Emulate multiple failures

[krizhanovsky]

This task is TBD, but here is what I observed from the traffic from the bunch of recent issues, where we faced many new crashes.

The traffic is quite modest in volume (just about 170RPS and 4Mbps). The only thing I could say about reproducing our problems in tests:

1. POST with relatively large binary body, I saw up to 97KB bodies. All are encoded with content-length.
2. high variability in URLs, e.g. https://github.com/tempesta-tech/tempesta/issues/2184 was caused by this reason. This should be also reproducable by #38 (random URL is a common attack pattern)
3. maybe long URLs, but frankly something about 300 bytes doesn't look interesting

[RomanBelozerov]

POST with relatively large binary body, I saw up to 97KB bodies. All are encoded with content-length.

We have tests sending one request with 500 MB body (t_long_body) and stress tests with a lot of request and 65KB body. I think we have enough tests for this case.

high variability in URLs, e.g. https://github.com/tempesta-tech/tempesta/issues/2184 was caused by this reason. This should be also reproducable by https://github.com/tempesta-tech/tempesta-test/issues/38 (random URL is a common attack pattern)

maybe long URLs, but frankly something about 300 bytes doesn't look interesting

These cases are almost not checked in our tests, but #438 should solve the problem. Also we can add a request/response generation for the deproxy and this allow us to simulate more real and different traffic. But we need good logging or reports for this (for example - allure)

I can add:

• our tests doesn't contain such a variety of headers (this is very important for http 2 which uses hpack). Probably #438 will help us with this too

[krizhanovsky]

Many headers case should be covered with https://github.com/tempesta-tech/tempesta/issues/1181