- Mar 05, 2024
-
-
Nicki Křížek authored
-
- Mar 04, 2024
-
-
Nicki Křížek authored
If the target server doesn't send any reply within the configured time limit (when some are expected), abort the traffic replay. This most commonly happens due to either incorrect configuration or server crash. In the case of premature exit, ensure that existing statistics are collected. The default (baked-in) server timeout limit is configured to be 16 seconds. This has proven to be the right magic number in the development of related tools (respdiff). It is unlikely that this limit would have to be reconfigured or disabled, so no way to adjust or turn off this limit is provided until a reasonable use-case is presented.
-
- Mar 01, 2024
-
-
Nicki Křížek authored
-
- Feb 19, 2024
- Feb 16, 2024
-
-
Oto Šťáva authored
Unfortunately, we are not yet able to detect the misconfiguration [1] ourselves, so when `%NO_TICKETS` is set in the priority string, `zero_rtt` also needs to be explicitly set to `false`, otherwise Shotgun misbehaves. [1] See GnuTLS mailing list: <https://lists.gnutls.org/pipermail/gnutls-help/2024-January/004844.html>
-
- Jan 29, 2024
- Jan 26, 2024
- Jan 17, 2024
-
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
This merges the two counters into conn_handshakes, as they are mutually exclusive anyway. Configuration can still refer to 'tcp_hs' for compatibility, but the preferred way is now 'conn_hs'. Note that 'tcp_hs' will also work for QUIC.
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
... instead of upon handshake confirmation.
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
When containerized, Shotgun would randomly fail with `Undefined symbol: output_dnssim_new`. When I add this assertion, it somehow seems to work reliably, so I assume that LuaJIT might be throwing away the seemingly unused result of `loader.load` sometimes, which causes the issue. Another solution *might* be just using the returned namespace directly, but I don't really want to go rewriting the whole `.lua` file for this, if the assert works.
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-
Oto Šťáva authored
-