1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
|
[appendix]
== Migrating from Ranch 1.7 to Ranch 2.0
Ranch 2.0 adds support for multiple connection supervisors.
Ranch 1.x had a bottleneck because it used only a single
connection supervisor. This was more evident when many
connections were dropped at once as the supervisor couldn't
keep up and failed to accept new connections while cleaning
up the old ones. Ranch 2.0 behaves much better in this scenario
by default. Multiple connection supervisors also helps with
concurrently accepting new connections.
Ranch 2.0 also adds experimental support for opening more
than one listening socket on a single port.
Starting with Ranch 2.0 we are also providing a Prometheus
collector as a separate project as well as a Grafana
dashboard.
Ranch 2.0 is compatible with Erlang/OTP 21.0 onward. Support
for Erlang/OTP 19 and 20 has been removed.
=== Features added
* The `num_conns_sups` option has been added. It allows
configuring the number of connection supervisors. It
now defaults to `num_accceptors`. The old behavior can
be obtained by setting this value to 1.
* The `logger` option is no longer experimental. It now
defaults to `logger` instead of `error_logger`.
* UNIX domain sockets are now supported.
* The active N socket option is now supported. It requires
Erlang/OTP 21.3 or above for TLS, however.
* Embedded listeners are now failing in a predictable
manner when `ranch_server` goes down. It is no longer
necessary to embed `ranch_sup` and the recommendation
is now to just start Ranch normally when using embedded
listeners.
* Two steps handshake is now supported. This allows
obtaining TLS extensions and updating options before
resuming the handshake. The handshake can also be
canceled.
=== Experimental features added
* The experimental `num_listen_sockets` option has been
added. It allows opening more than one listening socket
per listener. It can only be used alongside the Linux
`SO_REUSEPORT` socket option or equivalent. It allows
working around a bottleneck in the kernel and maximizes
resource usage, leading to increased rates for accepting
new connections.
=== Features removed
* The `socket` option was removed. A more viable solution
is to define a custom transport module that returns a fresh
socket when `Transport:listen/1` is called.
=== Changed behaviors
* The callback function `Transport:listen/1` and its
implementations in `ranch_tcp` and `ranch_ssl` have changed
to accept a map of transport options instead of only
socket options.
* The callback function `Transport:messages/0` return value
now include the tag used for passive messages.
* The `Socket` argument was removed from `Protocol:start_link/3`.
The socket must now be obtained by calling `ranch:handshake/1,2`.
=== Added functions
* The functions `ranch:handshake_continue/1,2` and
`ranch:handshake_cancel/1` can be used to perform
a two steps handshake. These functions may not be
supported by all transports.
=== Changed functions
* The `NumAcceptors` argument was removed from `ranch:start_listener/5`
and `ranch:child_spec/5` and moved to the transport options.
* Ranch options can no longer be passed along with socket options
as a proplist. The only forms allowed are now the `ranch:opts()`
map or only socket options as-is. Individual transport options
are now validated as well. The `ranch:opts()` map must
be used when socket options also use a map. This applies to the
`ranch:start_listener/5`, `ranch:child_spec/5` and
`ranch:set_transport_options/2` functions.
* The function `ranch:info/1,2` now returns a map containing
each listener's information rather than a list of key/values.
The key `num_acceptors` was removed as it can be found in the
transport options.
* The function `ranch:set_transport_options/2` no longer requires
the listener to be suspended. Which options apply immediately,
on suspend/resume or on restart has been documented. Some work
has also been done to make these option changes more predictable.
=== Removed functions
* The function `ranch:accept_ack/1` has been removed in favor
of `ranch:handshake/1,2`.
=== Bugs fixed
* Calling `ranch:remove_connection/1` will now resume a sleeping
acceptor process when applicable.
* Repeatedly calling `ranch:remove_connection/1` from a connection
process would crash the respective connection supervisor. This has
now been fixed.
* When a connection process was failing to start, the socket was
not closed and this lead to leaking sockets. This is now corrected.
=== Other changes
* Connection draining has now been documented in the guide
following user feedback and discussions.
* Ranch is now tested against `havoc`, a chaos monkey style
testing tool. Currently includes three scenarios: normal
TCP and TLS listeners and embedded TCP listener. This new
test suite helped uncover a misplaced `monitor/2` call
added during the development of Ranch 2.0.
* The supervisor for acceptors and the parent supervisor for
connection supervisors now have an adaptive restart
intensity limit set to `1 + ceil(math:log2(NumChildren))`
to allow room for errors when they have many children.
* Ranch now uses stricter compiler options. Missing function
specs were added to internal modules.
* Ranch now calls `ssl:handshake/1,2,3` instead of
`ssl:ssl_accept/1,2`.
* The `ranch_ssl:ssl_opt()` type has been updated to conform
with Erlang/OTP 22.0.
|
