These macros should replace the use of assert() in our entire codebase.
assert() have the following issues:
  - can be turned off at compilation time
  - they don't have consistent meaning in our code

kr_require() behaves similarly to assert - it checks a condition and
aborts if it fails. Unlike asserts, these aren't turned off by using
-DNDEBUG. kr_require() should be used for non-recoverable errors.

kr_assume() is a way to check for non-fatal errors which supports error
reporing, debugging and recovery. The function returns a boolean value
which the caller must use for error handling. An error log message is
produced when the condition fails. Optionally, when kr_debug_assumption
is set to true, the process will use fork() and the child will abort().
This generates a coredump for debugging purposes, while allowing the
parent process to keep running and recover from the non-fatal error.
This can be useful for debugging hard to reproduce errors in production
environments.

Technically, we return an actual lua table instead of this
pretty-printed representation, but I chose it to be consistent with the
rest of the docs.

Add HTTP headers to an optional kr_request.qsource.headers structure.
Headers are stored as name, value string pairs.

The following snippet can be used to access the headers in lua modules:

```
if (req.qsource.headers ~= nil) then
	for i = 1, tonumber(req.qsource.headers.len) do
		local name = ffi.string(req.qsource.headers.at[i - 1].name)
		local value = ffi.string(req.qsource.headers.at[i - 1].value)
		print(name, value)
	end
end
```

Fixes #616

I see no potential use for `set`; it's not a suitable abstraction.
And for `get` we want to use a pointer to the public type
instead of the private one.

Overall, worker.h has way too many stuff, but this branch is
not a good place to clean it up.

Tomas Krizek authored 3 years ago and Vladimír Čunát committed 3 years ago

This can happen for example when we want to send an answer, but the
http stream (or the connection?) is already closed.

Direct leak of 48 byte(s) in 1 object(s) allocated from:
    #0 0x7f5ad2445459 in __interceptor_malloc /build/gcc/src/gcc/libsanitizer/asan/asan_malloc_linux.cpp:145
    #1 0x55c0db3fc442 in http_write_pkt ../daemon/http.c:610
    #2 0x55c0db3fc882 in http_write ../daemon/http.c:651
    #3 0x55c0db3e9bb1 in qr_task_send ../daemon/worker.c:700
    #4 0x55c0db3ee86c in qr_task_finalize ../daemon/worker.c:1321
    #5 0x55c0db3f0123 in qr_task_step ../daemon/worker.c:1633
    #6 0x55c0db3f0982 in worker_submit ../daemon/worker.c:1755
    #7 0x55c0db3d992a in session_wirebuf_process ../daemon/session.c:759
    #8 0x55c0db3c5f01 in udp_recv ../daemon/io.c:89
    #9 0x7f5ad22b0e0e  (/usr/lib/libuv.so.1+0x20e0e)

Tomas Krizek authored 3 years ago and Lukas Jezek committed 3 years ago

The on_frame_recv() callback ins't guaranteed to be called by nghttp2.
This can happen e.g. in a case when nghttp2 issues a PROTOCOL_ERROR
RST_STREAM frame. Previously, it would leave the connection in a
stream-processing state, making it completely useless.

While this guarantees a cleanup will be called eventually, some streams
may still get ignored due to the order of various callbacks and data
processing procedures. Still, it's better than the previous
implementation.

Vladimír Čunát authored 3 years ago and Tomas Krizek committed 3 years ago

libknot >= 2.9 provides it and their version is less buggy :-)
In particular, it works with unaligned pointers.

This reverts commit 4079a1a9, reversing
changes made to a900fdbf.

This reverts commit 4dab349e, reversing
changes made to 4bcf335d.

This reverts commit 99e6e754, reversing
changes made to 65bed85f.