Here are the arguments you can pass to the waitress.serve` function or use in PasteDeploy configuration (interchangeably):
- hostname or IP address (string) on which to listen, default
0.0.0.0, which means “all IP addresses on this host”.
- TCP port (integer) on which to listen, default
Path of Unix socket (string), default is
None. If a socket path is specified, a Unix domain socket is made instead of the usual inet domain socket.
Not available on Windows.
- Octal permissions to use for the Unix domain socket (string), default is
600. Only used if
- number of threads used to process application logic (integer), default
- IP address of a client allowed to override
wsgi.url_schemevalue (string), default
http; can be overridden per-request by the value of the
X_FORWARDED_PROTOheader, but only if the client address matches
- server identity (string) used in “Server:” header in responses, default
- backlog is the value waitress passes to pass to socket.listen()
1024. This is the maximum number of incoming TCP connections that will wait in an OS queue for an available channel. From listen(1): “If a connection request arrives when the queue is full, the client may receive an error with an indication of ECONNREFUSED or, if the underlying protocol supports retransmission, the request may be ignored so that a later reattempt at connection succeeds.”
- recv_bytes is the argument waitress passes to socket.recv() (integer),
- send_bytes is the number of bytes to send to socket.send() (integer),
18000. Multiples of 9000 should avoid partly-filled TCP packets, but don’t set this larger than the TCP write buffer size. In Linux, /proc/sys/net/ipv4/tcp_wmem controls the minimum, default, and maximum sizes of TCP write buffers.
- A tempfile should be created if the pending output is larger than
outbuf_overflow, which is measured in bytes. The default is 1MB
1048576). This is conservative.
- A tempfile should be created if the pending input is larger than
inbuf_overflow, which is measured in bytes. The default is 512K
524288). This is conservative.
- Stop creating new channels if too many are already active (integer).
100. Each channel consumes at least one file descriptor, and, depending on the input and output body sizes, potentially up to three, plus whatever file descriptors your application logic happens to open. The default is conservative, but you may need to increase the number of file descriptors available to the Waitress process on most platforms in order to safely change it (see
ulimit -a“open files” setting). Note that this doesn’t control the maximum number of TCP connections that can be waiting for processing; the
backlogargument controls that.
- Minimum seconds between cleaning up inactive channels (integer), default
30. See “channel_timeout”.
- Maximum seconds to leave an inactive connection open (integer), default
120. “Inactive” is defined as “has received no data from a client and has sent no data to a client”.
- Boolean: turn off to not log premature client disconnect tracebacks.
- maximum number of bytes of all request headers combined (integer), 256K
- maximum number of bytes in request body (integer), 1GB (
- Boolean: expose tracebacks of unhandled exceptions to client. Default:
timeoutvalue (seconds) passed to
asyncore.loopto run the mainloop. Default: 1. (New in 0.8.3.)
- Boolean: switch from using select() to poll() in
asyncore.loop. By default asyncore.loop() uses select() which has a limit of 1024 file descriptors. Select() and poll() provide basically the same functionality, but poll() doesn’t have the file descriptors limit. Default: False (New in 0.8.6)
- String: the value used as the WSGI
SCRIPT_NAMEvalue. Setting this to anything except the empty string will cause the WSGI
SCRIPT_NAMEvalue to be the value passed minus any trailing slashes you add, and it will cause the
PATH_INFOof any request which is prefixed with this value to be stripped of the prefix. Default: the empty string.