.5 MB, not always the same. The test hangs there until mod_reqtimeout times out after a minute, complaining that it could not read more data from the client. If I reduce the multiplicator 1000000 to eg. 200000 it always passes.
On 10/11/2018 08:10 PM, Christophe JAILLET wrote:
No issue on my Ubuntu 18.04 VM.
On what configuration are you running your tests, Rüdiger? macOS, just like Jim?
Centos 7.5 64 Bit
If I start the test server using "t/TEST -start-httpd" and then use curl to POST data, I can even POST much bigger data and get the correct result back. I use
curl -v --data-binary @BIGFILE http://localhost:8529/apache/buffer_in/ > response-body
So I assume it is a problem of interaction between the server reading the POST body and the client sending it.
My test framework was freshly assembled recently, so lots of current modules.
The setup is based on OpenSSL 1.1.1 in the server and in the test framework, but the actual test runs over http, so I don't expect any OpenSSL related reason for the failure.
I did some more tests including using LWP directly and sniffing the packets on the network plus with mod_dumpio and also doing truss / strace.
I can reproduce even when sending using LWP directly or just the POST binary coming with LWP. I can not reproduce with curl.
With mod_dumpio and in a network sniff plus truss it looks like the client simply stops sending once it got the first response bytes. LWP seems to select the socket FD for read and write. As long as only write gets signalled, it happily sends data. Once it gets write plus read signalled, it switches over to read and no longer checks for write. Since our server side implementation is streaming and starts to send the reflected bytes right away, this LWP behavior breaks the request.
"A server listens on a connection for a request,
parses each message received, interprets the message semantics in
relation to the identified request target, and responds to that
request with one or more response messages”
I would interpret that “message received" as the server is expected to wait until the entire request is received, aside from the case of "Expect: 100-continue” and even that alludes to waiting.
"The server intends to send a final response after the request has been fully received and acted upon."
What do you think?