Directives

Core

rtmp

syntax: rtmp { ... }
context: root
The block which holds all RTMP settings

server

syntax: server { ... }
context: rtmp
Declares RTMP server instance

rtmp {
  server {
  }
}

listen

Adds listening socket to NGINX for accepting RTMP connections

server {
    listen 1935;
}

application

syntax: application name { ... }
context: server

Creates RTMP application. Unlike http location application name cannot be a pattern.

server {
    listen 1935;
    application myapp {
    }
}

timeout

syntax: timeout value
context: rtmp, server

Socket timeout. This value is primarily used for writing. Most of time RTMP module does not expect any activity on all sockets except for publisher socket. If you want broken socket to get quickly disconnected use active tools like keepalive or RTMP ping. Default is 1 minute.

timeout 60s;

ping

syntax: ping value
context: rtmp, server

RTMP ping interval. Zero turns ping off. RTMP ping is a protocol feature for active connection check. A special packet is sent to remote peer and a reply is expected within a timeout specified with ping_timeout directive. If ping reply is not received within this time then connection is closed. Default value for ping is 1 minute. Default ping timeout is 30 seconds.

ping 3m;
ping_timeout 30s;

ping_timeout

syntax: ping_timeout value
context: rtmp, server

See ping description above.

max_streams

syntax: max_streams value
context: rtmp, server

Sets maximum number of RTMP streams. Data streams are multiplexed into a single data stream. Different channels are used for sending commands, audio, video etc. Default value is 32 which is usually ok for many cases.

max_streams 32;

ack_window

syntax: ack_window value
context: rtmp, server

Sets RTMP acknowledge window size. It's the number of bytes received after which peer should send acknowledge packet to remote side. Default value is 5000000.

ack_window 5000000;

chunk_size

syntax: chunk_size value
context: rtmp, server

Maximum chunk size for stream multiplexing. Default is 4096. The bigger this value the lower CPU overhead. This value cannot be less than 128.

chunk_size 4096;

max_queue

max_message

syntax: max_queue value
context: rtmp, server

Maximum size of input data message. All input data comes split into messages (and further in chunks). A partial message is kept in memory while waiting for it to complete. In theory incoming message can be very large which can be a problem for server stability. Default value 1M is enough for many cases.

max_message 1M;

buflen

syntax: buflen time
context: rtmp, server

Sets default buffer length. Usually client sends RTMP set_buflen command before playing and resets this setting. Default is 1000 ms.

buflen 5s;

out_queue

out_cork

Access

allow

Syntax: allow [play|publish] address|subnet|all
Context: rtmp, server, application

Allow publishing/playing from addresses specified or from all addresses. Allow/deny directives are checked in order of appearance.

allow publish 127.0.0.1;
deny publish all;
allow play 192.168.0.0/24;
deny play all;

deny

Syntax: deny [play|publish] address|subnet|all
Context: rtmp, server, application

See allow for description.

Exec

exec_push

Syntax: exec_push command arg*
Context: rtmp, server, application

Specifies external command with arguments to be executed on every stream published. When publishing stops the process is terminated. Full path to binary should be specified as the first argument. There are no assumptions about what this process should do. However this feature is useful with ffmpeg for stream transcoding. FFmpeg is supposed to connect to nginx-rtmp as a client and output transcoded stream back to nginx-rtmp as publisher. Substitutions of form $var/${var} can be used within command line:

$name - stream name
$app - application name
$addr - client address
$flashver - client flash version
$swfurl - client swf url
$tcurl - client tc url
$pageurl - client page url

Shell-style redirects can be specified in exec_push directive for writing output and accepting input. Supported are

truncating output >file
appending output >>file
descriptor redirects like 1>&2
input <file

Make sure there's no space between redirection character and stream name/number.

You can specify full path to the command to execute or short command name. In the latter case binary is looked up in directories specified by the PATH environment variable. By default nginx clears the environment which will usually make rtmp module run only binaries located in standard directories like /bin and /usr/bin. To make this always work please keep the original PATH variable value with the following nginx directive.

env PATH;

The following ffmpeg call transcodes incoming stream to HLS-ready stream (H264/AAC). FFmpeg should be compiled with libx264 & libfaac support for this example to work.

application src {
    live on;
    exec_push ffmpeg -i rtmp://localhost/src/$name -vcodec libx264 -vprofile baseline -g 10 -s 300x200 -acodec libfaac -ar 44100 -ac 1 -f flv rtmp://localhost/hls/$name 2>>/var/log/ffmpeg-$name.log;
}

application hls {
    live on;
    hls on;
    hls_path /tmp/hls;
    hls_fragment 15s;
}

exec_pull

Syntax: exec_pull command arg*
Context: rtmp, server, application

Specifies external command with arguments to be executed on play event. The command is executed when first client connects to the stream and is killed when the last one disconnects. This directive makes it possible to pull remote stream in any format for local clients.

The feature works reliably only in single-worker mode. The reason for this is we cannot make sure external process always connects to the right worker. It will obviously connect to a random one. While this will still work in most cases it's not a recommended architecture, it will be unstable and buggy.

Directive arguments are the same as for exec_push.

application myapp {
    live on;
    exec_pull ffmpeg -i http://example.com/video_$name.ts -c copy -f flv rtmp://localhost/$app/$name;
}

In the above configuration exec_pull directive serves all streams. That leads to certain limitations on remote stream name format. It should be possible to construct the remote url using available variables like $app, $name etc. When it's not possible you can add exec_options on directive which permits setting additional stream options in exec-family directives. The only option supported now is name option.

application myapp {
    live on;
    exec_options on;
    exec_pull ffmpeg -i http://example.com/tv1.ts -c copy -f flv rtmp://localhost/$app/$name name=mystream;
    exec_pull ffmpeg -i http://another.example.com/video_plus.ts -c copy -f flv rtmp://localhost/$app/$name name=anotherstream;
}

exec

Syntax: exec command arg*
Context: rtmp, server, application

exec is an alias of exec_push

exec_options

Syntax: exec_options on|off
Context: rtmp, server, application

The directive toggles exec options mode. When activated you can add exec-family directive options. The only exec option supported is name. This option makes it possible to apply exec only to specified stream. Default if off.

exec_options on;
# call on_publish only for "mystream"
exec_publish http://localhost/on_publish name=mystream;

# call on_play only for "another"
exec_play http://localhost/on_play name=another;

# execute different ffmpeg's for different streams
exec_pull http://example.com/abc.ts -c copy -f flv rtmp://localhost/$name/$app name=mystream;
exec_pull http://my.example.com/tele.ts -c copy -f flv rtmp://localhost/$name/$app name=tv;
exec_pull http://enother.example.com/hello/f.ts -c copy -f flv rtmp://localhost/$name/$app name=fun;

exec_static

Syntax: exec_static command arg*
Context: rtmp, server, application

Similar to exec but runs specified command at nginx start. Does not support substitutions since has no session context.

exec_static ffmpeg -i http://example.com/video.ts -c copy -f flv rtmp://localhost/myapp/mystream;

exec_kill_signal

Syntax: exec_kill_signal signal
Context: rtmp, server, application

Sets process termination signal. Default is kill (SIGKILL). You can specify numeric or symbolic name (for POSIX.1-1990 signals).

exec_kill_signal term;
exec_kill_signal usr1;
exec_kill_signal 3;

respawn

Syntax: respawn on|off
Context: rtmp, server, application

If turned on respawns child process when it's terminated while publishing is still on. Default is on;

respawn off;

respawn_timeout

Syntax: respawn_timeout timeout
Context: rtmp, server, application

Sets respawn timeout to wait before starting new child instance. Default is 5 seconds.

respawn_timeout 10s;

exec_publish

Syntax: exec_publish command arg*
Context: rtmp, server, application

Specifies external command with arguments to be executed on publish event. Return code is not analyzed. Substitutions of exec are supported here as well. In addition args variable is supported holding query string arguments.

exec_play

Syntax: exec_play command arg*
Context: rtmp, server, application

Specifies external command with arguments to be executed on play event. Return code is not analyzed. Substitution list is the same as for exec_publish.

exec_play_done

Syntax: exec_play_done command arg*
Context: rtmp, server, application

Specifies external command with arguments to be executed on play_done event. Return code is not analyzed. Substitution list is the same as for exec_publish.

exec_publish_done

Syntax: exec_publish_done command arg*
Context: rtmp, server, application

Specifies external command with arguments to be executed on publish_done event. Return code is not analyzed. Substitution list is the same as for exec_publish.

exec_record_done

Syntax: exec_record_done command arg*
Context: rtmp, server, application, recorder

Specifies external command with arguments to be executed when recording is finished. Substitution of exec_publish are supported here as well as additional variables

recorder - recorder name
path - recorded file path (/tmp/rec/mystream-1389499351.flv)
filename - path with directory omitted (mystream-1389499351.flv)
basename - file name with extension omitted (mystream-1389499351)
dirname - directory path (/tmp/rec)

Examples

# track client info
exec_play bash -c "echo $addr $pageurl >> /tmp/clients";
exec_publish bash -c "echo $addr $flashver >> /tmp/publishers";

# convert recorded file to mp4 format
exec_record_done ffmpeg -y -i $path -acodec libmp3lame -ar 44100 -ac 1 -vcodec libx264 $dirname/$basename.mp4;

Live

live

Syntax: live on|off
Context: rtmp, server, application

Toggles live mode i.e. one-to-many broadcasting.

live on;

interleave

Syntax: interleave on|off
Context: rtmp, server, application

Toggles interleave mode. In this mode audio and video data is transmitted on the same RTMP chunk stream. Defaults to off.

interleave on;

wait_key

Syntax: wait_key on|off
Context: rtmp, server, application

Makes video stream start with a key frame. Defaults to off.

wait_key on;

wait_video

Syntax: wait_video on|off
Context: rtmp, server, application

Disable audio until first video frame is sent. Defaults to off. Can be combined with wait_key to make client receive video key frame with all other data following it. However this usually increases connection delay. You can tune keyframe interval in your encoder to reduce the delay.

Recent versions of IE need this option to be enabled for normal playback.

wait_video on;

publish_notify

Syntax: publish_notify on|off
Context: rtmp, server, application

Send NetStream.Play.PublishNotify and NetStream.Play.UnpublishNotify to subscribers. Defaults to off.

publish_notify on;

drop_idle_publisher

Syntax: drop_idle_publisher timeout
Context: rtmp, server, application

Drop publisher connection which has been idle (no audio/video data) within specified time. Default is off. Note this only works when connection is in publish mode (after sending publish command).

drop_idle_publisher 10s;

sync

Syntax: sync timeout
Context: rtmp, server, application

Synchronize audio and video streams. If subscriber bandwidth is not enough to receive data at publisher rate, some frames are dropped by server. This leads to synchronization problem. When timestamp difference exceeds the value specified as sync argument an absolute frame is sent fixing that. Default is 300ms.

sync 10ms;

play_restart

Syntax: play_restart on|off
Context: rtmp, server, application

If enabled nginx-rtmp sends NetStream.Play.Start and NetStream.Play.Stop to each subscriber every time publisher starts or stops publishing. If disabled each subscriber receives those notifications only at the start and end of playback. Default is off.

play_restart off;

idle_streams

Syntax: idle_streams on|off
Context: rtmp, server, application

If disabled nginx-rtmp prevents subscribers from connecting to idle/nonexistent live streams and disconnects all subscribers when stream publisher disconnects. Default is on.

idle_streams off;

Record

record

Toggles record mode. Stream can be recorded in flv file. This directive specifies what exactly should be recorded:

off - no recording at all
all - audio & video (everything)
audio - audio
video - video
keyframes - only key video frames
manual - never start recorder automatically, use control interface to start/stop

There can be any compatible combination of keys in a single record directive.

record all;

record audio keyframes;

record_path

syntax: record_path path
context: rtmp, server, application, recorder

Specifies record path to put recorded flv files to.

record_path /tmp/rec;

record_suffix

syntax: record_suffix value
context: rtmp, server, application, recorder

Sets record file suffix. Defaults to '.flv'.

record_suffix _recorded.flv;

Record suffix can be a pattern in strftime format. The following directive

record_suffix -%d-%b-%y-%T.flv;

will produce files of the form mystream-24-Apr-13-18:23:38.flv. All supported strftime format options can be found on strftime man page.

record_unique

syntax: record_unique on|off
context: rtmp, server, application, recorder

If turned on appends current timestamp to recorded files. Otherwise the same file is re-written each time new recording takes place. Default is off.

record_unique on;

record_append

syntax: record_append on|off
context: rtmp, server, application, recorder

Toggles file append mode. When turned on recorder appends new data to the old file or creates it when it's missing. There's no time gap between the old data and the new data in file. Default is off.

record_append on;

record_lock

syntax: record_lock on|off
context: rtmp, server, application, recorder

When turned on currently recorded file gets locked with fcntl call. That can be checked from elsewhere to find out which file is being recorded. Default is off.

record_lock on;

On FreeBSD you can use flock tool to check that. On Linux flock and fcntl are unrelated so you are left with writing a simple script checking file lock status. Here's an example of such script isunlocked.py.

#!/usr/bin/python

import fcntl, sys

sys.stderr.close()
fcntl.lockf(open(sys.argv[1], "a"), fcntl.LOCK_EX|fcntl.LOCK_NB)

record_max_size

syntax: record_max_size size
context: rtmp, server, application, recorder

Set maximum recorded file size.

record_max_size 128K;

record_max_frames

syntax: record_max_frames nframes
context: rtmp, server, application, recorder

Sets maximum number of video frames per recorded file.

record_max_frames 2;

record_interval

syntax: record_interval time
context: rtmp, server, application, recorder

Restart recording after this number of (milli)seconds. Off by default. Zero means no delay between recordings. If record_unique is off then all record fragments are written to the same file. Otherwise timestamp is appended which makes files differ (given record_interval is longer than 1 second).

record_interval 1s;

record_interval 15m;

recorder

syntax: recorder name {...}
context: application

Create recorder block. Multiple recorders can be created withing single application. All the above mentioned recording-related directives can be specified in recorder{} block. All settings are inherited from higher levels.

application {
    live on;

    # default recorder
    record all;
    record_path /var/rec;

    recorder audio {
        record audio;
        record_suffix .audio.flv;
    }

    recorder chunked {
        record all;
        record_interval 15s;
        record_path /var/rec/chunked;
    }
}

record_notify

syntax: record_notify on|off
context: rtmp, server, application, recorder

Toggles sending NetStream.Record.Start and NetStream.Record.Stop status messages (onStatus) to publisher when specific recorder starts or stops recording file. Status description field holds recorder name (empty for default recorder). Off by default.

recorder myrec {
    record all manual;
    record_path /var/rec;
    record_notify on;
}

Video on demand

play

Syntax: play dir|http://loc [dir|http://loc]*
Context: rtmp, server, application

Play flv or mp4 file from specified directory or HTTP location. If the argument is prefixed with http:// then it is assumed that file should be downloaded from remote http location before playing. Note playing is not started until the whole file is downloaded. You can use local nginx to cache files on local machine.

Multiple play locations can be specified in a single play directive. When multiple play directives are specified the location lists are merged and inherited from higher scopes. An attempt to play each location is made until a successful location is found. If such location is not found error status is sent to client.

Indexed FLVs are played with random seek capability. Unindexed FLVs are played with seek/pause disabled (restart-only mode). Use FLV indexer (for example, yamdi) for indexing.

If you play FLVs recorded with the record directive please do not forget to index them before playing. They are created unindexed.

Mp4 files can only be played if both video and audio codec are supported by RTMP. The most common case is H264/AAC.

application vod {
    play /var/flvs;
}

application vod_http {
    play http://myserver.com/vod;
}

application vod_mirror {
    # try local location first, then access remote location
    play /var/local_mirror http://myserver.com/vod;
}

Playing /var/flvs/dir/file.flv:

ffplay rtmp://localhost/vod//dir/file.flv

The two slashes after vod make ffplay use vod and application name and the rest of the url as playpath.

play_temp_path

Syntax: play_temp_path dir
Context: rtmp, server, application

Sets location where remote VOD files are stored before playing. Default is /tmp;

play_temp_path /www;
play http://example.com/videos;

play_local_path

Syntax: play_local_path dir
Context: rtmp, server, application

Sets location where remote VOD files copied from play_temp_path directory after they are completely downloaded. Empty value disables the feature. By default it's empty. The feature can be used for caching remote files locally.

This path should be on the same device as play_temp_path.

# search file in /tmp/videos.
# if not found play from remote location
# and store in /tmp/videos

play_local_path /tmp/videos;
play /tmp/videos http://example.com/videos;

Relay

pull

Syntax: pull url [key=value]*
Context: application

Creates pull relay. Stream is pulled from remote machine and becomes available locally. It only happens when at least one player is playing the stream locally.

Url syntax: [rtmp://]host[:port][/app[/playpath]]. If application is missing then local application name is used. If playpath is missing then current stream name is used instead.

The following parameters are supported:

app - explicit application name
name - local stream name to bind relay to; if empty or non-specified then all local streams within application are pulled
tcUrl - auto-constructed if empty
pageUrl - page url to pretend
swfUrl - swf url to pretend
flashVer - flash version to pretend, default is 'LNX.11,1,102,55'
playPath - remote play path
live - toggles special behavior for live streaming, values: 0,1
start - start time in seconds
stop - stop time in seconds
static - makes pull static, such pull is created at nginx start

If a value for a parameter contains spaces then you should use quotes around the WHOLE key=value pair like this : 'pageUrl=FAKE PAGE URL'.

pull rtmp://cdn.example.com/main/ch?id=12563 name=channel_a;

pull rtmp://cdn2.example.com/another/a?b=1&c=d pageUrl=http://www.example.com/video.html swfUrl=http://www.example.com/player.swf live=1;

pull rtmp://cdn.example.com/main/ch?id=12563 name=channel_a static;

push

Syntax: push url [key=value]*
Context: application

Push has the same syntax as pull. Unlike pull push directive publishes stream to remote server.

push_reconnect

Syntax: push_reconnect time
Context: rtmp, server, application

Timeout to wait before reconnecting pushed connection after disconnect. Default is 3 seconds.

push_reconnect 1s;

session_relay

Syntax: session_relay on|off
Context: rtmp, server, application

Toggles session relay mode. In this mode relay is destroyed when connection is closed. When the setting is off relay is destroyed when stream is closed so that another relay could possibly be created later. Default is off.

session_relay on;

Notify

on_connect

Syntax: on_connect url
Context: rtmp, server

Sets HTTP connection callback. When clients issues connect command an HTTP request is issued asynchronously and command processing is suspended until it returns result code. If HTTP 2xx code is returned then RTMP session continues. The code of 3xx makes RTMP redirect to another application whose name is taken from Location HTTP response header. Otherwise connection is dropped.

Note this directive is not allowed in application scope since application is still unknown at connection stage.

HTTP request receives a number of arguments. POST method is used with application/x-www-form-urlencoded MIME type. The following arguments are passed to caller:

call=connect
addr - client IP address
app - application name
flashVer - client flash version
swfUrl - client swf url
tcUrl - tcUrl
pageUrl - client page url

In addition to the above mentioned items all arguments passed explicitly to connect command are also sent with the callback. You should distinguish connect arguments from play/publish arguments. Players usually have a special way of setting connection string separate from play/publish stream name. As an example here's how these arguments are set in JWPlayer

...
streamer: "rtmp://localhost/myapp?connarg1=a&connarg2=b",
file: "mystream?strarg1=c&strarg2=d",
...

Ffplay (with librtmp) example

ffplay "rtmp://localhost app=myapp?connarg1=a&connarg2=b playpath=mystream?strarg1=c&strarg2=d"

Usage example

on_connect http://example.com/my_auth;

Redirect example

location /on_connect {
    if ($arg_flashver != "my_secret_flashver") {
        rewrite ^.*$ fallback? permanent;
    }
    return 200;
}

on_play

Syntax: on_play url
Context: rtmp, server, application

Sets HTTP play callback. Each time a clients issues play command an HTTP request is issued asynchronously and command processing is suspended until it returns result code. HTTP result code is then analyzed.

HTTP 2xx code continues RTMP session
HTTP 3xx redirects RTMP to another stream whose name is taken from Location HTTP response header. If new stream name is started with rtmp:// then remote relay is created instead. Relays require that IP address is specified instead of domain name and only work with nginx versions greater than 1.3.10. See also notify_relay_redirect.
Otherwise RTMP connection is dropped

Redirect example

http {
    ...
    location /local_redirect {
        rewrite ^.*$ newname? permanent;
    }
    location /remote_redirect {
        # no domain name here, only ip
        rewrite ^.*$ rtmp://192.168.1.123/someapp/somename? permanent;
    }
    ...
}

rtmp {
    ...
    application myapp1 {
        live on;
        # stream will be redirected to 'newname'
        on_play http://localhost:8080/local_redirect;
    }
    application myapp2 {
        live on;
        # stream will be pulled from remote location
        # requires nginx >= 1.3.10
        on_play http://localhost:8080/remote_redirect;
    }
    ...
}

HTTP request receives a number of arguments. POST method is used with application/x-www-form-urlencoded MIME type. The following arguments are passed to caller:

call=play
addr - client IP address
clientid - nginx client id (displayed in log and stat)
app - application name
flashVer - client flash version
swfUrl - client swf url
tcUrl - tcUrl
pageUrl - client page url
name - stream name

In addition to the above mentioned items all arguments passed explicitly to play command are also sent with the callback. For example if stream is accessed with the url rtmp://localhost/app/movie?a=100&b=face&foo=bar then a, b & foo are also sent with callback.

on_play http://example.com/my_callback;

on_publish

Syntax: on_publish url
Context: rtmp, server, application

The same as on_play above with the only difference that this directive sets callback on publish command. Instead of remote pull push is performed in this case.

on_done

Syntax: on_done url
Context: rtmp, server, application

Sets play/publish terminate callback. All the above applies here. However HTTP status code is not checked for this callback.

on_play_done

Syntax: on_play_done url
Context: rtmp, server, application

Same behavior as on_done but only for play end event.

on_publish_done

Syntax: on_publish_done url
Context: rtmp, server, application

Same behavior as on_done but only for publish end event.

on_record_done

syntax: on_record_done url
context: rtmp, server, application, recorder

Set record_done callback. In addition to common HTTP callback variables it receives the following values

recorder - recorder name in config or empty string for inline recorder
path - recorded file path

Example

on_record_done http://example.com/recorded;

on_update

syntax: on_update url
context: rtmp, server, application

Set update callback. This callback is called with period of notify_update_timeout. If a request returns HTTP result other than 2xx connection is terminated. This can be used to synchronize expired sessions. Two additional arguments time and timestamp are passed to this handler:

time is the number of seconds since play/publish call
timestamp is RTMP timestamp of the last audio/video packet sent to the client

You can use timestamp argument to individually limit playback duration for each user.

on_update http://example.com/update;

notify_update_timeout

syntax: notify_update_timeout timeout
context: rtmp, server, application

Sets timeout between on_update callbacks. Default is 30 seconds.

notify_update_timeout 10s;
on_update http://example.com/update;

notify_update_strict

syntax: notify_update_strict on|off
context: rtmp, server, application

Toggles strict mode for on_update callbacks. Default is off. When turned on all connection errors, timeouts as well as HTTP parse errors and empty responses are treated as update failures and lead to connection termination. When off only valid HTTP response codes other that 2xx lead to failure.

notify_update_strict on;
on_update http://example.com/update;

notify_relay_redirect

syntax: notify_relay_redirect on|off
context: rtmp, server, application

Enables local stream redirect for on_play and on_publish remote redirects. New stream name is MD5 hash of RTMP URL used for remote redirect. Default is off.

notify_relay_redirect on;

notify_method

syntax: notify_method get|post
context: rtmp, server, application, recorder

Sets HTTP method for notifications. Default is POST with application/x-www-form-urlencoded content type. In certain cases GET is preferable, for example if you plan to handle the call in http{} section of nginx. In this case you can use arg_* variables to access arguments.

notify_method get;

With GET method handling notifications in http{} section can be done this way

location /on_play {
    if ($arg_pageUrl ~* localhost) {
        return 200;
    }
    return 500;
}

HLS

hls

Syntax: hls on|off
Context: rtmp, server, application

Toggles HLS on the application.

hls on;
hls_path /tmp/hls;
hls_fragment 15s;

In http{} section set up the following location for clients to play HLS.

http {
    ...
    server {
        ...
        location /hls {
            types {
                application/vnd.apple.mpegurl m3u8;
            }
            root /tmp;
            add_header Cache-Control no-cache;
        }
    }
}

hls_path

Syntax: hls_path path
Context: rtmp, server, application

Sets HLS playlist and fragment directory. If the directory does not exist it will be created.

hls_fragment

Syntax: hls_fragment time
Context: rtmp, server, application

Sets HLS fragment length. Defaults to 5 seconds.

hls_playlist_length

Syntax: hls_playlist_length time
Context: rtmp, server, application

Sets HLS playlist length. Defaults to 30 seconds.

hls_playlist_length 10m;

hls_sync

Syntax: hls_sync time
Context: rtmp, server, application

Sets HLS timestamp synchronization threshold. Default is 2ms. This feature prevents crackling noises after conversion from low-resolution RTMP (1KHz) to high-resolution MPEG-TS (90KHz).

hls_sync 100ms;

hls_continuous

Syntax: hls_continuous on|off
Context: rtmp, server, application

Toggles HLS continuous mode. In this mode HLS sequence number is started from where it stopped last time. Old fragments are keeped. Default is off.

hls_continuous on;

hls_nested

Syntax: hls_nested on|off
Context: rtmp, server, application

Toggles HLS nested mode. In this mode a subdirectory of hls_path is created for each stream. Playlist and fragments are created in that subdirectory. Default is off.

hls_nested on;

hls_base_url

Syntax: hls_base_url url
Context: rtmp, server, application

Sets base url for HLS playlist items. When empty those items have no prefix and assumed to be at the same location as parent playlist or one level lower when hls_nested is used. This feature applies both to master (variant) and slave HLS playlists. It can let you download the playlist and play it locally since it contains full references to child playlists or fragments. Empty by default.

hls_base_url http://myserver.com/hls/;

hls_cleanup

Syntax: hls_cleanup on|off
Context: rtmp, server, application

Toggles HLS cleanup. By default the feature is on. In this mode nginx cache manager process removes old HLS fragments and playlists from HLS directory.

hls_cleanup off;

hls_fragment_naming

Syntax: hls_fragment_naming sequential|timestamp|system
Context: rtmp, server, application

Sets fragment naming mode.

sequential - use increasing integers
timestamp - use stream timestamp
system - use system time

Default is sequential.

hls_fragment_naming system;

hls_fragment_naming_granularity

Syntax: hls_fragment_naming_granularity number
Context: rtmp, server, application

Sets granularity for hls fragment ids. If above zero, changes ids to divide the provided value. Default is zero.

# use system time rounded to 500ms as fragment names
hls_fragment_naming system;
hls_fragment_naming_granularity 500;

hls_fragment_slicing

Syntax: hls_fragment_slicing plain|aligned
Context: rtmp, server, application

Sets fragment slicing mode.

plain - switch fragment when target duration is reached
aligned - switch fragment when incoming timestamp is a multiple of fragment duration. This mode makes it possible to generate identical fragments on different nginx instances

Default is plain.

hls_fragment_slicing aligned;

hls_variant

Syntax: hls_variant suffix [param*]
Context: rtmp, server, application

Adds HLS variant entry. When suffix is matched on stream name then variant playlist is created for the current stream with all entries specified by hls_variant directives in current application. Stripped name without suffix is used as variant stream name. The original stream is processed as usual.

Optional parameters following the suffix are appended to EXT-X-STREAM-INF in m3u8 playlist. See HLS spec. 3.3.10. EXT-X-STREAM-INF for the full list of supported parameters.

rtmp {
    server {
        listen 1935;
    
        application src {
            live on;

            exec ffmpeg -i rtmp://localhost/src/$name
              -c:a libfdk_aac -b:a 32k  -c:v libx264 -b:v 128K -f flv rtmp://localhost/hls/$name_low
              -c:a libfdk_aac -b:a 64k  -c:v libx264 -b:v 256k -f flv rtmp://localhost/hls/$name_mid
              -c:a libfdk_aac -b:a 128k -c:v libx264 -b:v 512K -f flv rtmp://localhost/hls/$name_hi;
        }

        application hls {
            live on;

            hls on;
            hls_path /tmp/hls;
            hls_nested on;

            hls_variant _low BANDWIDTH=160000;
            hls_variant _mid BANDWIDTH=320000;
            hls_variant _hi  BANDWIDTH=640000;
        }
    }
}

hls_type

Syntax: hls_type live|event
Context: rtmp, server, application

Sets HLS playlist type specified in X-PLAYLIST-TYPE playlist directive. Live HLS stream is usually played from the current live position which is several fragments to the end of playlist. Event HLS stream is always played from the start of playlist. When in event mode make sure playlist length is enough for the whole event. Default is live;

hls_type event;

hls_keys

Syntax: hls_keys on|off
Context: rtmp, server, application

Enables HLS encryption. AES-128 method is used to encrypt the whole HLS fragments. Off by default.

hls_keys on;

hls_key_path

Syntax: hls_key_path path
Context: rtmp, server, application

Sets HLS key directory. If the directory does not exist it will be created. By default hls_path directory is used for key files.

hls_keys on;
hls_key_path /tmp/keys;

hls_key_url

Syntax: hls_key_url url
Context: rtmp, server, application

Sets url for HLS key file entries. When empty those items have no prefix and keys are assumed to be at the same location as the playlist. Empty by default.

hls_key_url https://myserver.com/keys/;

Example playlist with the above setting

#EXT-X-KEY:METHOD=AES-128,URI="https://myserver.com/keys/337.key",IV=0x00000000000000000000000000000151

hls_fragments_per_key

Syntax: hls_fragments_per_key value
Context: rtmp, server, application

Sets the number of HLS fragments encrypted with the same key. Zero means only one key will be created at the publish start and all fragments within the session will be encrypted with this key.

hls_fragments_per_key 10;

MPEG-DASH

dash

Syntax: dash on|off
Context: rtmp, server, application

Toggles MPEG-DASH on the application.

dash on;
dash_path /tmp/dash;
dash_fragment 15s;

In http{} section set up the following location for clients to play MPEG-DASH.

http {
    ...
    server {
        ...
        location /dash {
            root /tmp;
            add_header Cache-Control no-cache;
        }
    }
}

dash_path

Syntax: dash_path path
Context: rtmp, server, application

Sets MPEG-DASH playlist and fragment directory. If the directory does not exists it will be created.

dash_fragment

Syntax: dash_fragment time
Context: rtmp, server, application

Sets MPEG-DASH fragment length. Defaults to 5 seconds.

dash_playlist_length

Syntax: dash_playlist_length time
Context: rtmp, server, application

Sets MPEG-DASH playlist length. Defaults to 30 seconds.

dash_playlist_length 10m;

dash_nested

Syntax: dash_nested on|off
Context: rtmp, server, application

Toggles MPEG-DASH nested mode. In this mode a subdirectory of dash_path is created for each stream. Playlist and fragments are created in that subdirectory. Default is off.

dash_nested on;

dash_cleanup

Syntax: dash_cleanup on|off
Context: rtmp, server, application

Toggles MPEG-DASH cleanup. By default the feature is on. In this mode nginx cache manager process removes old MPEG-DASH fragments and manifests from MPEG-DASH directory. Init fragments are deleted after stream manifest is deleted.

dash_cleanup off;

Access log

access_log

Syntax: access_log off|path [format_name]
Context: rtmp, server, application

Sets access log parameters. Logging is turned on by default. To turn it off use access_log off directive. By default access logging is done to the same file as HTTP access logger (logs/access.log). You can specify another log file path in access_log directive. Second argument is optional. It can be used to specify logging format by name. See log_format directive for more details about formats.

log_format new '$remote_addr';
access_log logs/rtmp_access.log new;
access_log logs/rtmp_access.log;
access_log off;

log_format

Syntax: log_format format_name format
Context: rtmp

Creates named log format. Log formats look very much the same as nginx HTTP log formats. Several variables are supported within log format:

connection - connection number
remote_addr - client address
app - application name
name - last stream name
args - last stream play/publish arguments
flashver - client flashVer
swfurl - client swfUrl
tcurl - client tcUrl
pageurl - client pageUrl
command - play/publish commands sent by client: NONE, PLAY, PUBLISH, PLAY+PUBLISH
bytes_sent - number of bytes sent to client
bytes_received - number of bytes received from client
time_local - local time at the end of client connection
session_time - connection duration in seconds
session_readable_time - connection duration in human-readable format
msec - current unix timestamp in SEC.MSEC format

Default log format has the name combined. Here's the definition of this format

$remote_addr [$time_local] $command "$app" "$name" "$args" - 
$bytes_received $bytes_sent "$pageurl" "$flashver" ($session_readable_time)

Limits

max_connections

Syntax: max_connections number
Context: rtmp, server, application

Sets maximum number of connections for rtmp engine. Off by default.

max_connections 100;

Statistics

Statistics module is NGINX HTTP module unlike all other modules listed here. Hence statistics directives should be located within http{} block.

rtmp_stat

Syntax: rtmp_stat all
Context: http, server, location

Sets RTMP statistics handler to the current HTTP location. RTMP statistics is dynamic XML document. To watch this document in browser as XHTML page use rtmp_stat_stylesheet directive.

http {
    server {
        location /stat {
            rtmp_stat all;
            rtmp_stat_stylesheet stat.xsl;
        }
        location /stat.xsl {
            root /path/to/stat/xsl/file;
        }
    }
}

rtmp_stat_stylesheet

Syntax: rtmp_stat_stylesheet path
Context: http, server, location

Adds XML stylesheet reference to statistics XML to make it viewable in browser. See rtmp_stat description and example for more information.

Multi-worker live streaming

Multi-worker live streaming is implemented through pushing stream to remaining nginx workers.

rtmp_auto_push

Syntax: rtmp_auto_push on|off
Context: root

Toggles auto-push (multi-worker live streaming) mode. Default is off.

rtmp_auto_push_reconnect

Syntax: rtmp_auto_push_reconnect timeout
Context: root

Sets auto-push reconnect timeout when worker is killed. Default is 100 milliseconds.

rtmp_socket_dir

Syntax: rtmp_socket_dir dir
Context: root

Sets directory for UNIX domains sockets used for stream pushing. Default is /tmp.

rtmp_auto_push on;
rtmp_auto_push_reconnect 1s;
rtmp_socket_dir /var/sock;

rtmp {
    server {
        listen 1935;
        application myapp {
            live on;
        }
    }
}

Control

Control module is NGINX HTTP module and should be located within http{} block.

rtmp_control

Syntax: rtmp_control all
Context: http, server, location

Sets RTMP control handler to the current HTTP location.

http {
    server {
        location /control {
            rtmp_control all;
        }
    }
}

More details about control module