RNG is not thread safe.

[evoskuil]

Police Terror:

I'm using the current master branch to sync with testnet, but I'm
getting this error:

...
14:15:56.657549 WARNING [server] Failed to connect secure notification
worker: operation failed
14:15:56.756083 WARNING [server] Failed to connect secure notification
worker: operation failed
terminate called after throwing an instance of 'std::runtime_error'
what(): random_device::random_device(const std::string&)
14:15:57.415901 INFO [server] Stop signal detected (code: 6).
Aborted (core dumped)

Here's what debug.log is showing:

14:15:57.415478 DEBUG [network] Failure connecting outbound: operation
failed
14:15:57.415501 DEBUG [network] Failure connecting outbound: operation
failed
14:15:57.415570 DEBUG [node] Redundant block
[0000000057bd296621bc2f314e3594c412a0cd9e94f1083554c3184160612450] from
[177.189.250.198:18333]
14:15:57.415605 DEBUG [network] Connecting to [121.40.72.188:18333]
14:15:57.415703 DEBUG [network] Valid block payload from
[177.189.250.198:18333](207 bytes)
14:15:57.415757 DEBUG [network] Connecting to [212.93.226.185:18333]
14:15:57.415816 DEBUG [network] Connecting to [2.55.186.124:18333]
14:15:57.415901 INFO [server] Stop signal detected (code: 6).
14:15:57.415937 DEBUG [node] Redundant block
[000000008a675b534ea86b312f29c9fdcce7b7b07ca1879c144b8e20215ac201] from
[177.189.250.198:18333]
14:15:57.416008 DEBUG [network] Valid block payload from
[177.189.250.198:18333](207 bytes)
g

I've attached the .cfg file I'm using.

Looking at libbitcoin/src/utility/random.cpp, I see that
nonzero_pseudo_random() is throwing. Why isn't that an assert instead?

https://github.com/libbitcoin/libbitcoin/blob/master/src/utility/random.cpp#L46

Here is the backtrace:

9 libbitcoin::pseudo_random () at src/utility/random.cpp:39

10 0x00007ffff5a3d48d in libbitcoin::network::hosts::fetch

(this=this@entry=0x657310,
out=...) at src/collections/hosts.cpp:76

11 0x00007ffff5a27e7a in

libbitcoin::network::p2p::fetch_address(std::function<void
(std::error_code const&, libbitcoin::message::network_address const&)>)
const (this=0x656ad0,
handler=...) at src/p2p.cpp:430

12 0x00007ffff5a5ae5f in

libbitcoin::network::session::fetch_address(std::function<void
(std::error_code const&, libbitcoin::config::authority const&)>) const (
this=this@entry=0x7fff280d6330, handler=...) at
src/sessions/session.cpp:83

13 0x00007ffff5a6064e in

libbitcoin::network::session_batch::new_connect(std::shared_ptrlibbitcoin::network::connector,
std::function<void (std::error_code const&,
std::shared_ptrlibbitcoin::network::channel)>)
(this=this@entry=0x7fff280d6330,
connect=std::shared_ptr (count 115, weak 1) 0x7ffef03f5760, handler=...)
at src/sessions/session_batch.cpp:70

14 0x00007ffff5a60b53 in

libbitcoin::network::session_batch::connect(std::shared_ptrlibbitcoin::network::connector,
std::function<void (std::error_code const&,
std::shared_ptrlibbitcoin::network::channel)>) (this=0x7fff280d6330,
connect=std::shared_ptr (count 115, weak 1) 0x7ffef03f5760, handler=...)
at src/sessions/session_batch.cpp:56

15 0x00007ffff5a6f335 in

libbitcoin::network::session_outbound::new_connection (
this=this@entry=0x7fff280d6330,
connect=std::shared_ptr (count 115, weak 1) 0x7ffef03f5760)
at src/sessions/session_outbound.cpp:87

16 0x00007ffff5a6f7d1 in

libbitcoin::network::session_outbound::handle_connect (
this=0x7fff280d6330, ec=..., channel=...,
connect=std::shared_ptr (count 115, weak 1) 0x7ffef03f5760)
at src/sessions/session_outbound.cpp:97

[evoskuil]

The following indicates that the method threw due to encountering 100 zeros in succession. The only explanations I can think of are lack of thread safety (which is documented) or flawed RNG implementation.

terminate called after throwing an instance of 'std::runtime_error'
what(): random_device::random_device(const std::string&)

[evoskuil]

The zero return failure can be resolved by bounding the distribution:

std::uniform_int_distribution<uint64_t> distribution(begin, end);

Although this does not resolve the core issue (of 100 zero returns from the RNG).

[evoskuil]

In order to provide efficient thread safety the generator needs to be moved to thread local storage.

[evoskuil]

See: libbitcoin/libbitcoin-system#559