try to fix#3978
**Background**
check #3978
**Research**
I referred the Android platform's solution, because I have android
background, and there is a loop to handle message inside android.
ff007a03c0/core/java/android/os/Handler.java (L701-L706C6)
```
public final boolean sendMessageDelayed(@NonNull Message msg, long delayMillis) {
if (delayMillis < 0) {
delayMillis = 0;
}
return sendMessageAtTime(msg, SystemClock.uptimeMillis() + delayMillis);
}
```
59d9dc1f50/libutils/SystemClock.cpp (L37-L51)
```
/*
* native public static long uptimeMillis();
*/
int64_t uptimeMillis()
{
return nanoseconds_to_milliseconds(uptimeNanos());
}
/*
* public static native long uptimeNanos();
*/
int64_t uptimeNanos()
{
return systemTime(SYSTEM_TIME_MONOTONIC);
}
```
59d9dc1f50/libutils/Timers.cpp (L32-L55)
```
#if defined(__linux__)
nsecs_t systemTime(int clock) {
checkClockId(clock);
static constexpr clockid_t clocks[] = {CLOCK_REALTIME, CLOCK_MONOTONIC,
CLOCK_PROCESS_CPUTIME_ID, CLOCK_THREAD_CPUTIME_ID,
CLOCK_BOOTTIME};
static_assert(clock_id_max == arraysize(clocks));
timespec t = {};
clock_gettime(clocks[clock], &t);
return nsecs_t(t.tv_sec)*1000000000LL + t.tv_nsec;
}
#else
nsecs_t systemTime(int clock) {
// TODO: is this ever called with anything but REALTIME on mac/windows?
checkClockId(clock);
// Clock support varies widely across hosts. Mac OS doesn't support
// CLOCK_BOOTTIME (and doesn't even have clock_gettime until 10.12).
// Windows is windows.
timeval t = {};
gettimeofday(&t, nullptr);
return nsecs_t(t.tv_sec)*1000000000LL + nsecs_t(t.tv_usec)*1000LL;
}
#endif
```
For Linux system, we can use `clock_gettime` api, but it's first
appeared in Mac OSX 10.12.
`man clock_gettime`
The requirement is to find an alternative way to get the timestamp in
microsecond unit, but the `clock_gettime` get nanoseconds, the math
formula is the nanoseconds / 1000 = microsecond. Then I check the
performance of this api + math division.
I used those code to check the `clock_gettime` performance.
```
#include <sys/time.h>
#include <time.h>
#include <stdio.h>
#include <unistd.h>
int main() {
struct timeval tv;
struct timespec ts;
clock_t start;
clock_t end;
long t;
while (1) {
start = clock();
gettimeofday(&tv, NULL);
end = clock();
printf("gettimeofday clock is %lu\n", end - start);
printf("gettimeofday is %lld\n", (tv.tv_sec * 1000000LL + tv.tv_usec));
start = clock();
clock_gettime(CLOCK_MONOTONIC, &ts);
t = ts.tv_sec * 1000000L + ts.tv_nsec / 1000L;
end = clock();
printf("clock_monotonic clock is %lu\n", end - start);
printf("clock_monotonic: seconds is %ld, nanoseconds is %ld, sum is %ld\n", ts.tv_sec, ts.tv_nsec, t);
start = clock();
clock_gettime(CLOCK_MONOTONIC_RAW, &ts);
t = ts.tv_sec * 1000000L + ts.tv_nsec / 1000L;
end = clock();
printf("clock_monotonic_raw clock is %lu\n", end - start);
printf("clock_monotonic_raw: nanoseconds is %ld, sum is %ld\n", ts.tv_nsec, t);
sleep(3);
}
return 0;
}
```
Here is output:
env: Mac OS M2 chip.
```
gettimeofday clock is 11
gettimeofday is 1709775727153949
clock_monotonic clock is 2
clock_monotonic: seconds is 1525204, nanoseconds is 409453000, sum is 1525204409453
clock_monotonic_raw clock is 2
clock_monotonic_raw: nanoseconds is 770493000, sum is 1525222770493
```
We can see the `clock_gettime` is faster than `gettimeofday`, so there
are no performance risks.
**MacOS solution**
`clock_gettime` api only available until mac os 10.12, for the mac os
older than 10.12, just keep the `gettimeofday`.
check osx version in `auto/options.sh`, then add MACRO in
`auto/depends.sh`, the MACRO is `MD_OSX_HAS_NO_CLOCK_GETTIME`.
**CYGWIN**
According to google search, it seems the
`clock_gettime(CLOCK_MONOTONIC)` is not support well at least 10 years
ago, but I didn't own an windows machine, so can't verify it. so keep
win's solution.
---------
Co-authored-by: winlin <winlinvip@gmail.com>
For coroutine, we should use `__sanitizer_start_switch_fiber` which
similar to`VALGRIND_STACK_REGISTER`, see
https://github.com/google/sanitizers/issues/189#issuecomment-1346243598
for details. If not fix this, asan will output warning:
```
==72269==WARNING: ASan is ignoring requested __asan_handle_no_return: stack type: default top: 0x00016f638000; bottom 0x000106bec000; size: 0x000068a4c000 (1755627520)
False positive error reports may follow
For details see https://github.com/google/sanitizers/issues/189
```
It will cause asan failed to get the stack, see
`research/st/asan-switch.cpp` for example:
```
==71611==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x000103600733 at pc 0x0001009d3d7c bp 0x000100b4bd40 sp 0x000100b4bd38
WRITE of size 1 at 0x000103600733 thread T0
#0 0x1009d3d78 in foo(void*) asan-switch.cpp:13
```
After fix this issue, it should provide the full stack when crashing:
```
==73437==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x000103300733 at pc 0x000100693d7c bp 0x00016f76f550 sp 0x00016f76f548
WRITE of size 1 at 0x000103300733 thread T0
#0 0x100693d78 in foo(void*) asan-switch.cpp:13
#1 0x100693df4 in main asan-switch.cpp:23
#2 0x195aa20dc (<unknown module>)
```
For primordial coroutine, if not set the stack by
`st_set_primordial_stack`, then the stack is NULL and asan can't get the
stack tracing. Note that it's optional and only make it fail to display
the stack information, no other errors.
---
Co-authored-by: john <hondaxiao@tencent.com>
Improvements for ST(State Threads):
1. ST: Use g++ for CXX compiler.
2. ST: Remove macros for clist.
3. ST: Remove macros for global thread and vp.
4. ST: Remove macros for vp queue operations.
5. ST: Remove macros for context switch.
6. ST: Remove macros for setjmp/longjmp.
7. ST: Remove macro for stack pad.
8. ST: Refine macro for valgrind.
---------
Co-authored-by: Jacob Su <suzp1984@gmail.com>
To manage an object:
```cpp
// Before
MyClass* ptr = new MyClass();
SrsAutoFree(MyClass, ptr);
ptr->do_something();
// Now
SrsUniquePtr<MyClass> ptr(new MyClass());
ptr->do_something();
```
To manage an array of objects:
```cpp
// Before
char* ptr = new char[10];
SrsAutoFreeA(char, ptr);
ptr[0] = 0xf;
// Now
SrsUniquePtr<char[]> ptr(new char[10]);
ptr[0] = 0xf;
```
In fact, SrsUniquePtr is a limited subset of SrsAutoFree, mainly
managing pointers and arrays. SrsUniquePtr is better than SrsAutoFree
because it has the same API to standard unique ptr.
```cpp
SrsUniquePtr<MyClass> ptr(new MyClass());
ptr->do_something();
MyClass* p = ptr.get();
```
SrsAutoFree actually uses a pointer to a pointer, so it can be set to
NULL, allowing the pointer's value to be changed later (this usage is
different from SrsUniquePtr).
```cpp
// OK to free ptr correctly.
MyClass* ptr;
SrsAutoFree(MyClass, ptr);
ptr = new MyClass();
// Crash because ptr is an invalid pointer.
MyClass* ptr;
SrsUniquePtr<MyClass> ptr(ptr);
ptr = new MyClass();
```
Additionally, SrsAutoFreeH can use specific release functions, which
SrsUniquePtr does not support.
---------
Co-authored-by: Jacob Su <suzp1984@gmail.com>
1. Add live benchmark support in srs-bench, which only connects and
disconnects without any media transport, to test source creation and
disposal and verify source memory leaks.
2. SmartPtr: Support cleanup of HTTP-FLV stream. Unregister the HTTP-FLV
handler for the pattern and clean up the objects and resources.
3. Support benchmarking RTMP/SRT with srs-bench by integrating the gosrt
and oryx RTMP libraries.
4. Refine SRT and RTC sources by using a timer to clean up the sources,
following the same strategy as the Live source.
---------
Co-authored-by: Haibo Chen <495810242@qq.com>
Co-authored-by: Jacob Su <suzp1984@gmail.com>
Try to fix two blackbox test:
1. TestSlow_SrtPublish_HttpTsPlay_HEVC_Basic: fixed by enlarge the wait
time from 3 seconds to 4 before run ffprobe task, which will record the
stream by ffmpeg first.
2 TestSlow_SrtPublish_HlsPlay_HEVC_Basic: fixed by wait 16 seconds
before run ffprobe task.
About the 2 case: it seems ridiculous to add 16 seconds delay before run
ffprobe task.
> So, I start #4088 to check the github action workflow process, I start
this branch from a very earlier version `6.0.113
(srs/core/srs_core_version6.hpp)`, what I found it that, inside `SRS
blackbox-test`, the srs version `6.0.128`, the latest version, was
printed. That's really wired.
I confirmed this is not the SRS source code's problem, check
https://github.com/suzp1984/srs/actions/runs/9511600525/job/26218025559
the srs code 6.0.113 was checkout and running actions based on them,
still met same problem.
---------
Co-authored-by: winlin <winlinvip@gmail.com>
When debugging the RTMP protocol, we can capture packets using tcpdump
and then replay the pcap file. For example:
```bash
cd ~/git/srs/trunk/3rdparty/srs-bench/pcap
tcpdump -i any -w t.pcap tcp port 1935
go run . -f ./t.pcap -s 127.0.0.1:1935
```
However, sometimes due to poor network conditions between the server and
the client, there may be many retransmitted packets. In such cases,
setting up a transparent TCP proxy that listens on port 1935 and
forwards to port 19350 can be a solution:
```bash
./objs/srs -c conf/origin.conf
cd 3rdparty/srs-bench/tcpproxy/ && go run main.go
tcpdump -i any -w t.pcap tcp port 19350
```
This approach allows for the implementation of packet dumping,
multipoint replication, or the provision of detailed timestamps and byte
information at the proxy. It enables the collection of debugging
information without the need to modify the server.
---------
`TRANS_BY_GPT4`
---------
Co-authored-by: john <hondaxiao@tencent.com>
1. After enabling FFmpeg opus, the transcoding time for each opus packet
is around 4ms.
2. To speed up case execution, our test publisher sends 400 opus packets
at intervals of 1ms.
3. After the publisher starts, wait for 30ms, then the player starts.
4. Due to the lengthy processing time for each opus packet, SRS
continuously receives packets from the publisher, so it doesn't switch
coroutines and can't accept the player's connection.
5. Only after all opus packets are processed will it accept the player
connection. Therefore, the player doesn't receive any data, leading to
the failure of the case.
---------
Co-authored-by: winlin <winlinvip@gmail.com>
When accessing the SRS Stack, you should log in and use a token for each
request, or utilize the HTTP API with a secret Bearer token included in
every request. The SRS Stack HTTP API proxies both /api/v1 and /rtc/v1
to the SRS HTTP API while ensuring secure authentication. Additionally,
there is a console in the SRS Stack that requires the same token to
request the SRS Stack HTTP API, which is then proxied to the SRS HTTP
API.
The SRS Stack runs SRS with the HTTP API listening at 127.0.0.1:1985 on
the local loopback interface, allowing only the SRS Stack to access it
without authentication. All other users must login and access the SRS
Stack through its interface, rather than directly accessing the SRS HTTP
API within the SRS Stack.
---------
Co-authored-by: panda <542638787@qq.com>
For the DJI M30, there is a bug where empty NALU packets with a size of
zero are causing issues with HLS streaming. This bug leads to random
unpublish events due to the SRS disconnecting the connection for the HLS
module when it fails to handle empty NALU packets.
To address this bug, we have patched the system to ignore any empty NALU
packets with a size of zero. Additionally, we have created a tool in the
srs-bench to replay pcapng files captured by tcpdump or Wireshark. We
have also added utest using mprotect and asan to detect any memory
corruption.
It is important to note that this bug has been fixed in versions 4.0.271
6477f31004 and 5.0.170
939f6b484b. This patch specifically
addresses the issue in SRS 6.0.
Please be aware that there is another commit related to this bug that
partially fixes the issue but still leaves a small problem for asan to
detect memory corruption. This commit,
577cd299e1, only ignores empty NALU
packets but still reads beyond the memory.
---------
Co-authored-by: chundonglinlin <chundonglinlin@163.com>
1. Enable blackbox test for each PR and push.
2. Refine Makefile and README for srs-bench.
3. Add blackbox using FFmpeg and ffprobe.
4. Add blackbox basic test for RTMP stream.
5. Add blackbox basic test for HTTP-FLV stream.
6. Fix utest rand seed issue.
PICK 2141d220b4
1. Build regression test tool for gb28181.
2. Run regression test for gb28181.
3. Format go code and eliminate logs.
4. Change base docker to ubuntu20.
PICK 7750bdae10
* FLV: Support set default has_av and disable guessing. v5.0.110
1. Support config default has_audio and has_video.
2. Support disable guessing has_audio or has_video.
* FLV: Reset to false if start to guess has_av.
* FLV: Add regression test for FLV header av metadata.
1. Ignore audo or video packets if FLV header disable it.
2. Run: Add regression test config and run for IDEA.
3. Test: Refine regression test to allow no audio/video for FLV
4. Config: Whether drop packet if not match header.
1. Support github actions on Windows Server 2022.
2. Use cygwin64 in windows-latest to build SRS.
3. Package SRS-xxx-setup.exe by NSIS.exe
4. Patch crypto/include/config.h for libsrtp.
5. Support run as administrator.
6. Apply utest for cygwin.
7. Enable srtp over openssl.
01. Support GB config as StreamCaster.
02. Support disable GB by --gb28181=off.
03. Add utests for SIP examples.
04. Wireshark plugin to decode TCP/9000 as rtp.rfc4571
05. Support MPEGPS program stream codec.
06. Add utest for PS stream codec.
07. Decode MPEGPS packet stream.
08. Carry RTP and PS packet as helper in PS message.
09. Support recover from error mode.
10. Support process by a pack of PS/TS messages.
11. Add statistic for recovered and msgs dropped.
12. Recover from err position fastly.
13. Define state machine for GB session.
14. Bind context to GB session.
15. Re-invite when media disconnected.
16. Update GitHub actions with GB28181.
17. Support parse CANDIDATE by env or pip.
18. Support mux GB28181 to RTMP.
19. Support regression test by srs-bench.
ChenGH
Jacob Su
Winlin
john
Haibo Chen
Loken
panda
chundonglinlin
mapengfei53
Winlin