assert() enabled by default

[allen lee]

Tested in libmemcached 0.53, but also apllied in 1.0.4 after check the source code.

assert is enabled by default, so any assert_msg will be a core dump.

If something bad with the connection(ie EHOSTUNREACH, ECONNRESET) when connect_poll, libmemecached will set errno. However it only cares ENOTCONN and ECONNREFUSED, other errors will be considered as MEMCACHED_ERRNO, which means "Programmer error".
function call route: memcached_mget_by_key_real() -> memcached_connect() -> network_connect() -> connect_poll()

Please see the information below:

(gdb) p *instance->error_messages
$1 = {root = 0xfb4bf0, query_id = 16748106, next = 0x0, rc = MEMCACHED_ERRNO, local_errno = 113, size = 111, message = "SYSTEM ERROR(No route to host), host: 10.11.22.33:11212 -> ../libmemcached-0.53-work/libmemcached/connect.cc:75

backtrace:

#0 0x000000302af2e2ed in raise () from /lib64/tls/libc.so.6
#1 0x000000302af2fa3e in abort () from /lib64/tls/libc.so.6
#2 0x00000000007726bf in memcached_set_error (self=@0xfb9348, rc=MEMCACHED_ERRNO, at=0x7e6c20 "../libmemcached-0.53-work/libmemcached/get.cc:235")
    at ../libmemcached-0.53-work/libmemcached/error.cc:246
#3 0x000000000077411a in memcached_mget_by_key_real (ptr=0xfb4bf0, group_key=0x0, group_key_length=0, keys=0x46206490, key_length=0x46206488, number_of_keys=1, mget_mode=false)
    at ../libmemcached-0.53-work/libmemcached/get.cc:310
#4 0x0000000000773826 in memcached_get_by_key (ptr=0xfb4bf0, group_key=0x0, group_key_length=0, key=0x46206620 "Count:<dsttype:1,dstid:625713147,mtype:1>", key_length=41,
    value_length=0x46206554, flags=0x46206558, error=0x4620655c) at ../libmemcached-0.53-work/libmemcached/get.cc:86
#5 0x000000000077376d in memcached_get (ptr=0xfb4bf0, key=0x46206620 "Count:<dsttype:1,dstid:625713147,mtype:1>", key_length=41, value_length=0x46206554, flags=0x46206558,
    error=0x4620655c) at ../libmemcached-0.53-work/libmemcached/get.cc:49
#6 ...

[sinny (sinnydono)]

got struck by the same thing, brief scenario:
 - working with membase backend
 - membase restart
 - call memcached_delete()

this leads to the following:
---------------------------------------------------------------------------------------------------------------------
(gdb) bt
#0 0xffffe410 in __kernel_vsyscall ()
#1 0x00555df0 in raise () from /lib/libc.so.6
#2 0x00557701 in abort () from /lib/libc.so.6
#3 0x080e1289 in memcached_set_error (self=@0x9beacf8, rc=MEMCACHED_ERRNO, at=0x810043a "libmemcached/delete.cc:215") at libmemcached/error.cc:308
#4 0x080e0050 in memcached_delete_by_key (ptr=0x9beacf8, group_key=0x9bea5b4 "79169918414.turbo_2_unlim_internet_dpi", group_key_length=38, key=0x9bea5b4 "79169918414.turbo_2_unlim_internet_dpi", key_length=38, expiration=0)
    at libmemcached/delete.cc:215
#5 0x080e017e in memcached_delete (ptr=0x9beacf8, key=0x9bea5b4 "79169918414.turbo_2_unlim_internet_dpi", key_length=38, expiration=0) at libmemcached/delete.cc:44
#6 0x08070975 in Box::Backend::MemcachedConn::z_del (this=0x9b773c0, key=@0xf7f5ef58) at backend/memcached.cpp:1034
#7 0x080795e5 in Box::Backend::ConnInterface::del (this=0x9b773c0, key=@0xf7f5ef58, silent=Box::Backend::ConnInterface::BECS_NONE) at backend/interface.cpp:170
...
#14 0x0067a5ab in start_thread () from /lib/libpthread.so.0
#15 0x005fecfe in clone () from /lib/libc.so.6
(gdb) f 3
#3 0x080e1289 in memcached_set_error (self=@0x9beacf8, rc=MEMCACHED_ERRNO, at=0x810043a "libmemcached/delete.cc:215") at libmemcached/error.cc:308
warning: Source file is more recent than executable.
308 assert_msg(rc != MEMCACHED_ERRNO, "Programmer error, MEMCACHED_ERRNO was set to be returned to client");
(gdb) p *self->error_messages
$2 = {root = 0x9beacf8, query_id = 2, next = 0x0, rc = MEMCACHED_ERRNO, local_errno = 115, size = 104,
  message = "(163491064) SYSTEM ERROR(Operation now in progress), host: 192.168.64.94:11211 -> libmemcached/io.cc:447\000"...}
(gdb) bt
---------------------------------------------------------------------------------------------------------------------

[Brian Aker (brianaker)]

This will be fixed in the next version. Somewhere in the history this got switched on "no" by default, it should be yes.