Alexander Shlemin

Response example from the WordPress (POST /wp-login.php): ```http HTTP/1.1 200 OK Date: Thu, 08 Sep 2022 11:50:34 GMT Server: Apache/2.4.54 (Debian) X-Powered-By: PHP/7.4.30 Expires: Wed, 11 Jan 1984 05:00:00 GMT...

Related issue: #490

Crash is also reproduced with HTTPS and HTTP/2 tests: [stress.test_stress.H2CurlStress](https://github.com/tempesta-tech/tempesta-test/blob/c70b3d26457a759099a8e9628b1cb8c32a0e512f/stress/test_stress.py#L102) A few more crashes occurs (not reproduced): skb_copy_and_csum_bits ... kernel BUG at net/core/skbuff.c:2169 [ 12.080038] ------------[ cut here ]------------ [...

It turned out that special network conditions are required to reproduce crashes. Testing was conducted under high TCP fragmentations conditions. Fragmentation seems to have a big impact on Tempesta performance....

Crash when running `stress.test_stress.WrkStress` with MTU = 80 kernel BUG at net/core/skbuff.c:111! [ 350.647003] [tempesta fw] Tempesta FW is ready [ 350.774350] [tempesta fw] new client: cli=000000003ea793a3 [ 350.774677] [tempesta...

Related to the previous comment, HTTP/2 requests hangs forever with large output (sometimes) or when MTU 80 is set (reproducible). Can be reproduced with `stress.test_stress.H2CurlStress` and MTU 80. curl -k...

Another way to reproduce crashes when running in local testing mode, is to run Tempesta and backend server on different network interfaces, which has different MTU. Can be reproduced with...

More related tests, interleaving multiple hosts in multiplexed connections: [TlsSNIwithHttpTableMulti](https://github.com/tempesta-tech/tempesta-test/blob/d3e2a2088c36e07aa1d18cedde3f61f81f95bff4/tls/test_tls_cert.py#L792)

List of related issues after tests: https://github.com/tempesta-tech/tempesta-test/issues/212#issuecomment-1239118068

If connection is not closed, then chunked body is expected by a client, but body is cutted off by Tempesta. And even if `Connection: close` is set by a client,...