consecutive delete and set

[jsia]

got this error during a consecutive delete and set

After a period of time of executing these commands I get a segfault

key = "test"
for (i = 0; i< 100; i++) {
conn.delete(key)
conn.set(key, "test")
}

Note: multiple threads

terminate called after throwing an instance of 'std::length_error'
what(): basic_string::_S_create

Program received signal SIGABRT, Aborted.
[Switching to Thread 801c07400 (LWP 100406/redis-leveldb)]
0x00000008016e0a7c in thr_kill () from /lib/libc.so.7
(gdb)
(gdb) bt
#0 0x00000008016e0a7c in thr_kill () from /lib/libc.so.7
#1 0x000000080177dd3b in abort () from /lib/libc.so.7
#2 0x0000000800fc4c94 in __gnu_cxx::__verbose_terminate_handler () from /usr/lib/libstdc++.so.6
#3 0x0000000800fc9193 in std::set_unexpected () from /usr/lib/libstdc++.so.6
#4 0x0000000800fc91d3 in std::terminate () from /usr/lib/libstdc++.so.6
#5 0x0000000800fc913a in __cxa_throw () from /usr/lib/libstdc++.so.6
#6 0x000000080101a972 in std::__throw_length_error () from /usr/lib/libstdc++.so.6
#7 0x0000000800fca5d9 in std::string::_Rep::_S_create () from /usr/lib/libstdc++.so.6
#8 0x0000000800fcb305 in std::string::_S_copy_chars () from /usr/lib/libstdc++.so.6
#9 0x0000000800fcb4ba in std::basic_string<char, std::char_traits, std::allocator >::basic_string () from /usr/lib/libstdc++.so.6
#10 0x00000000004072fe in RLConnection::do_read ()
#11 0x0000000000407739 in RLConnection::on_readable ()
#12 0x00000008008aac52 in ev_invoke_pending (loop=0x801c6f600) at ev.c:2994
#13 0x00000008008af37c in ev_run (loop=0x801c6f600, flags=Variable "flags" is not available.

) at ev.c:3394
#14 0x00000000004049fb in RLServer::start ()

[jsia]

I think it has an issue when the sent command is too long I saw ing RLConnection::on_readable that there is a case 0 but there is a TODO is this the fix for this problem?

[KDr2]

@jsia

Yes, it's a known issue about std::length_error exception when the command is too long(longer than the buffer size READ_BUFFER defined in src/rl.h).

This may be fixed in the future, but I'm not sure about it: receive big command should use big memory block, and I should allocate memory dynamically for it, this may hurt the performance.

[KDr2]

@jsia

But your code:

key = "test"
for (i = 0; i< 100; i++) {
conn.delete(key)
conn.set(key, "test")
}

will not trigger this issue, am I right?

[KDr2]

I did some commits in the develop branch, now the server will close the connection to the client who sends a big command(has an argument larger than READ_BUFFER) , could you try this again?

[jsia]

Yup my code will not trigger this issue, i'm trying your commit now will keep you posted

[bitthegeek]

Tried the new development branch but still returns the following under heavy load:

HOSTADDR:()
DB_PATH:(redis.db)
PORT:(8323)
Server running successfully
terminate called after throwing an instance of 'std::length_error'
  what():  basic_string::_S_create

[KDr2]

I am closing this one, because somebody is working on #13.